{"id":696,"date":"2025-10-23T21:04:27","date_gmt":"2025-10-23T13:04:27","guid":{"rendered":"http:\/\/anyone.anuoblog.cn\/?p=696"},"modified":"2025-10-26T11:00:12","modified_gmt":"2025-10-26T03:00:12","slug":"interactive-concept-network-enhanced-transformer-for-remote-sensing-image-captioning","status":"publish","type":"post","link":"http:\/\/anyone.anuoblog.cn\/index.php\/2025\/10\/23\/interactive-concept-network-enhanced-transformer-for-remote-sensing-image-captioning\/","title":{"rendered":"Interactive Concept Network Enhanced Transformer for Remote Sensing Image Captioning"},"content":{"rendered":"\n<p><a href=\"https:\/\/ieeexplore.ieee.org\/document\/10816624\">Interactive Concept Network Enhanced Transformer for Remote Sensing Image Captioning | IEEE Journals &amp; Magazine | IEEE Xplore<\/a><\/p>\n\n\n\n<p><\/p>\n\n\n\n<p>\u671f\u520a\uff1aIEEE Transactions on Geoscience and Remote Sensing&nbsp;(IEEE TGRS)<\/p>\n\n\n\n<p>\u53d1\u5e03\u65e5\u671f\uff1a2024.12.27<\/p>\n\n\n\n<h1 class=\"wp-block-heading\">\u7b80\u4ecb<\/h1>\n\n\n\n<p>\u672c\u6587ICNET\u805a\u7126\u4e8e\u9065\u611f\u56fe\u50cfcaptioning\uff08\u56fe\u50cf\u63cf\u8ff0\u751f\u6210\uff09\u4efb\u52a1\uff0c\u6700\u7ec8\u7684\u76ee\u6807\u662f\u751f\u6210\u81ea\u7136\u8bed\u8a00\u6587\u672c\u63cf\u8ff0\uff0c\u6838\u5fc3\u662f\u201c\u56fe\u50cf&#8211;&gt;\u8bed\u8a00\u201d\u7684\u8de8\u6a21\u6001\u8f6c\u6362\u3002<\/p>\n\n\n\n<h1 class=\"wp-block-heading\">0 \u6458\u8981<\/h1>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u9065\u611f\u56fe\u50cf captioning\uff08\u56fe\u50cf\u63cf\u8ff0\u751f\u6210\uff09\u5728\u63a8\u52a8\u9065\u611f\u56fe\u50cf\u7406\u89e3\u4e0e\u81ea\u7136\u8bed\u8a00\u751f\u6210\u7ed3\u5408\u65b9\u9762\u53d1\u6325\u7740\u91cd\u8981\u4f5c\u7528\u3002\u7136\u800c\uff0c\u7531\u4e8e\u9065\u611f\u56fe\u50cf\u8986\u76d6\u8303\u56f4\u5e7f\u3001\u4fe1\u606f\u91cf\u5927\uff0c\u96be\u4ee5\u5bf9\u5173\u952e\u76ee\u6807\u53ca\u5176\u5173\u7cfb\u751f\u6210\u51c6\u786e\u7684\u8bed\u4e49\u63cf\u8ff0\u3002\u9488\u5bf9\u8fd9\u4e9b\u95ee\u9898\uff0c\u672c\u6587\u63d0\u51fa\u4e00\u79cd\u65b0\u9896\u7684\u4ea4\u4e92\u5f0f\u6982\u5ff5\u7f51\u7edc\u589e\u5f3a\u578b Transformer\uff08ICNET\uff09\u7528\u4e8e\u9065\u611f\u56fe\u50cf captioning\u3002\u9996\u5148\uff0c\u5728\u5c40\u90e8\u4e0e\u5168\u5c40\u7279\u5f81\u63d0\u53d6\u6a21\u5757\u4e2d\u63d0\u53d6\u591a\u7ea7\u89c6\u89c9\u7279\u5f81\uff1b\u4e3a\u5145\u5206\u6355\u6349\u5c40\u90e8\u7279\u5f81\u4e2d\u7684\u5173\u952e\u76ee\u6807\uff0c\u6784\u5efa\u6982\u5ff5\u6620\u5c04\u7f51\u7edc\uff08CMN\uff09\uff0c\u5c06\u591a\u5c3a\u5ea6\u5c40\u90e8\u7279\u5f81\u6620\u5c04\u5230\u76ee\u6807\u7684\u9ad8\u5c42\u8bed\u4e49\u6982\u5ff5\uff0c\u4f7f\u89c6\u89c9\u7279\u5f81\u6620\u5c04\u4e2d\u7684\u76f8\u5173\u7279\u5f81\u5411\u91cf\u6574\u5408\u4e3a\u591a\u4e2a\u76f8\u5bf9\u72ec\u7acb\u7684\u8bcd\u7279\u5f81\uff0c\u4ece\u800c\u5f25\u5408\u89c6\u89c9\u7279\u5f81\u4e0e\u8bed\u4e49\u6982\u5ff5\u4e4b\u95f4\u7684\u9e3f\u6c9f\uff1b\u968f\u540e\uff0c\u5f15\u5165\u5168\u5c40\u7279\u5f81\u589e\u5f3a\uff08GFE\uff09\u6a21\u5757\uff0c\u63d0\u5347\u5168\u5c40\u5173\u7cfb\u7684\u533a\u5206\u5ea6\u5e76\u8fc7\u6ee4\u65e0\u5173\u5185\u5bb9\uff1b\u6700\u540e\uff0c\u4e3a\u878d\u5408\u8bed\u4e49\u6982\u5ff5\u4e0e\u5168\u5c40\u7279\u5f81\uff0c\u8bbe\u8ba1\u914d\u5907\u6982\u5ff5\u4ea4\u4e92\u6a21\u5757\uff08CIM\uff09\u7684 Transformer\uff0c\u4fc3\u8fdb\u7279\u5f81\u5bf9\u9f50\uff0c\u751f\u6210\u76ee\u6807\u7c7b\u522b\u4e0e\u5173\u7cfb\u8868\u8ff0\u51c6\u786e\u7684\u56fe\u50cf\u63cf\u8ff0\u3002\u5728\u4e09\u4e2a\u9065\u611f\u56fe\u50cf captioning \u6570\u636e\u96c6\u4e0a\u7684\u5b9e\u9a8c\u7ed3\u679c\u8868\u660e\uff0c\u6240\u63d0\u65b9\u6cd5\u5177\u6709\u4f18\u8d8a\u6027\u3002<\/p>\n<\/details>\n\n\n\n<p>1\u3001\u9065\u611f\u56fe\u50cf\u63cf\u8ff0\u501f\u52a9Transformer\u6280\u672f\u5728\u9065\u611f\u56fe\u50cf\u7406\u89e3\u65b9\u9762\u53d1\u6325\u91cd\u8981\u4f5c\u7528\u3002<\/p>\n\n\n\n<p>2\u3001\uff08\u52a8\u673a\uff09\u7531\u4e8e\u9065\u611f\u56fe\u50cf\u8986\u76d6\u8303\u56f4\u5e7f\u3001\u5305\u542b\u4fe1\u606f\u91cf\u5927\uff0c\u56e0\u6b64\u96be\u4ee5\u51c6\u786e\u751f\u6210\u5173\u952e\u76ee\u6807\u53ca\u5176\u76f8\u4e92\u5173\u7cfb\u7684\u8bed\u4e49\u63cf\u8ff0\u3002<\/p>\n\n\n\n<p>3\u3001\u672c\u6587\u63d0\u51fa\u4e86ICNET\u65b0\u578b\u7f51\u7edc\uff0c\u4e00\u79cd\u4ea4\u4e92\u5f0f\u6982\u5ff5\u7f51\u7edc\u589e\u5f3aTransformer\uff0c\u5176\u4e2d\u4e09\u5927\u5173\u952e\u6a21\u5757\uff1a<\/p>\n\n\n\n<p><strong>\u6982\u5ff5\u6620\u5c04\u7f51\u7edc\uff08CMN\uff09<\/strong>\uff1a\u5c06\u56fe\u50cf\u5c40\u90e8\u7279\u5f81\u6620\u5c04\u5230\u9ad8\u5c42\u8bed\u4e49\u6982\u5ff5\uff0c\u628a\u89c6\u89c9\u7279\u5f81\u8f6c\u5316\u4e3a\u591a\u4e2a\u72ec\u7acb\u7684\u8bcd\u7279\u5f81\uff0c\u89e3\u51b3\u201c\u89c6\u89c9-\u8bed\u4e49\u201d\u7684\u7406\u89e3\u9e3f\u6c9f\uff0c\u91cd\u70b9\u6355\u6349\u5173\u952e\u76ee\u6807\u3002<\/p>\n\n\n\n<p><strong>\u5168\u5c40\u7279\u5f81\u589e\u5f3a\uff08GFE\uff09\u6a21\u5757<\/strong>\uff1a\u5f3a\u5316\u56fe\u50cf\u5168\u5c40\u5173\u7cfb\u7684\u533a\u5206\u5ea6\uff0c\u540c\u65f6\u8fc7\u6ee4\u65e0\u5173\u4fe1\u606f\uff0c\u907f\u514d\u5197\u4f59\u5185\u5bb9\u5e72\u6270\u63cf\u8ff0\u751f\u6210\u3002<\/p>\n\n\n\n<p><strong>\u6982\u5ff5\u4ea4\u4e92\u6a21\u5757\uff08CIM\uff09<\/strong>\uff1a\u96c6\u6210\u5230 Transformer \u4e2d\uff0c\u5b9e\u73b0\u8bed\u4e49\u6982\u5ff5\u4e0e\u5168\u5c40\u7279\u5f81\u7684\u878d\u5408\u5bf9\u9f50\uff0c\u786e\u4fdd\u751f\u6210\u7684\u63cf\u8ff0\u80fd\u51c6\u786e\u4f53\u73b0\u76ee\u6807\u7c7b\u522b\u548c\u76f8\u4e92\u5173\u7cfb\u3002<\/p>\n\n\n\n<h1 class=\"wp-block-heading\">1 \u5f15\u8a00<\/h1>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u968f\u7740\u6df1\u5ea6\u795e\u7ecf\u7f51\u7edc\u7684\u53d1\u5c55\uff0c\u9065\u611f\u56fe\u50cf\u5904\u7406\u5728\u76ee\u6807\u68c0\u6d4b [1]-[3]\u3001\u53d8\u5316\u68c0\u6d4b [4]-[6]\u3001\u56fe\u50cf\u5206\u5272 [7]-[9] \u7b49\u8ba1\u7b97\u673a\u89c6\u89c9\u4efb\u52a1\u4e2d\u53d6\u5f97\u4e86\u663e\u8457\u8fdb\u5c55\u3002\u7136\u800c\uff0c\u8fd9\u4e9b\u4efb\u52a1\u901a\u5e38\u4ec5\u751f\u6210\u5355\u4e2a\u6216\u591a\u4e2a\u79bb\u6563\u6807\u7b7e\u6765\u63cf\u8ff0\u4e3b\u8981\u76ee\u6807\uff0c\u5bf9\u4e8e\u9065\u611f\u56fe\u50cf\u4e2d\u590d\u6742\u7684\u76ee\u6807\u96c6\u5408\u4e0e\u573a\u666f\u6240\u5305\u542b\u7684\u4fe1\u606f\uff0c\u5176\u63cf\u8ff0\u80fd\u529b\u5341\u5206\u6709\u9650\u3002\u4e3a\u89e3\u51b3\u8fd9\u4e00\u5c40\u9650\uff0c\u7814\u7a76\u8005\u5c06\u76ee\u5149\u8f6c\u5411\u81ea\u7136\u8bed\u8a00\u5904\u7406\u9886\u57df [10]-[11]\u3002\u9065\u611f\u56fe\u50cf captioning \u4f5c\u4e3a\u878d\u5408\u8ba1\u7b97\u673a\u89c6\u89c9 [12] \u4e0e\u81ea\u7136\u8bed\u8a00\u5904\u7406 [13] \u7684\u8de8\u6a21\u6001\u4efb\u52a1\uff0c\u65e8\u5728\u901a\u8fc7\u81ea\u7136\u8bed\u8a00\u63cf\u8ff0\u9065\u611f\u56fe\u50cf\u7684\u89c6\u89c9\u5185\u5bb9 [14]\u3002\u5b83\u7ed3\u5408\u6587\u672c\u751f\u6210\u4e0e\u89c6\u89c9\u7406\u89e3\uff0c\u7a81\u7834\u79bb\u6563\u6807\u7b7e\u7684\u9650\u5236\uff0c\u4e3a\u9065\u611f\u573a\u666f\u89e3\u8bfb\u63d0\u4f9b\u652f\u6301\uff0c\u80fd\u591f\u66f4\u5168\u9762\u5730\u53cd\u6620\u591a\u4e2a\u76ee\u6807\u7684\u62d3\u6251\u7ed3\u6784\u3001\u5206\u5e03\u60c5\u51b5\u53ca\u5176\u5c5e\u6027\u4fe1\u606f [15]\u3002\u5728\u81ea\u7136\u707e\u5bb3\u8bc4\u4f30\u3001\u571f\u5730\u8986\u76d6\u76d1\u6d4b\u7b49\u4f17\u591a\u5e94\u7528\u9886\u57df\uff0c\u9065\u611f\u56fe\u50cf captioning \u90fd\u80fd\u63d0\u4f9b\u76f4\u89c2\u7684\u6307\u5bfc\u3002<\/p>\n\n\n\n<p>\u4e0e\u81ea\u7136\u56fe\u50cf captioning \u76f8\u6bd4\uff0c\u9065\u611f\u56fe\u50cf\u63cf\u8ff0\u751f\u6210\u96be\u5ea6\u66f4\u5927\u3002\u751f\u6210\u7684\u63cf\u8ff0\u4e0d\u4ec5\u9700\u8981\u6355\u6349\u4e0d\u540c\u5c3a\u5ea6 [16] \u548c\u7c7b\u522b\u7684\u5730\u7269\u7279\u5f81 [17]\uff0c\u8fd8\u9700\u4f53\u73b0\u5730\u7269\u95f4\u7684\u5173\u8054\u5c5e\u6027\u4e0e\u6a21\u6001\u4fe1\u606f [18]\u3002\u9065\u611f\u56fe\u50cf captioning \u7684\u4e3b\u8981\u6311\u6218\u53ef\u5f52\u7eb3\u4e3a\u4e24\u65b9\u9762\uff1a\u5176\u4e00\uff0c\u9065\u611f\u56fe\u50cf\u5177\u6709\u72ec\u7279\u7279\u5f81\uff0c\u96be\u4ee5\u5bf9\u91cd\u8981\u76ee\u6807\u53ca\u5176\u76f8\u4e92\u5173\u7cfb\u8fdb\u884c\u6709\u6548\u62bd\u8c61\u4e0e\u8868\u5f81\uff1b\u5176\u4e8c\uff0c\u63d0\u53d6\u7684\u9065\u611f\u56fe\u50cf\u7279\u5f81\u4e0e\u6587\u672c\u7279\u5f81\u4e4b\u95f4\u7f3a\u4e4f\u4ea4\u4e92\u4e0e\u5bf9\u9f50\uff0c\u5bfc\u81f4\u751f\u6210\u8bed\u53e5\u5b58\u5728\u8bed\u4e49\u504f\u5dee [19]\u3002<\/p>\n\n\n\n<p>\u9488\u5bf9\u7b2c\u4e00\u4e2a\u6311\u6218\uff0c\u73b0\u6709\u5e38\u7528\u65b9\u6cd5\u901a\u5e38\u901a\u8fc7\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\uff08CNN\uff09[20]-[21] \u6216\u89c6\u89c9 Transformer\uff08ViT\uff09[22] \u63d0\u53d6\u9065\u611f\u56fe\u50cf\u7279\u5f81\u3002\u4e3a\u6355\u6349\u4e0d\u540c\u5c3a\u5ea6\u7684\u76ee\u6807\u3001\u907f\u514d\u9057\u6f0f\u5173\u952e\u76ee\u6807 [23]\uff0c\u7814\u7a76\u4eba\u5458\u5229\u7528 CNN \u7684\u4e0d\u540c\u5377\u79ef\u5c42\u63d0\u53d6\u591a\u5c3a\u5ea6\u7279\u5f81 [24]\u3002\u4f46 CNN \u5728\u5efa\u6a21\u56fe\u50cf\u5168\u5c40\u4fe1\u606f\u65b9\u9762\u80fd\u529b\u4e0d\u8db3\uff0c\u5c24\u5176\u5728\u5168\u5c40\u5173\u7cfb\u7684\u63d0\u53d6\u4e0e\u8868\u5f81\u4e0a\u5b58\u5728\u5c40\u9650\u3002\u4e0e CNN \u64c5\u957f\u6355\u6349\u5c40\u90e8\u76ee\u6807\u7684\u7279\u70b9\u76f8\u6bd4\uff0cViT \u901a\u8fc7\u81ea\u6ce8\u610f\u529b\u673a\u5236\u80fd\u591f\u6355\u6349\u56fe\u50cf\u4e2d\u7684\u5168\u5c40\u5173\u7cfb\uff0c\u7136\u800c\u503c\u5f97\u6ce8\u610f\u7684\u662f\uff0cViT \u65e0\u6cd5\u6709\u6548\u5229\u7528\u5c40\u90e8\u4fe1\u606f\uff0c\u5bfc\u81f4\u5176\u5bf9\u4e0d\u540c\u5c3a\u5ea6\u548c\u7c7b\u522b\u7684\u76ee\u6807\u6355\u6349\u80fd\u529b\u53d7\u9650\u3002<\/p>\n\n\n\n<p>\u800c\u7b2c\u4e8c\u4e2a\u6311\u6218\u5219\u6d89\u53ca\u7528\u4e8e\u63cf\u8ff0\u751f\u6210\u7684\u89c6\u89c9\u7279\u5f81\u4e0e\u6587\u672c\u8bed\u4e49\u4e4b\u95f4\u7684\u4ea4\u4e92\u95ee\u9898\u3002\u901a\u5e38\u60c5\u51b5\u4e0b\uff0c\u89c6\u89c9\u7279\u5f81\u4f1a\u88ab\u7f16\u7801\u4e3a\u9ad8\u7ef4\u7279\u5f81\u5411\u91cf\uff0c\u968f\u540e\u57fa\u4e8e\u5faa\u73af\u795e\u7ecf\u7f51\u7edc\uff08RNN\uff09[25]-[26] \u548c Transformer [27]-[28] \u76f8\u5173\u6280\u672f\u76f4\u63a5\u751f\u6210\u63cf\u8ff0\u6587\u672c\uff1b\u53e6\u4e00\u79cd\u4e3b\u6d41\u65b9\u6cd5\u662f\u901a\u8fc7\u4ea4\u53c9\u6ce8\u610f\u529b\u673a\u5236\u5b9e\u73b0\u89c6\u89c9\u7279\u5f81\u4e0e\u6587\u672c\u7279\u5f81\u7684\u4ea4\u4e92 [29]-[31]\u3002\u4f46\u7531\u4e8e\u6a21\u578b\u4ea4\u4e92\u80fd\u529b\u4e0d\u8db3\uff0c\u5bf9\u76ee\u6807\u95f4\u5173\u8054\u5173\u7cfb\u7684\u63cf\u8ff0\u6548\u679c\u53d7\u5230\u9650\u5236\u3002<\/p>\n\n\n\n<p>\u56e0\u6b64\uff0c\u5982\u4f55\u9488\u5bf9\u4e0d\u540c\u5c42\u7ea7\u7684\u56fe\u50cf\u7279\u5f81\uff0c\u6709\u6548\u8fd0\u7528\u7279\u5f81\u878d\u5408\u65b9\u6cd5\uff0c\u5e76\u5bf9\u5173\u952e\u76ee\u6807\u4e0e\u5173\u7cfb\u7684\u8de8\u6a21\u6001\u7279\u5f81\u4ea4\u4e92\u8fdb\u884c\u5efa\u6a21\uff0c\u5f62\u6210\u5168\u9762\u7684\u89c6\u89c9\u8868\u5f81\uff0c\u5bf9\u4e8e\u9065\u611f\u56fe\u50cf captioning \u81f3\u5173\u91cd\u8981\u3002\u4e3a\u7f13\u89e3\u4e0a\u8ff0\u95ee\u9898\uff0c\u672c\u6587\u63d0\u51fa\u4e00\u79cd\u878d\u5408\u591a\u7ea7\u7279\u5f81\u805a\u5408\u4e0e\u8de8\u6a21\u6001\u7279\u5f81\u4ea4\u4e92\u7684\u89e3\u51b3\u65b9\u6848\u3002\u4e3a\u83b7\u53d6\u591a\u5c3a\u5ea6\u76ee\u6807\u4e0e\u5168\u5c40\u5173\u7cfb\u7684\u5168\u9762\u7279\u5f81\uff0c\u8054\u5408\u4f7f\u7528 CNN \u4e0e ViT \u7279\u5f81\u4ee5\u63d0\u5347\u7279\u5f81\u8868\u5f81\u80fd\u529b\uff1b\u63d0\u51fa\u6982\u5ff5\u6620\u5c04\u7f51\u7edc\uff08CMN\uff09\uff0c\u5bf9\u4e0d\u540c\u5c3a\u5ea6\u7684\u5c40\u90e8\u7279\u5f81\u8fdb\u884c\u6709\u6548\u805a\u5408\uff0c\u751f\u6210\u9ad8\u5c42\u8bed\u4e49\u6982\u5ff5\uff1b\u91c7\u7528\u5168\u5c40\u7279\u5f81\u589e\u5f3a\uff08GFE\uff09\u6a21\u5757\uff0c\u83b7\u53d6\u66f4\u5168\u9762\u7684\u5168\u5c40\u8bed\u4e49\u4fe1\u606f\u5e76\u63d0\u5347\u5176\u533a\u5206\u5ea6\uff1b\u6b64\u5916\uff0c\u4e3a\u89e3\u51b3\u8de8\u6a21\u6001\u7279\u5f81\u4ea4\u4e92\u95ee\u9898\uff0c\u9075\u5faa Transformer \u7684\u6807\u51c6\u8f93\u5165\u6216\u8f93\u51fa\u5f62\u5f0f\uff0c\u5b9e\u73b0\u5168\u5c40\u589e\u5f3a\u7279\u5f81\u7684\u4ea4\u4e92\uff0c\u63d0\u9ad8\u6a21\u578b\u5bf9\u5168\u5c40\u5173\u7cfb\u7684\u533a\u5206\u80fd\u529b\u3002\u5177\u4f53\u800c\u8a00\uff0c\u8bbe\u8ba1\u6982\u5ff5\u4ea4\u4e92\u6a21\u5757\uff08CIM\uff09\uff0c\u901a\u8fc7\u5c06\u89c6\u89c9\u7279\u5f81\u6620\u5c04\u4e2d\u7684\u76f8\u5173\u7279\u5f81\u5411\u91cf\u6574\u5408\u4e3a\u591a\u4e2a\u76f8\u5bf9\u72ec\u7acb\u7684\u8bcd\u7279\u5f81\uff0c\u5b9e\u73b0\u8bed\u4e49\u6982\u5ff5\u4e0e\u8bcd\u5411\u91cf\u7684\u4ea4\u4e92\u5efa\u6a21\u3002<\/p>\n\n\n\n<p>\u57fa\u4e8e\u4e0a\u8ff0\u601d\u8def\uff0c\u672c\u6587\u63d0\u51fa\u4e00\u79cd\u7528\u4e8e\u9065\u611f\u56fe\u50cf captioning \u7684\u4ea4\u4e92\u5f0f\u6982\u5ff5\u7f51\u7edc\u589e\u5f3a\u578b Transformer\uff08ICNET\uff09\uff0c\u5176\u7ed3\u6784\u5982\u56fe 1 \u6240\u793a\u3002\u8be5\u6846\u67b6\u4e3b\u8981\u7531\u56db\u4e2a\u90e8\u5206\u7ec4\u6210\uff1a\u5c40\u90e8\u4e0e\u5168\u5c40\u7279\u5f81\u63d0\u53d6\u6a21\u5757\u3001CMN \u6a21\u5757\u3001GFE \u6a21\u5757\u4ee5\u53ca\u914d\u5907 CIM \u7684 Transformer\u3002\u9996\u5148\uff0c\u5229\u7528 CNN \u63d0\u53d6\u56fe\u50cf\u4e0d\u540c\u5c3a\u5ea6\u7684\u5c40\u90e8\u7279\u5f81\uff0c\u901a\u8fc7 ViT \u63d0\u53d6\u56fe\u50cf\u7684\u5168\u5c40\u7279\u5f81\uff1b\u7136\u540e\uff0c\u63d0\u51fa CMN\uff0c\u501f\u52a9\u7a7a\u95f4\u548c\u901a\u9053\u7ef4\u5ea6\u7684\u6ce8\u610f\u529b\u673a\u5236\u5bf9\u5c40\u90e8\u7279\u5f81\u8fdb\u884c\u6709\u6548\u805a\u5408\uff1b\u5229\u7528 GFE \u6a21\u5757\u83b7\u53d6\u66f4\u5168\u9762\u3001\u4e30\u5bcc\u7684\u7279\u5f81\u8868\u5f81\uff1b\u6700\u540e\uff0c\u63d0\u51fa\u914d\u5907 CIM \u7684 Transformer\uff0c\u901a\u8fc7 CIM \u5b9e\u73b0\u8bed\u4e49\u6982\u5ff5\u4e0e\u8bcd\u5411\u91cf\u7684\u4ea4\u4e92\uff0c\u751f\u6210\u8fde\u8d2f\u4e14\u4e0a\u4e0b\u6587\u76f8\u5173\u7684\u56fe\u50cf\u63cf\u8ff0\u3002\u672c\u6587\u5de5\u4f5c\u7684\u4e3b\u8981\u8d21\u732e\u5982\u4e0b\uff1a<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>\u63d0\u51fa\u65b0\u9896\u7684 ICNET\uff0c\u7528\u4e8e\u9065\u611f\u56fe\u50cf captioning \u4efb\u52a1\uff0c\u5b9e\u73b0\u5bf9\u76ee\u6807\u53ca\u5176\u5173\u7cfb\u7684\u51c6\u786e\u63cf\u8ff0\u751f\u6210\uff1b<\/li>\n\n\n\n<li>\u4e3a\u63d0\u5347\u5173\u952e\u76ee\u6807\u63cf\u8ff0\u751f\u6210\u6548\u679c\uff0c\u63d0\u51fa\u805a\u5408\u4e0d\u540c\u5c3a\u5ea6\u7279\u5f81\u7684 CMN \u6a21\u5757\uff0c\u5e76\u8fdb\u4e00\u6b65\u5f15\u5165 CIM \u6a21\u5757\uff0c\u589e\u5f3a\u8bed\u4e49\u6982\u5ff5\u4e0e\u751f\u6210\u8bcd\u8bed\u95f4\u7684\u5173\u8054\u6027\uff0c\u63d0\u9ad8\u6a21\u578b\u5bf9\u63cf\u8ff0\u4e2d\u6bcf\u4e2a\u8bcd\u8bed\u7684\u533a\u5206\u80fd\u529b\uff1b<\/li>\n\n\n\n<li>\u8bbe\u8ba1 GFE \u6a21\u5757\u4ee5\u589e\u5f3a\u5168\u5c40\u8bed\u4e49\u4fe1\u606f\uff0c\u5e76\u5c06\u5176\u6ce8\u5165\u89e3\u7801\u7f51\u7edc\uff0c\u63d0\u5347\u6a21\u578b\u5bf9\u5173\u7cfb\u63cf\u8ff0\u7684\u533a\u5206\u80fd\u529b\u3002<\/li>\n<\/ol>\n\n\n\n<p>\u672c\u6587\u5176\u4f59\u90e8\u5206\u7ed3\u6784\u5982\u4e0b\uff1a\u7b2c\u4e8c\u90e8\u5206\u56de\u987e\u76f8\u5173\u5de5\u4f5c\uff1b\u7b2c\u4e09\u90e8\u5206\u9610\u8ff0\u6a21\u578b\u6784\u5efa\u4e0e\u56fe\u50cf\u63cf\u8ff0\u751f\u6210\u8fc7\u7a0b\uff1b\u7b2c\u56db\u90e8\u5206\u901a\u8fc7\u8bbe\u8ba1\u5b9e\u9a8c\u5bf9\u6240\u63d0\u65b9\u6cd5\u8fdb\u884c\u8bc4\u4f30\uff1b\u6700\u540e\uff0c\u7b2c\u4e94\u90e8\u5206\u5bf9\u5168\u6587\u8fdb\u884c\u603b\u7ed3\u3002<\/p>\n<\/details>\n\n\n\n<p><\/p>\n\n\n\n<h1 class=\"wp-block-heading\">2 \u76f8\u5173\u5de5\u4f5c<\/h1>\n\n\n\n<p>\u672c\u8282\u5c06\u7b80\u8981\u4ecb\u7ecd\u81ea\u7136\u56fe\u50cf captioning \u4e0e\u9065\u611f\u56fe\u50cf captioning \u9886\u57df\u7684\u76f8\u5173\u7814\u7a76\u5de5\u4f5c\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">2.1 \u81ea\u7136\u56fe\u50cf captioning<\/h2>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u65e9\u671f\u7684\u56fe\u50cf captioning \u65b9\u6cd5\u5305\u62ec\u8574\u542b\u63cf\u8ff0\u68c0\u7d22 [32]-[35]\u3001\u6a21\u677f\u586b\u5145\u4ee5\u53ca\u4eba\u5de5\u8bbe\u8ba1\u7684\u81ea\u7136\u8bed\u8a00\u751f\u6210\u6280\u672f [36]-[40]\u3002\u76ee\u524d\u4e3b\u6d41\u7684\u8bed\u53e5\u751f\u6210\u7c7b\u56fe\u50cf captioning \u65b9\u6cd5\u901a\u5e38\u57fa\u4e8e\u6df1\u5ea6\u5b66\u4e60\u751f\u6210\u6a21\u578b\u6784\u5efa\u3002\u5176\u4e2d\uff0c\u6700\u7b80\u5355\u7684\u5b9e\u73b0\u65b9\u5f0f\u662f\u5229\u7528 CNN \u6700\u540e\u4e00\u5c42\u7684\u6fc0\u6d3b\u503c\u63d0\u53d6\u9ad8\u5c42\u8868\u5f81\uff0c\u5e76\u5c06\u5176\u4f5c\u4e3a\u8bed\u8a00\u6a21\u578b\u7684\u6761\u4ef6\u53cd\u5c04\u5143\u7d20\u3002\u5f00\u521b\u6027\u7814\u7a76 \u201cShow and tell\u201d[41] \u4fbf\u91c7\u7528\u4e86\u8fd9\u79cd\u601d\u8def\uff0c\u5c06 GoogLeNet [42] \u7684\u8f93\u51fa\u7ed3\u679c\u8f93\u5165\u5230\u8bed\u8a00\u6a21\u578b\u7684\u521d\u59cb\u9690\u85cf\u72b6\u6001\u4e2d\u3002\u540c\u5e74\uff0cKarpathy \u4e0e Fei-Fei [43] \u5c06 AlexNet [44] \u63d0\u53d6\u7684\u5168\u5c40\u7279\u5f81\u4f5c\u4e3a\u8bed\u8a00\u6a21\u578b\u7684\u8f93\u5165\u3002\u6b64\u5916\uff0cMao \u7b49\u4eba [45] \u548c Donahue \u7b49\u4eba [46] \u5728\u8bed\u8a00\u6a21\u578b\u7684\u6bcf\u4e2a\u65f6\u95f4\u6b65\u90fd\u6ce8\u5165\u4ece VGG \u7f51\u7edc [47] \u63d0\u53d6\u7684\u5168\u5c40\u7279\u5f81\u3002<\/p>\n\n\n\n<p>\u7531\u4e8e\u5168\u5c40\u8868\u5f81\u5b58\u5728\u5c40\u9650\u6027\uff0c\u540e\u7eed\u591a\u6570\u65b9\u6cd5\u901a\u8fc7\u63d0\u9ad8\u89c6\u89c9\u7f16\u7801\u7684\u7c92\u5ea6\u6765\u6539\u8fdb\u6a21\u578b [48]-[50]\u3002\u4f8b\u5982\uff0cDai \u7b49\u4eba [51] \u91c7\u7528\u4e8c\u7ef4\u6fc0\u6d3b\u56fe\u66ff\u4ee3\u4e00\u7ef4\u5168\u5c40\u7279\u5f81\u5411\u91cf\uff0c\u76f4\u63a5\u5411\u8bed\u8a00\u6a21\u578b\u5f15\u5165\u7a7a\u95f4\u7ed3\u6784\u4fe1\u606f\u3002\u5728\u673a\u5668\u7ffb\u8bd1\u9886\u57df\u7684\u7814\u7a76\u542f\u53d1\u4e0b\uff0c\u8bb8\u591a\u56fe\u50cf\u63cf\u8ff0\u7814\u7a76\u56e2\u961f\u5f15\u5165\u989d\u5916\u7684\u6ce8\u610f\u529b\u673a\u5236\uff0c\u4f7f\u56fe\u50cf\u63cf\u8ff0\u67b6\u6784\u5177\u5907\u65f6\u53d8\u89c6\u89c9\u7279\u5f81\u7f16\u7801\u80fd\u529b\uff0c\u4ece\u800c\u62e5\u6709\u66f4\u9ad8\u7684\u7075\u6d3b\u6027\u4e0e\u66f4\u7ec6\u7684\u7c92\u5ea6\u3002\u81ea\u6ce8\u610f\u529b\u673a\u5236\u4f5c\u4e3a\u4e00\u79cd\u80fd\u8ba9\u96c6\u5408\u4e2d\u6bcf\u4e2a\u5143\u7d20\u4e0e\u5176\u4ed6\u6240\u6709\u5143\u7d20\u5efa\u7acb\u5173\u8054\u7684\u6ce8\u610f\u529b\u673a\u5236\uff0c\u901a\u8fc7\u6b8b\u5dee\u8fde\u63a5\u53ef\u8ba1\u7b97\u8be5\u96c6\u5408\u5143\u7d20\u7684\u7ec6\u7c92\u5ea6\u8868\u5f81\u3002Vaswani \u7b49\u4eba [29] \u9996\u6b21\u5c06\u81ea\u6ce8\u610f\u529b\u673a\u5236\u5e94\u7528\u4e8e\u673a\u5668\u7ffb\u8bd1\u4e0e\u8bed\u8a00\u7406\u89e3\u4efb\u52a1\uff0c\u7531\u6b64\u63d0\u51fa Transformer \u67b6\u6784\u53ca\u5176\u53d8\u4f53\u3002\u8be5\u67b6\u6784\u4e0d\u4ec5\u5728\u81ea\u7136\u8bed\u8a00\u5904\u7406\u9886\u57df\u5360\u636e\u4e3b\u5bfc\u5730\u4f4d\uff0c\u968f\u540e\u4e5f\u88ab\u5e7f\u6cdb\u5e94\u7528\u4e8e\u8ba1\u7b97\u673a\u89c6\u89c9\u9886\u57df\u3002<\/p>\n\n\n\n<p>\u5728\u5168\u5c40\u7279\u5f81\u4e0e\u7f51\u683c\u7279\u5f81\u51fa\u73b0\u4e4b\u540e\uff0c\u57fa\u4e8e\u533a\u57df\u7684\u7279\u5f81\u51ed\u501f\u51fa\u8272\u7684\u6027\u80fd\uff0c\u591a\u5e74\u6765\u4e00\u76f4\u662f\u56fe\u50cf captioning \u4efb\u52a1\u7684\u6700\u4f18\u9009\u62e9\u3002\u7136\u800c\uff0c\u8fd1\u5e74\u6765\u591a\u79cd\u56e0\u7d20\u4fc3\u4f7f\u7814\u7a76\u8005\u91cd\u65b0\u63a2\u8ba8\u4f55\u79cd\u7279\u5f81\u6a21\u578b\u6700\u9002\u7528\u4e8e\u56fe\u50cf captioning \u4efb\u52a1\uff0c\u4f8b\u5982\u8bad\u7ec3\u66f4\u5145\u5206\u7684\u7f51\u683c\u7279\u5f81\u6027\u80fd\u63d0\u5347 [52]\u3001\u81ea\u6ce8\u610f\u529b\u89c6\u89c9\u7f16\u7801\u5668\u7684\u51fa\u73b0 [53] \u4ee5\u53ca\u5bf9\u6bd4\u8bed\u8a00 &#8211; \u56fe\u50cf\u9884\u8bad\u7ec3\uff08CLIP\uff09\u7b49\u5927\u89c4\u6a21\u591a\u6a21\u6001\u6a21\u578b\u7684\u5174\u8d77 [54]\u3002\u5f53\u524d\u7684\u7814\u7a76\u7b56\u7565\u5305\u62ec\u5728\u5927\u89c4\u6a21\u6570\u636e\u4e0a\u8bad\u7ec3\u66f4\u4f18\u7684\u76ee\u6807\u68c0\u6d4b\u5668 [55]\uff0c\u6216\u91c7\u7528\u4ece\u5934\u8bad\u7ec3\u7684\u7aef\u5230\u7aef\u89c6\u89c9\u6a21\u578b [56]\u3002\u6b64\u5916\uff0c\u7c7b BERT \u6a21\u578b\u901a\u8fc7\u5b9e\u73b0\u56fe\u50cf\u4e0e\u6587\u672c\u7684\u65e9\u671f\u878d\u5408\u53d6\u5f97\u6210\u529f\uff0c\u8fd9\u4e00\u6210\u679c\u4e5f\u8bc1\u660e\u4e86\u878d\u5408\u6587\u672c\u4fe1\u606f\u7684\u89c6\u89c9\u8868\u5f81\u5177\u6709\u826f\u597d\u7684\u9002\u7528\u6027\u3002<\/p>\n<\/details>\n\n\n\n<h2 class=\"wp-block-heading\">2.2 \u9065\u611f\u56fe\u50cf captioning<\/h2>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u4e0e\u81ea\u7136\u56fe\u50cf captioning \u76f8\u6bd4\uff0c\u9065\u611f\u56fe\u50cf captioning \u7684\u5dee\u5f02\u4e3b\u8981\u4f53\u73b0\u5728\u6570\u636e\u6765\u6e90\u3001\u5185\u5bb9\u7279\u5f81\u4e0e\u8bed\u8a00\u8868\u5f81\u4e09\u4e2a\u65b9\u9762\u3002\u65e9\u5e74\u95f4\uff0cQu \u7b49\u4eba [57] \u63d0\u51fa\u4e00\u79cd\u591a\u6a21\u6001\u7f16\u7801\u5668 &#8211; \u89e3\u7801\u5668\u67b6\u6784\uff0c\u7528\u4e8e\u4ece\u8bed\u4e49\u5c42\u9762\u89e3\u8bfb\u9ad8\u5206\u8fa8\u7387\u9065\u611f\u56fe\u50cf\u3002\u5728\u8be5\u6a21\u578b\u4e2d\uff0c\u5c06 CNN \u63d0\u53d6\u7684\u7279\u5f81\u4e0e RNN \u751f\u6210\u7684\u63cf\u8ff0\u6587\u672c\u76f8\u7ed3\u5408\uff0c\u5f97\u5230\u6700\u7ec8\u7684\u8bed\u53e5\uff0c\u5e76\u5728\u5176\u63d0\u51fa\u7684 UCM-captions \u548c Sydney-captions \u6570\u636e\u96c6\u4e0a\u9a8c\u8bc1\u4e86\u8be5\u67b6\u6784\u7684\u6709\u6548\u6027\u3002<\/p>\n\n\n\n<p>Shi \u4e0e Zou [58] \u9488\u5bf9\u4e0d\u540c\u5730\u7406\u5c3a\u5ea6\u4e0b\u9065\u611f\u56fe\u50cf\u4e2d\u540c\u4e00\u5730\u7269\u5143\u7d20\u5b58\u5728\u7684\u8bed\u4e49\u5dee\u5f02\u95ee\u9898\uff0c\u63d0\u51fa\u4e86\u4e00\u79cd\u89e3\u51b3\u65b9\u6cd5\u3002\u4ed6\u4eec\u5f15\u5165\u5168\u5377\u79ef\u7f51\u7edc\uff08FCN\uff09\uff0c\u4e3a\u4ece\u8c37\u6b4c\u5730\u7403\u548c\u9ad8\u5206\u4e8c\u53f7\u536b\u661f\u83b7\u53d6\u7684\u9065\u611f\u56fe\u50cf\u751f\u6210\u5168\u9762\u4e14\u9c81\u68d2\u7684\u63cf\u8ff0\u6587\u672c\uff0c\u540c\u65f6\u4fdd\u8bc1\u6a21\u578b\u5177\u6709\u826f\u597d\u7684\u901f\u5ea6\u6027\u80fd\u3002\u8be5\u65b9\u6cd5\u9996\u5148\u5b8c\u6210\u5173\u952e\u76ee\u6807\u68c0\u6d4b\u3001\u73af\u5883\u5206\u6790\u4e0e\u666f\u89c2\u5206\u6790\u4e09\u4e2a\u5b50\u4efb\u52a1\uff0c\u968f\u540e\u878d\u5408\u5404\u9636\u6bb5\u7ed3\u679c\u751f\u6210\u6700\u7ec8\u63cf\u8ff0\u3002<\/p>\n\n\n\n<p>\u4e0e\u4e4b\u4e0d\u540c\uff0cZhang \u7b49\u4eba [59] \u5229\u7528 CNN \u68c0\u6d4b\u9065\u611f\u56fe\u50cf\u4e2d\u7684\u4e3b\u8981\u76ee\u6807\uff0c\u5e76\u901a\u8fc7 RNN \u4e3a\u68c0\u6d4b\u5230\u7684\u76ee\u6807\u751f\u6210\u81ea\u7136\u8bed\u8a00\u63cf\u8ff0\u3002Zhang \u7b49\u4eba [60] \u63d0\u51fa\u4e00\u79cd\u89c6\u89c9\u5bf9\u9f50\u6ce8\u610f\u529b\u6a21\u578b\uff0c\u8be5\u6a21\u578b\u5f15\u5165\u7cbe\u5fc3\u8bbe\u8ba1\u7684\u89c6\u89c9\u5bf9\u9f50\u635f\u5931\u3002\u8be5\u635f\u5931\u901a\u8fc7\u660e\u786e\u8ba1\u7b97\u53c2\u4e0e\u4ea4\u4e92\u7684\u56fe\u50cf\u7279\u5f81\u4e0e\u5bf9\u5e94\u8bcd\u5d4c\u5165\u5411\u91cf\u4e4b\u95f4\u7684\u7279\u5f81\u76f8\u4f3c\u5ea6\u6765\u786e\u5b9a\uff1b\u4e3a\u89e3\u51b3\u975e\u89c6\u89c9\u8bcd\u6c47\u5bf9\u6ce8\u610f\u529b\u5c42\u8bad\u7ec3\u7684\u5f71\u54cd\uff0c\u4ed6\u4eec\u8fd8\u6784\u5efa\u4e86\u4e00\u4e2a\u89c6\u89c9\u8bcd\u6c47\u8868\uff0c\u5728\u8ba1\u7b97\u89c6\u89c9\u5bf9\u9f50\u635f\u5931\u65f6\u5c06\u975e\u89c6\u89c9\u8bcd\u6c47\u4ece\u8bed\u53e5\u4e2d\u5254\u9664\u3002<\/p>\n\n\n\n<p>\u6b64\u5916\uff0c\u4e3a\u5f25\u5408\u9065\u611f\u56fe\u50cf\u4f4e\u5c42\u6b21\u7279\u5f81\u4e0e\u9ad8\u5c42\u6b21\u8bed\u4e49\u4e4b\u95f4\u7684\u9e3f\u6c9f\uff0cZhang \u7b49\u4eba [18] \u5229\u7528 FCN \u751f\u6210\u56fe\u50cf\u7279\u5f81\uff0c\u5e76\u901a\u8fc7\u6ce8\u610f\u529b\u673a\u5236\u83b7\u53d6\u4e2d\u95f4\u5411\u91cf\uff0c\u5c06\u5176\u4f5c\u4e3a\u957f\u77ed\u671f\u8bb0\u5fc6\uff08LSTM\uff09\u89e3\u7801\u5668\u7684\u8f93\u5165\uff0c\u5b9e\u73b0\u9065\u611f\u56fe\u50cf\u63cf\u8ff0\u751f\u6210\u3002\u5728\u8fd9\u9879\u7814\u7a76\u7684\u57fa\u7840\u4e0a\uff0cLi \u7b49\u4eba [61] \u63d0\u51fa\u4e00\u79cd\u591a\u7ea7\u6ce8\u610f\u529b\u6a21\u578b\uff0c\u8be5\u6a21\u578b\u5305\u542b\u4e09\u79cd\u6ce8\u610f\u529b\u7ed3\u6784\uff0c\u5206\u522b\u7528\u4e8e\u5173\u6ce8\u56fe\u50cf\u7684\u4e0d\u540c\u533a\u57df\u3001\u4e0d\u540c\u8bcd\u6c47\u4ee5\u53ca\u5b9e\u73b0\u89c6\u89c9\u4e0e\u8bed\u4e49\u6ce8\u610f\u529b\u3002Zhang \u7b49\u4eba [18] \u5219\u91c7\u7528\u57fa\u4e8e Transformer \u7684\u7f51\u7edc\u8fdb\u884c\u9065\u611f\u56fe\u50cf captioning\uff1b\u7279\u522b\u5730\uff0c\u4e3a\u5e94\u5bf9\u8bad\u7ec3\u6570\u636e\u6709\u9650\u7684\u95ee\u9898\uff0c\u5f15\u5165\u8f85\u52a9\u89e3\u7801\u5668\u7528\u4e8e\u591a\u6807\u7b7e\u573a\u666f\u5206\u7c7b\uff0c\u8be5\u8f85\u52a9\u89e3\u7801\u5668\u5229\u7528\u7f16\u7801\u5668\u4e0e\u56fe\u50cf\u63cf\u8ff0\u4e4b\u95f4\u7684\u6982\u5ff5\u76f8\u4f3c\u6027\u4ee5\u53ca\u7a81\u51fa\u8bed\u4e49\u7c7b\u522b\u7684\u80fd\u529b\uff0c\u5728\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\u4e3a\u7f16\u7801\u5668\u63d0\u4f9b\u8f85\u52a9\u3002<\/p>\n\n\n\n<p>Wang \u7b49\u4eba [62] \u63d0\u51fa\u4e00\u79cd\u53ef\u89e3\u91ca\u7684\u9065\u611f\u56fe\u50cf captioning \u6846\u67b6\uff0c\u8be5\u6846\u67b6\u6452\u5f03\u4f20\u7edf\u601d\u8def\uff0c\u91c7\u7528\u7531\u8bcd\u63d0\u53d6\u5668\u548c\u8bed\u53e5\u751f\u6210\u5668\u4e24\u4e2a\u7f51\u7edc\u6784\u6210\u7684\u8bcd &#8211; \u53e5\u53ef\u89e3\u91ca\u6846\u67b6\u3002\u53e6\u5916\uff0cLi \u7b49\u4eba [63] \u63d0\u51fa\u4e00\u79cd\u65b0\u7684\u622a\u65ad\u4ea4\u53c9\u71b5\u635f\u5931\uff0c\u4ee5\u89e3\u51b3\u9065\u611f\u56fe\u50cf captioning \u4efb\u52a1\u4e2d\u4ea4\u53c9\u71b5\u635f\u5931\u6613\u5bfc\u81f4\u7684\u8fc7\u62df\u5408\u95ee\u9898\u3002\u9488\u5bf9\u4ee5\u5f80\u65b9\u6cd5\u5ffd\u89c6\u9886\u57df\u77e5\u8bc6\u7684\u95ee\u9898\uff0cSumbul \u7b49\u4eba [24] \u63d0\u51fa\u4e00\u79cd\u57fa\u4e8e\u7ec6\u7c92\u5ea6\u7ed3\u6784\u5316\u6ce8\u610f\u529b\u7684\u6a21\u578b\uff0c\u7528\u4e8e\u5145\u5206\u5229\u7528\u9ad8\u5206\u8fa8\u7387\u9065\u611f\u56fe\u50cf\u4e2d\u8bed\u4e49\u5185\u5bb9\u7684\u7ed3\u6784\u7279\u5f81\u3002\u6b64\u5916\uff0cZia \u7b49\u4eba [25] \u9996\u5148\u901a\u8fc7\u591a\u5c3a\u5ea6\u89c6\u89c9\u7279\u5f81\u7f16\u7801\u5668\u4ece\u9065\u611f\u56fe\u50cf\u4e2d\u63d0\u53d6\u8be6\u7ec6\u4fe1\u606f\uff0c\u7136\u540e\u5229\u7528\u81ea\u9002\u5e94\u591a\u5934\u6ce8\u610f\u529b\u89e3\u7801\u5668\uff0c\u57fa\u4e8e\u63d0\u53d6\u7684\u591a\u5c3a\u5ea6\u7279\u5f81\u4f18\u5316\u63cf\u8ff0\u751f\u6210\u8fc7\u7a0b\u3002<\/p>\n\n\n\n<p>\u5c3d\u7ba1\u591a\u5c3a\u5ea6\u65b9\u6cd5\u5df2\u53d6\u5f97\u663e\u8457\u8fdb\u5c55\uff0c\u4f46\u8fd9\u4e9b\u65b9\u6cd5\u901a\u5e38\u65e0\u6cd5\u5145\u5206\u8003\u8651\u4e0d\u540c\u5c3a\u5ea6\u4e0b\u89c6\u89c9\u7279\u5f81\u4e0e\u6587\u672c\u7279\u5f81\u4e4b\u95f4\u7684\u4ea4\u4e92\uff0c\u5bfc\u81f4\u63cf\u8ff0\u751f\u6210\u4e0d\u51c6\u786e\u3002\u53d7 [19] \u548c [27] \u7684\u542f\u53d1\uff0c\u672c\u6587\u63d0\u51fa\u4e00\u79cd\u57fa\u4e8e Transformer \u7684\u65b0\u578b ICNET \u6846\u67b6\u6765\u89e3\u51b3\u8fd9\u4e00\u95ee\u9898\u3002\u8be5\u6846\u67b6\u901a\u8fc7 CMN \u83b7\u53d6\u9ad8\u5c42\u6982\u5ff5\uff0c\u5e76\u501f\u52a9 CIM \u878d\u5408\u8fd9\u4e9b\u4ea4\u4e92\u7684\u4f5c\u7528\uff1b\u6b64\u5916\uff0c\u5728\u63d0\u53d6\u7684\u5168\u5c40\u8868\u5f81\u4e2d\u5f15\u5165 GFE \u6a21\u5757\uff0c\u4ee5\u589e\u5f3a\u6a21\u578b\u7684\u5168\u5c40\u533a\u5206\u6027\u8868\u5f81\u80fd\u529b\u3002<\/p>\n<\/details>\n\n\n\n<h1 class=\"wp-block-heading\">3 \u65b9\u6cd5\u539f\u7406<\/h1>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u6240\u63d0 ICNET \u7684\u67b6\u6784\u5982\u56fe 1 \u6240\u793a\uff0c\u6574\u4e2a\u6a21\u578b\u4e3b\u8981\u7531\u5c40\u90e8\u4e0e\u5168\u5c40\u7279\u5f81\u63d0\u53d6\u6a21\u5757\u3001CMN \u6a21\u5757\u3001GFE \u6a21\u5757\u4ee5\u53ca\u914d\u5907 CIM \u7684 Transformer \u7ec4\u6210\u3002ICNET \u4e3b\u8981\u901a\u8fc7 CMN \u6784\u5efa\u6982\u5ff5\uff0c\u5e76\u5c06\u5176\u8f93\u5165\u5230 CIM \u4e2d\u4e0e\u8bcd\u5d4c\u5165\u8fdb\u884c\u4ea4\u4e92\uff0c\u4ee5\u63d0\u5347 Transformer \u5bf9\u5173\u952e\u76ee\u6807\u7684\u6355\u6349\u80fd\u529b\u3002\u8be5\u57fa\u4e8e Transformer \u7684\u7f51\u7edc\u9996\u5148\u5229\u7528 ViT \u63d0\u53d6\u56fe\u50cf\u7684\u5168\u5c40\u7279\u5f81\uff0c\u7136\u540e\u901a\u8fc7\u4ea4\u53c9\u6ce8\u610f\u529b\u673a\u5236\u5bf9\u5173\u7cfb\u8fdb\u884c\u5efa\u6a21\uff0c\u6700\u7ec8\u751f\u6210\u63cf\u8ff0\u6587\u672c\u3002\u672c\u8282\u5c06\u8be6\u7ec6\u9610\u8ff0\u5168\u5c40\u4e0e\u5c40\u90e8\u7279\u5f81\u63d0\u53d6\u3001CMN\u3001GFE\u3001\u914d\u5907 CIM \u7684 Transformer \u4ee5\u53ca\u6a21\u578b\u8bad\u7ec3\u8fc7\u7a0b\u3002<\/p>\n<\/details>\n\n\n\n<figure class=\"wp-block-image size-full\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-39.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1011\" height=\"540\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-39.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-733\"  sizes=\"auto, (max-width: 1011px) 100vw, 1011px\" \/><\/div><\/figure>\n\n\n\n<p>\u56fe1. \u6240\u63d0 ICNET \u6a21\u578b\u6846\u67b6\u56fe\u3002\u5c40\u90e8\u4e0e\u5168\u5c40\u7279\u5f81\u63d0\u53d6\u6a21\u5757\u5206\u522b\u91c7\u7528\u9884\u8bad\u7ec3\u7684 CNN\uff08\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\uff09\u548c ViT\uff08\u89c6\u89c9 Transformer\uff09\u56fe\u50cf\u7f16\u7801\u5668\uff0c\u4ece\u8f93\u5165\u56fe\u50cf\u4e2d\u63d0\u53d6\u591a\u5c3a\u5ea6\u5c40\u90e8\u7279\u5f81\u548c\u5168\u5c40\u7279\u5f81\u3002CMN\uff08\u6982\u5ff5\u6620\u5c04\u7f51\u7edc\uff09\u5c06\u591a\u5c3a\u5ea6\u5c40\u90e8\u7279\u5f81\u6620\u5c04\u5230\u9ad8\u5c42\u8bed\u4e49\u6982\u5ff5\u3002GFE\uff08\u5168\u5c40\u7279\u5f81\u589e\u5f3a\uff09\u6a21\u5757\u4e30\u5bcc\u5168\u5c40\u7279\u5f81\u8868\u793a\u3002\u5e26\u6709 CIM\uff08\u6982\u5ff5\u4ea4\u4e92\u6a21\u5757\uff09\u7684 Transformer \u4fc3\u8fdb\u7279\u5f81\u5bf9\u9f50\uff0c\u5e76\u751f\u6210\u5177\u6709\u6070\u5f53\u7c7b\u522b\u548c\u5173\u8054\u5173\u7cfb\u7684\u63cf\u8ff0\u6587\u672c\u3002<\/p>\n\n\n\n<p><strong>ICNET\u7684\u6574\u4f53\u67b6\u6784\u53ef\u5206\u4e3a\u56db\u4e2a\u9636\u6bb5\uff1a\u5c40\u90e8-\u5168\u5c40\u7279\u5f81\u63d0\u53d6&#8212;&gt;\u6982\u5ff5\u6620\u5c04\u7f51\u7edc&#8212;&gt;\u5168\u5c40\u7279\u5f81\u589e\u5f3a&#8212;&gt;\u5e26CIM\u7684Transformer<\/strong><\/p>\n\n\n\n<p><strong>\u5c40\u90e8-\u5168\u5c40\u7279\u5f81\u63d0\u53d6\u6a21\u5757<\/strong>\uff0c\u5176\u6838\u5fc3\u529f\u80fd\u5373\u4e3a\u540c\u65f6\u83b7\u53d6\u9065\u611f\u56fe\u50cf\u7684\u201c\u5c40\u90e8\u76ee\u6807\u7279\u5f81\u201d\u4e0e\u201c\u5168\u5c40\u5173\u7cfb\u7279\u5f81\u201d\uff0c\u5177\u4f53\u7ec6\u8282\uff1a\u5c40\u90e8\u7279\u5f81\u65b9\u9762\uff0c\u7528\u9884\u8bad\u7ec3ResNet-152\uff08CNN\uff09\u63d0\u53d6\u4e09\u4e2a\u4e0d\u540c\u5206\u8fa8\u7387\u7684\u7279\u5f81\u56fe\uff0828\u00d728\u300114\u00d714\u30017\u00d77\uff09\uff0c\u805a\u7126\u5927\u3001\u4e2d\u3001\u5c0f\u5c3a\u5ea6\u76ee\u6807\uff1b\u5168\u5c40\u7279\u5f81\u65b9\u9762\uff0c\u7528\u9884\u8bad\u7ec3ViT\uff08CLIP-ViT-b\/32 backbone\uff09\uff08\u89c6\u89c9Transformer\uff09\u63d0\u53d6\u5168\u5c40\u8bed\u4e49\uff0c\u4fdd\u8bc1\u5e73\u79fb\/\u65cb\u8f6c\u4e0d\u53d8\u6027\u3002\u6355\u6349\u76ee\u6807\u95f4\u7684\u6574\u4f53\u5173\u7cfb\u3002<\/p>\n\n\n\n<p><strong>\u6982\u5ff5\u6620\u5c04\u7f51\u7edc<\/strong>\uff0c\u5176\u6838\u5fc3\u529f\u80fd\u5373\u4e3a\u5c06\u201c\u5c40\u90e8\u7279\u5f81\u201d\u8f6c\u5316\u4e3a\u201c\u9ad8\u5c42\u8bed\u4e49\u6982\u5ff5\u201d\uff0c\u7f29\u5c0f\u89c6\u89c9-\u6587\u672c\u8bed\u4e49\u9e3f\u6c9f\uff0c\u5efa\u7acb\u56fe\u50cf\u7279\u5f81\u4e0e\u8bed\u4e49\u6982\u5ff5\u7684\u6620\u5c04\u5173\u7cfb\u3002\u5176\u4e2d\u5305\u542b<strong>\u7a7a\u95f4\u6ce8\u610f\u529bSA<\/strong>\u5bf9\u7279\u5f81\u56feF2\u8fdb\u884c\u5904\u7406\uff0c\u751f\u6210\u4f4d\u7f6e\u6743\u91cd\uff08postion weights\uff09\uff0c\u6355\u6349\u7a7a\u95f4\u7ef4\u5ea6\u7684\u91cd\u8981\u6027\uff0c<strong>1\u00d71\u5377\u79ef<\/strong>\u5bf9\u4e8eF3\u548cF4\u7279\u5f81\u8fdb\u884c\u901a\u9053\u8c03\u6574\uff0c\u5176\u4e2dF4\u8fd8\u7ecf\u8fc7\u50cf\u7d20\u6ce8\u610f\u529bPA\u548c\u591a\u5c42\u7f51\u7edc\u5904\u7406\u751f\u6210\u8bed\u4e49\u6743\u91cd\uff08semantic weights\uff09\uff0c\u7ecf\u8fc71\u00d71\u5377\u79ef\u5904\u7406\u540e\u7684F3\u7279\u5f81\u4e0e\u4f4d\u7f6e\u6743\u91cd\u8fdb\u884c\u7279\u5f81\u878d\u5408\uff0c\u52a0\u5f3a\u7a7a\u95f4-\u8bed\u4e49\u5173\u8054\u6027\u4f18\u5316\u7279\u5f81\u8868\u5f81\u3002\u4f4d\u7f6e\u6743\u91cd\u4e0e\u8bed\u4e49\u6743\u91cd\u901a\u8fc7\u52a0\u6743\u6620\u5c04\u878d\u5408\uff0c\u518d\u7ecf\u8fc7\u5168\u5c40\u5e73\u5747\u6c60\u5316\u64cd\u4f5c\u5f97\u5230\u591a\u7ec4\u7279\u5f81\u5411\u91cfL\uff0c\u540c\u65f6\u4f4d\u7f6e\u4fe1\u606f\u901a\u8fc7\u4f4d\u7f6e\u5171\u4eab\uff08PS\uff09\u6a21\u5757\u5168\u9012\u7ed9\u540e\u7eed\u6a21\u5757\u3002<\/p>\n\n\n\n<p><strong>\u5168\u5c40\u7279\u5f81\u589e\u5f3a<\/strong>\u9488\u5bf9ViT\u5904\u7406\u5f97\u5230\u7684\u5168\u5c40\u7279\u5f81\u7279\u5f81\u56fe\u8fdb\u884c\u589e\u5f3a\uff0c\u5177\u4f53\u800c\u8a00\uff1a\u4e24\u4e2aLinear\u7ebf\u6027\u5c42\u5206\u522b\u5bf9\u5168\u5c40\u7279\u5f81\u8fdb\u884c\u53d8\u6362\uff0c\u5176\u4e2d\u4e00\u4e2a\u5206\u652f\u53c8\u5f15\u5165\u4e86\u6df1\u5ea6\u53ef\u5206\u79bb\u5377\u79ef\u548csigmod\u6fc0\u6d3b\u51fd\u6570\uff0c\u751f\u6210\u95e8\u63a7\u56e0\u5b50G\uff08Gating Factor\uff09\uff0c\u7136\u540e\u4e24\u4e2a\u5206\u652f\u5904\u7406\u540e\u7684\u7279\u5f81\u901a\u8fc7\u95e8\u63a7\u5355\u5143\u8fdb\u884c\u878d\u5408\uff0c\u5f97\u5230\u589e\u5f3a\u540e\u7684\u7279\u5f81\u3002\uff08\u6ce8\u610f\u529b\u95e8\u80fd\u591f\u9009\u62e9\u8f93\u51fa\u7279\u5f81\u56fe\u4e2d\u91cd\u8981\u7684\u90e8\u5206\uff09\u5c06\u589e\u5f3a\u540e\u7684\u7279\u5f81\u4f20\u5165\u5230Transformer\u4e2d\uff0c\u53c2\u4e0e\u540e\u7eed\u56fe\u6587\u4ea4\u4e92\u4e0e\u6587\u672c\u751f\u6210\u3002<\/p>\n\n\n\n<p><strong>\u5e26\u6982\u5ff5\u4ea4\u4e92\u6a21\u5757\u7684Transformer<\/strong>\u8be5\u6a21\u5757\u8d1f\u8d23\u5c06\u56fe\u50cf\u7279\u5f81\u4e0e\u6587\u672c\u751f\u6210\u5173\u8054\uff0c\u5177\u4f53\u800c\u8a00\uff1a\u6587\u672c\u7aef\u7ecf\u8fc7\u8bcd\u5d4c\u5165\u548c\u4f4d\u7f6e\u7f16\u7801\u5f97\u5230\u521d\u59cb\u8868\u793a\uff1b\u56fe\u50cf\u7aef\u7684\u7279\u5f81\u4e0e\u4f4d\u7f6e\u4fe1\u606f\u8f93\u5165\u5230\u6982\u5ff5\u4ea4\u4e92\u6a21\u5757\uff0c\u5b9e\u73b0\u56fe\u6587\u6982\u5ff5\u7684\u4ea4\u4e92\uff0c\u968f\u540e\u5c31\u662f\u8fdb\u5165\u7f16\u7801\u5668Encoder\u548c\u89e3\u7801\u5668Decoder\u7684\u591a\u5c42\u5806\u53e0\u6a21\u5f0f\uff0c\u6700\u7ec8\u751f\u6210\u6587\u672c\u63cf\u8ff0\u6587\u672c\uff08\u5982\uff1a\u5341\u5b57\u8def\u53e3\u5468\u56f4\u6709\u7eff\u6811\u548c\u7eff\u8349\u576a\uff09\uff0c\u5176\u4e2d\u5168\u5c40\u7279\u5f81\u589e\u5f3a\u8f93\u51fa\u7684\u5168\u5c40\u7279\u5f81\u8f93\u5165\u5230Transformer\u7684\u7f16\u7801\u5668\u63d0\u4f9b\u5168\u5c40\u8bed\u4e49\u652f\u6491\uff0c\u5e2e\u52a9\u6a21\u578b\u7406\u89e3\u56fe\u7247\u7684\u6574\u4f53\u5185\u5bb9\uff0c\u5e76\u4e0e\u6587\u672c\u7279\u5f81\u8fdb\u884c\u4ea4\u4e92\u52a9\u529b\u56fe\u6587\u8bed\u4e49\u5bf9\u9f50\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">3.1 \u5168\u5c40\u4e0e\u5c40\u90e8\u7279\u5f81\u8868\u5f81<\/h2>\n\n\n\n<figure class=\"wp-block-image size-full\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-16.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"987\" height=\"722\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-16.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-700\"  sizes=\"auto, (max-width: 987px) 100vw, 987px\" \/><\/div><\/figure>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u5728\u9065\u611f\u56fe\u50cf\u4e2d\uff0c\u76ee\u6807\u901a\u5e38\u8868\u73b0\u51fa\u663e\u8457\u7684\u5c3a\u5ea6\u5dee\u5f02\u3002\u56fe\u50cf\u8868\u5f81\u4f5c\u4e3a\u5c06\u9065\u611f\u56fe\u50cf\u4e2d\u6709\u4ef7\u503c\u7684\u7279\u5f81\u4e0e\u76ee\u6807\u7f16\u7801\u4e3a\u9ad8\u5c42\u7279\u5f81\u56fe\u7684\u8fc7\u7a0b\uff0c\u5bf9\u4e8e\u9065\u611f\u56fe\u50cf\u7406\u89e3\u81f3\u5173\u91cd\u8981\u3002\u9ad8\u8d28\u91cf\u7684\u56fe\u50cf\u8868\u5f81\u6709\u52a9\u4e8e\u63d0\u53d6\u66f4\u6709\u6548\u7684\u5168\u5c40\u548c\u5c40\u90e8\u7279\u5f81\uff0c\u8fdb\u800c\u63d0\u5347\u63cf\u8ff0\u751f\u6210\u6027\u80fd [26]\u3002<\/p>\n\n\n\n<p>\u53d7\u4e0a\u8ff0\u601d\u8def\u542f\u53d1\uff0c\u672c\u6587\u5206\u522b\u5229\u7528 CNN \u548c ViT \u63d0\u53d6\u5c40\u90e8\u7279\u5f81\u4e0e\u5168\u5c40\u7279\u5f81\uff0c\u4ee5\u83b7\u53d6\u9065\u611f\u56fe\u50cf\u7684\u5168\u9762\u8bed\u4e49\u8868\u5f81\u3002\u5176\u4e2d\uff0c\u5c40\u90e8\u7279\u5f81\u7528\u4e8e\u6355\u6349\u9065\u611f\u56fe\u50cf\u4e2d\u7684\u76ee\u6807\u4fe1\u606f\uff0c\u5168\u5c40\u7279\u5f81\u5219\u7528\u4e8e\u6982\u62ec\u6574\u4e2a\u9065\u611f\u56fe\u50cf\u4e2d\u76ee\u6807\u95f4\u7684\u5173\u7cfb\u4fe1\u606f\u3002<\/p>\n\n\n\n<p>\u5bf9\u4e8e\u5168\u5c40\u7279\u5f81\uff0c\u5176\u9700\u5177\u5907\u5e73\u79fb\u548c\u65cb\u8f6c\u4e0d\u53d8\u6027\u3002\u4e3a\u5b9e\u73b0\u8fd9\u4e00\u7279\u6027\uff0c\u4ece\u539f\u59cb RGB \u56fe\u50cf\\(X \\in \\mathbb{R}^{W \\times H \\times 3}\\)\u4e2d\uff0c\u901a\u8fc7\u9884\u8bad\u7ec3\u7684 ViT \u63d0\u53d6\u5168\u5c40\u7279\u5f81\\(F_{1} \\in \\mathbb{R}^{d_{1}}\\)\uff0c\u5176\u4e2d\\(W \\times H\\)\u4e3a\u56fe\u50cf\u5c3a\u5bf8\uff0c3 \u4ee3\u8868 RGB \u5149\u5b66\u901a\u9053\u3002\u8be5\u8fc7\u7a0b\u53ef\u8868\u793a\u4e3a\uff1a\\(F_{1}=\\text{ViT}(X) \\tag{1}\\)<\/p>\n\n\n\n<p>\u5bf9\u4e8e\u5c40\u90e8\u7279\u5f81\uff0c\u4e0d\u540c\u7684\u5c40\u90e8\u7279\u5f81\u9700\u5173\u6ce8\u9065\u611f\u56fe\u50cf\u7684\u4e0d\u540c\u90e8\u5206\u3002\u4e3a\u6b64\uff0c\u901a\u8fc7\u9884\u8bad\u7ec3\u7684 ResNet152 \u56fe\u50cf\u7f16\u7801\u5668\u63d0\u53d6\u6574\u4e2a\u56fe\u50cf\u7684\u5c40\u90e8\u8868\u5f81\uff0c\u5f97\u5230\u56db\u4e2a\u4e0d\u540c\u5206\u8fa8\u7387\u7684\u7279\u5f81\u56fe [12]\u3002\u7531\u4e8e\u8bed\u8a00\u7279\u5f81\u6bd4\u89c6\u89c9\u7279\u5f81\u66f4\u5177\u533a\u5206\u6027\uff0c\u4e24\u79cd\u6a21\u6001\u7279\u5f81\u4e4b\u95f4\u5b58\u5728\u8f83\u5927\u7684\u8bed\u4e49\u9e3f\u6c9f\uff0c\u5c24\u5176\u662f\u5728\u8f83\u4f4e\u5c42\u7ea7 [53]\u3002\u56e0\u6b64\uff0c\u9009\u53d6\u4e09\u4e2a\u76f8\u5bf9\u8f83\u6df1\u7684\u5c42\u7ea7\u4ee5\u5b9e\u73b0\u66f4\u5168\u9762\u7684\u7279\u5f81\u8868\u5f81\uff0c\u5c06\u5176\u8bb0\u4e3a\\(F_{2}, F_{3}, F_{4} \\in \\mathbb{R}^{d_{2}}\\)\uff0c\u5bf9\u5e94\u7684\u5c3a\u5bf8\u5206\u522b\u4e3a 28\u00d728\u300114\u00d714 \u548c 7\u00d77\u3002\u8fd9\u79cd\u9009\u62e9\u6709\u52a9\u4e8e\u89c6\u89c9\u7279\u5f81\u4e0e\u8bed\u8a00\u7279\u5f81\u4e4b\u95f4\u5b9e\u73b0\u66f4\u6709\u6548\u7684\u4ea4\u4e92\u3002\u5377\u79ef\u8fc7\u7a0b\u53ef\u8868\u793a\u4e3a\uff1a\\(F_{2}, F_{3}, F_{4}=\\text{ResNet}(X) \\tag{2}\\)<\/p>\n\n\n\n<p>\u901a\u8fc7\u91c7\u7528\u8fd9\u4e9b\u4e0d\u540c\u5c42\u7ea7\u7684\u7279\u5f81\u8868\u5f81\uff0c\u80fd\u591f\u4ece\u9065\u611f\u56fe\u50cf\u4e2d\u83b7\u53d6\u4e30\u5bcc\u7684\u5c40\u90e8\u548c\u5168\u5c40\u8bed\u4e49\u4fe1\u606f\u3002\u6b64\u5916\uff0c\u89c6\u89c9\u4e0e\u6587\u672c\u7684\u878d\u5408\u8fdb\u4e00\u6b65\u4e30\u5bcc\u4e86\u4e24\u79cd\u6a21\u6001\u7684\u4fe1\u606f\u5d4c\u5165\u3002<\/p>\n<\/details>\n\n\n\n<h2 class=\"wp-block-heading\">3.2 \u6982\u5ff5\u6620\u5c04\u7f51\u7edc CMN<\/h2>\n\n\n\n<figure class=\"wp-block-image size-full\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-17.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"643\" height=\"691\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-17.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-703\"  sizes=\"auto, (max-width: 643px) 100vw, 643px\" \/><\/div><\/figure>\n\n\n\n<figure class=\"wp-block-image size-full\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-18.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"643\" height=\"254\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-18.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-704\"  sizes=\"auto, (max-width: 643px) 100vw, 643px\" \/><\/div><\/figure>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>CMN \u7684\u4e3b\u8981\u76ee\u7684\u662f\u6784\u5efa\u8bed\u4e49\u6982\u5ff5\uff0c\u4ee5\u4fbf\u5728 Transformer \u89e3\u7801\u5668\u90e8\u5206\u4e0e\u8bcd\u5d4c\u5165\u8fdb\u884c\u4ea4\u4e92\uff0c\u5b9e\u73b0\u8de8\u6a21\u6001\u7279\u5f81\u5bf9\u9f50\u3002\u4e3a\u5145\u5206\u805a\u5408\u4e0d\u540c\u5c3a\u5ea6\u5c40\u90e8\u7279\u5f81\u4e2d\u7684\u91cd\u8981\u4fe1\u606f\uff0c\u6240\u63d0 CMN \u91c7\u7528\u52a8\u6001\u901a\u9053\u4e0e\u7a7a\u95f4\u6ce8\u610f\u529b\u673a\u5236\uff0c\u805a\u7126\u5173\u952e\u8bed\u4e49\u4e0e\u7ed3\u6784\u4fe1\u606f\uff1b\u968f\u540e\u5c06\u6743\u91cd\u6620\u5c04\u5230\u90bb\u57df\u7279\u5f81\u56fe\u4e0a\uff0c\u4ee5\u8865\u5145\u4e0d\u540c\u5c42\u7ea7\u7684\u7279\u5f81\u7ec6\u8282\uff1b\u6700\u540e\u901a\u8fc7\u5168\u5c40\u5e73\u5747\u6c60\u5316\u64cd\u4f5c\u8fdb\u4e00\u6b65\u4f18\u5316\u6982\u5ff5\u7279\u5f81\u3002<\/p>\n\n\n\n<p>\u5177\u4f53\u800c\u8a00\uff0c\u9996\u5148\u5bf9 CNN \u63d0\u53d6\u7684\u591a\u5c3a\u5ea6\u5c40\u90e8\u7279\u5f81\\(F_{2}-F_{4}\\)\u91c7\u7528\u591a\u4e2a 1\u00d71 \u5927\u5c0f\u7684\u5c0f\u5377\u79ef\u6838\u8fdb\u884c\u5904\u7406\uff0c\u8be5\u8fc7\u7a0b\u53ef\u8868\u793a\u4e3a\uff08\u5176\u4e2d Conv (\u30fb) \u8868\u793a\u5377\u79ef\u64cd\u4f5c\uff09\uff1a\uff08\u539f\u6587\u6b64\u5904\u516c\u5f0f\u8868\u8ff0\u4e0d\u5b8c\u6574\uff0c\u672a\u660e\u786e\u5199\u51fa\u5177\u4f53\u8868\u8fbe\u5f0f\uff0c\u6545\u4fdd\u6301\u539f\u6837\uff09<\/p>\n\n\n\n<p>\u968f\u540e\uff0c\u5c06\u7279\u5f81\u56fe\u8c03\u6574\u4e3a\u4e0e\\(F_{3}\\)\u76f8\u540c\u7684\u5c3a\u5bf8\\(M \\times K\\)\uff0c\u4e14\u901a\u9053\u6570\u4e3a 512\u3002\u5982\u56fe 1 \u6240\u793a\uff0c\u8bbe\u6d45\u5c42\u3001\u4e2d\u5c42\u548c\u6df1\u5c42\u7279\u5f81\u56fe\u5206\u522b\u4e3a\\(F_{2}&#8217; &#8211; F_{4}&#8217;\\)\u3002\u5176\u4e2d\uff0c\\(F_{2}&#8217;\\)\u5305\u542b\u4e30\u5bcc\u7684\u7ec6\u8282\u7279\u5f81\uff0c\u6709\u52a9\u4e8e\u6a21\u578b\u8bc6\u522b\u663e\u8457\u76ee\u6807\u3002\u4e3a\u83b7\u53d6\u9690\u85cf\u7279\u5f81\u56fe\u7684\u4f4d\u7f6e\u6743\u91cd\uff0c\u5148\u5728\u901a\u9053\u7ef4\u5ea6\u4e0a\u8fdb\u884c\u5e73\u5747\u6c60\u5316\u64cd\u4f5c\uff0c\u518d\u6267\u884c sigmoid \u64cd\u4f5c\uff1b\u7136\u540e\u901a\u8fc7\u77e9\u9635\u4e58\u6cd5\u5c06\u4f4d\u7f6e\u6743\u91cd\u7528\u4e8e\\(F_{2}&#8217;\\)\u4e0e\\(F_{3}&#8217;\\)\u7684\u878d\u5408\u3002\u901a\u8fc7\u8fd9\u79cd\u65b9\u5f0f\uff0c\u80fd\u591f\u57fa\u4e8e\u4f4d\u7f6e\u6ce8\u610f\u529b\u5b9e\u73b0\u4e24\u4e2a\u7279\u5f81\u56fe\u7684\u4fe1\u606f\u878d\u5408\uff0c\u589e\u5f3a\u878d\u5408\u7279\u5f81\u56fe\u4e2d\u7eb9\u7406\u7279\u5f81\u7684\u8868\u5f81\u3002\u8be5\u8fc7\u7a0b\u53ef\u5b9a\u4e49\u4e3a\uff1a\\(F_{3}&#8221;=\\frac{1}{C} \\sum_{k=0}^{C} F_{2[i, j, k]}&#8217; \\odot F_{3}&#8217;\\)\u5176\u4e2d\uff0cC \u8868\u793a\u901a\u9053\u6570\uff0c\\(\\odot\\)\u8868\u793a\u77e9\u9635\u4e58\u6cd5\u3002<\/p>\n\n\n\n<p>\u6b64\u5916\uff0c\u6df1\u5c42\u7279\u5f81\u56fe\\(F_{4}&#8217;\\)\u80fd\u591f\u6355\u6349\u5bc6\u96c6\u4e14\u5927\u5c3a\u5ea6\u7684\u76ee\u6807\u3002\u50cf\u7d20\u6ce8\u610f\u529b\u901a\u5e38\u5e94\u7528\u4e8e\u56fe\u50cf\u7684\u6700\u7ec6\u7c92\u5ea6\u5c42\u7ea7\uff0c\u56e0\u6b64\u5728\u6df1\u5c42\u5229\u7528\u50cf\u7d20\u6ce8\u610f\u529b\u751f\u6210 N \u4e2a\u901a\u9053\u7684\u50cf\u7d20\u6743\u91cd\uff0c\u8be5\u6743\u91cd\u5173\u6ce8\u6bcf\u4e2a\u50cf\u7d20\u7684\u91cd\u8981\u6027\uff0c\u5e76\u901a\u8fc7\u5b66\u4e60\u6bcf\u4e2a\u50cf\u7d20\u76f8\u5bf9\u4e8e\u5176\u4ed6\u50cf\u7d20\u7684\u91cd\u8981\u6027\u6765\u8fdb\u884c\u7279\u5f81\u63d0\u53d6\uff0c\u8fd9\u4e00\u8fc7\u7a0b\u6709\u52a9\u4e8e\u89e3\u7801\u5668\u5b66\u4e60\u9065\u611f\u56fe\u50cf\u4e2d\u7684\u76ee\u6807\u3002\u57fa\u4e8e\u6b64\uff0c\u5bf9\\(F_{4}&#8217;\\)\u5e94\u7528 N \u4e2a\u8f6f\u6ce8\u610f\u529b\u6a21\u5757\uff0c\u751f\u6210\u591a\u4e2a\u5173\u952e\u8bed\u4e49\u6743\u91cd\u3002\u6784\u5efa N \u4e2a\u7a7a\u95f4\u5c3a\u5bf8\u4e3a\\(M \\times K\\)\u7684\u6ce8\u610f\u529b\u6743\u91cd\u56fe\u662f\u5b9e\u73b0\u8fd9\u4e00\u8fc7\u7a0b\u7684\u524d\u63d0\uff0c\u5c06\u5176\u8bb0\u4e3a\\(\\gamma \\in \\mathbb{R}^{M \\times K \\times N}\\)\u3002\u7b2c n \u4e2a\u6ce8\u610f\u529b\u56fe\u4e2d\u4f4d\u7f6e\\((i, j)\\)\u5904\u7684\u6743\u91cd\u7cfb\u6570\u8ba1\u7b97\u5982\u4e0b\uff1a\\(\\gamma_{[i, j, n]}=\\frac{1}{C} \\sum_{k=0}^{C} F_{4[i, j, k]}&#8217; \\times w_{[n, k]}\\)\u5176\u4e2d\uff0c\\(\\gamma_{[i, j, n]}\\)\u4e3a\u4e2d\u95f4\u6743\u91cd\u56fe\uff0c\u7528\u4e8e\u5c06 C \u4e2a\u7279\u5f81\u901a\u9053\u6620\u5c04\u5230 N \u4e2a\u6ce8\u610f\u529b\u901a\u9053\u3002<\/p>\n\n\n\n<p>\u4e3a\u80fd\u66f4\u597d\u5730\u4e0e Transformer \u89e3\u7801\u5668\u4e2d\u7684\u8bcd\u5d4c\u5165\u5411\u91cf\u5bf9\u9f50\uff0c\u5e76\u907f\u514d\u5f15\u5165\u4eba\u5de5\u53c2\u6570\uff0c\u672c\u6587\u63d0\u51fa\u4f4d\u7f6e\u5171\u4eab\u673a\u5236\uff0c\u5229\u7528\u8bcd\u5d4c\u5165\u5411\u91cf\u7684\u957f\u5ea6\u5bf9\u6ce8\u610f\u529b\u6743\u91cd N \u8fdb\u884c\u5bf9\u9f50\uff1b\u968f\u540e\u901a\u8fc7 SoftMax \u64cd\u4f5c\u5bf9\u5176\u8fdb\u884c\u7f29\u653e\uff0c\u5f97\u5230\\(\\bar{\\gamma}_{[i, j, n]}\\)\uff0c\u5177\u4f53\u8ba1\u7b97\u5982\u4e0b\uff1a\\(\\overline{\\gamma}_{[i, j, n]}=\\frac{\\exp \\left(\\gamma_{[i, j, n]}\\right)}{\\sum_{i=0}^{M} \\sum_{j=0}^{K} \\exp \\left(\\gamma_{[i, j, n]}\\right)}\\)<\/p>\n\n\n\n<p>\u6700\u540e\uff0c\u5229\u7528\u5168\u5c40\u5e73\u5747\u6c60\u5316\u5c42\u5c06\u5c40\u90e8\u7279\u5f81\u56fe\u805a\u5408\u4e3a\u591a\u4e2a\u6982\u5ff5\u7279\u5f81\u5411\u91cf\u3002L \u4e2d\u7b2c n \u4e2a\u5411\u91cf\u7684\u7b2c k \u4e2a\u5143\u7d20\u8ba1\u7b97\u5982\u4e0b\uff1a\\(L_{[n, k]}=\\frac{1}{H \\times W} \\sum_{i=0}^{H} \\sum_{j=0}^{W} F_{3[i, j, k]}&#8221; \\otimes \\overline{\\gamma}_{[i, j, n]}\\)\u5176\u4e2d\uff0c\\(\\otimes\\)\u8868\u793a\u54c8\u8fbe\u739b\u79ef\u3002<\/p>\n<\/details>\n\n\n\n<h2 class=\"wp-block-heading\">3.3 \u5168\u5c40\u7279\u5f81\u589e\u5f3a GFE<\/h2>\n\n\n\n<figure class=\"wp-block-image size-full is-resized\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-19.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"646\" height=\"332\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-19.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-705\" style=\"width:840px;height:auto\"  sizes=\"auto, (max-width: 646px) 100vw, 646px\" \/><\/div><\/figure>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u4e3a\u63d0\u5347\u5168\u5c40\u8bed\u4e49\u4fe1\u606f\u7684\u533a\u5206\u80fd\u529b\uff0c\u8fdb\u800c\u83b7\u5f97\u66f4\u5168\u9762\u3001\u4e30\u5bcc\u7684\u7279\u5f81\u8868\u5f81\uff0c\u672c\u6587\u9996\u5148\u5bf9\u5168\u5c40\u7279\u5f81\\(F_{1}\\)\u4e2d\u65e0\u5173\u6216\u5f31\u76f8\u5173\u7684\u5185\u5bb9\u8fdb\u884c\u8fc7\u6ee4\u3002\u5177\u4f53\u800c\u8a00\uff0c\u91c7\u7528 3\u00d73 \u6df1\u5ea6\u53ef\u5206\u79bb\u5377\u79ef\u5757\u5c06\u5176\u8f6c\u6362\u4e3a\u57fa\u4e8e\u8fd1\u90bb\u7279\u5f81\u7684\u95e8\u63a7\u901a\u9053\u6ce8\u610f\u529b\u673a\u5236\uff0c\u4f7f\u5176\u7ed3\u6784\u4e0e\u95e8\u63a7\u901a\u9053\u6ce8\u610f\u529b\u7684\u8bbe\u8ba1\u7406\u5ff5\u4fdd\u6301\u4e00\u81f4\uff0c\u8be5\u8fc7\u7a0b\u53ef\u8868\u793a\u4e3a\uff08\u539f\u6587\u6b64\u5904\u516c\u5f0f\u8868\u8ff0\u4e0d\u5b8c\u6574\uff0c\u672a\u660e\u786e\u5199\u51fa\u5177\u4f53\u8868\u8fbe\u5f0f\uff0c\u6545\u4fdd\u6301\u539f\u6837\uff09\uff1a<\/p>\n\n\n\n<p>\u968f\u540e\uff0c\u5f15\u5165\u57fa\u4e8e sigmoid \u6fc0\u6d3b\u51fd\u6570\u7684\u6ce8\u610f\u529b\u95e8\\(G \\in \\mathbb{R}^{N \\times d}\\)\uff0c\u5176\u8ba1\u7b97\u5982\u4e0b\uff1a\\(G=\\sigma\\left(\\hat{F}_{1} W_{g}+b_{g}\\right)\\)\u5176\u4e2d\uff0c\\(W_{g} \\in \\mathbb{R}^{d \\times d}\\)\u4e3a\u53ef\u5b66\u4e60\u53c2\u6570\u77e9\u9635\uff0c\\(b_{g} \\in \\mathbb{R}^{d}\\)\u4e3a\u504f\u7f6e\u9879\uff0c\\(\\sigma(\\cdot)\\)\u4e3a sigmoid \u6fc0\u6d3b\u51fd\u6570\u3002\u6ce8\u610f\u529b\u95e8 G \u80fd\u591f\u9009\u62e9\u6027\u5730\u7a81\u51fa\\(F_{1}\\)\u4e2d\u7684\u91cd\u8981\u4fe1\u606f\uff0c\u4e3a\u56fe\u50cf\u63d0\u4f9b\u66f4\u5177\u533a\u5206\u6027\u7684\u8868\u5f81\u3002\u57fa\u4e8e\u8be5\u95e8\u63a7\uff0c\u4f18\u5316\u540e\u7684\u7279\u5f81\\(\\tilde{F}_{1} \\in \\mathbb{R}^{d_{1}}\\)\u53ef\u901a\u8fc7\uff08\u539f\u6587\u6b64\u5904\u516c\u5f0f\u8868\u8ff0\u4e0d\u5b8c\u6574\uff0c\u672a\u660e\u786e\u5199\u51fa\u5177\u4f53\u8868\u8fbe\u5f0f\uff0c\u6545\u4fdd\u6301\u539f\u6837\uff09\u8ba1\u7b97\u5f97\u5230\uff0c\u5176\u4e2d\\(\\otimes\\)\u8868\u793a\u54c8\u8fbe\u739b\u79ef\u3002\u4e4b\u540e\uff0c\u5c06\u4f18\u5316\u540e\u7684\u7279\u5f81\\(\\tilde{F}_{1}\\)\u8f93\u5165\u5230 Transformer \u4e2d\u3002<\/p>\n<\/details>\n\n\n\n<h2 class=\"wp-block-heading\">3.4 \u914d\u5907CIM\u7684Transformer<\/h2>\n\n\n\n<figure class=\"wp-block-image size-full\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-20.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"647\" height=\"77\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-20.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-706\"  sizes=\"auto, (max-width: 647px) 100vw, 647px\" \/><\/div><\/figure>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u4e3a\u5145\u5206\u5229\u7528\u8bed\u4e49\u6982\u5ff5\u4e0e\u5168\u5c40\u7279\u5f81\uff0c\u672c\u6587\u8bbe\u8ba1\u7684 Transformer \u7531\u7f16\u7801\u5668\u3001\u89e3\u7801\u5668\u548c CIM \u7ec4\u6210\uff0c\u5982\u56fe 2 \u6240\u793a\u3002\u589e\u5f3a\u540e\u7684\u5168\u5c40\u7279\u5f81\\(\\tilde{F}_{1}\\)\u4ee5\u6807\u51c6\u65b9\u5f0f\u8f93\u5165\u5230 Transformer \u4e2d\uff0c\u63d0\u5347\u6a21\u578b\u5bf9\u5168\u5c40\u5173\u7cfb\u7684\u5efa\u6a21\u80fd\u529b\uff1b\u5bf9\u4e8e\u8bed\u4e49\u6982\u5ff5\uff0c\u901a\u8fc7\u6240\u63d0 CIM \u589e\u5f3a\u5176\u4e0e\u8bcd\u5d4c\u5165\u5411\u91cf\u7684\u5173\u8054\u6027\uff0cCIM \u76f4\u63a5\u8fde\u63a5\u5728\u89e3\u7801\u5668\u7684\u5934\u90e8\uff0c\u6700\u7ec8\u4e24\u8005\u5171\u540c\u7528\u4e8e\u8bcd\u6c47\u9884\u6d4b<\/p>\n<\/details>\n\n\n\n<figure class=\"wp-block-image size-full\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-21.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"652\" height=\"309\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-21.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-707\"  sizes=\"auto, (max-width: 652px) 100vw, 652px\" \/><\/div><\/figure>\n\n\n\n<figure class=\"wp-block-image size-full is-resized\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-53.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"462\" height=\"403\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-53.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-798\" style=\"width:840px;height:auto\"  sizes=\"auto, (max-width: 462px) 100vw, 462px\" \/><\/div><\/figure>\n\n\n\n<p>\u56fe 2. \u5e26\u6709\u6982\u5ff5\u4ea4\u4e92\u6a21\u5757\uff08CIM\uff09\u7684 Transformer \u67b6\u6784\u3002\u5728\u539f\u59cb Transformer [29] \u7684\u57fa\u7840\u4e0a\uff0c\u89e3\u7801\u5668\u7684\u524d\u7aef\u589e\u8bbe\u4e86\u4e00\u4e2a\u6982\u5ff5\u4ea4\u4e92\u6a21\u5757\u3002<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">3.4.1 \u6982\u5ff5\u4ea4\u4e92\u6a21\u5757 CIM<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u7ed9\u5b9a\u751f\u6210\u7684\u8bed\u53e5 S\uff0c\u901a\u8fc7 word2vec [64] \u8bcd\u5d4c\u5165\u65b9\u6cd5\u83b7\u53d6\u5305\u542b\u4f4d\u7f6e\u7f16\u7801\u7684\u7279\u5f81\u5411\u91cf E\uff0c\u5176\u8868\u793a\u5982\u4e0b\uff08\u539f\u6587\u6b64\u5904\u516c\u5f0f\u8868\u8ff0\u4e0d\u5b8c\u6574\uff0c\u672a\u660e\u786e\u5199\u51fa\u5177\u4f53\u8868\u8fbe\u5f0f\uff0c\u6545\u4fdd\u6301\u539f\u6837\uff09\uff1a<\/p>\n\n\n\n<p>\u8bed\u8a00\u751f\u6210\u8fc7\u7a0b\u9700\u8981\u5728\u8bed\u4e49\u6982\u5ff5\u7684\u5f15\u5bfc\u4e0b\u8fdb\u884c\uff0c\u56e0\u6b64\u57fa\u4e8e\u6ce8\u610f\u529b\u673a\u5236\uff0c\u5f15\u5165\u524d\u7f00\u56e0\u679c\u8bed\u8a00\u5efa\u6a21\uff08PCLM\uff09\u3002\u8be5\u5efa\u6a21\u65b9\u6cd5\u501f\u9274 UniLM [65] \u4e2d\u524d\u7f00\u7684\u4f7f\u7528\u601d\u8def\uff0c\u901a\u8fc7\u9002\u914d\u7ed9\u5b9a\u7684\u56fe\u50cf\u524d\u7f00\u6765\u751f\u6210\u6587\u672c\uff0c\u5f97\u5230\u7684\u603b\u6ce8\u610f\u529b\u6743\u91cd\u5982\u4e0b\uff1a\\(\\hat{X}=\\text{SoftMax}\\left(\\frac{\\left(E W_{q}\\right)\\left(L W_{k}\\right)^{T}}{\\sqrt{d}}\\right) \\times\\left(L W_{v}\\right)\\)\u5176\u4e2d\uff0cE \u4e3a\u8bcd\u5411\u91cf\uff0cL \u4e3a\u6982\u5ff5\u5411\u91cf\uff0c\\(W_{q}\\)\u3001\\(W_{k}\\)\u548c\\(W_{v}\\)\u4e3a\u53ef\u5b66\u4e60\u53c2\u6570\u77e9\u9635\uff0cd \u4e3a\u7f29\u653e\u56e0\u5b50\u3002<\/p>\n\n\n\n<p>CIM \u91c7\u7528\u70b9\u79ef\u8fd0\u7b97\u8ba1\u7b97\u5411\u91cf E \u4e0e L \u4e4b\u95f4\u7684\u76f8\u4f3c\u5ea6\u5206\u5e03\uff0c\u5e76\u901a\u8fc7\u5e26\u6709\u76f8\u4f3c\u5ea6\u5ea6\u91cf\u7684\u5411\u91cf L \u8fdb\u4e00\u6b65\u8ba1\u7b97\u52a0\u6743\u548c\\(\\hat{X}\\)\u3002\u5982\u56fe 3 \u6240\u793a\uff0c\u6bcf\u4e2a\u6587\u672c\u6807\u8bb0\u5d4c\u5165\u90fd\u80fd\u5904\u7406\u6240\u6709\u6982\u5ff5\u53ca\u5176\u4e4b\u524d\u7684\u6587\u672c\u6807\u8bb0\u5d4c\u5165\u3002PCLM \u4e0e\u4f20\u7edf\u56e0\u679c\u8bed\u8a00\u5efa\u6a21 [66] \u7c7b\u4f3c\uff0c\u6a21\u578b\u5728\u751f\u6210\u6587\u672c\u65f6\uff0c\u6bcf\u6b21\u4ec5\u751f\u6210\u4e00\u4e2a\u6807\u8bb0\uff0c\u4e14\u751f\u6210\u8fc7\u7a0b\u4f9d\u8d56\u4e8e\u4e4b\u524d\u7684\u6807\u8bb0\u3002\u9700\u8981\u6ce8\u610f\u7684\u662f\uff0c\u4f5c\u4e3a\u524d\u7f00\u6dfb\u52a0\u7684\u6982\u5ff5\u5e76\u4e0d\u53c2\u4e0e PCLM \u635f\u5931\u7684\u8ba1\u7b97\uff0c\u8be5\u635f\u5931\u672c\u8d28\u4e0a\u4ecd\u662f\u5e8f\u5217\u5230\u5e8f\u5217\u7684\u8bed\u8a00\u5efa\u6a21\u635f\u5931\u3002<\/p>\n<\/details>\n\n\n\n<figure class=\"wp-block-image size-full\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-54.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"647\" height=\"256\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-54.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-799\"  sizes=\"auto, (max-width: 647px) 100vw, 647px\" \/><\/div><\/figure>\n\n\n\n<p>\u56fe 3. \u6982\u5ff5\u4ea4\u4e92\u6a21\u5757\uff08CIM\uff09\u793a\u610f\u56fe\u3002\uff08a\uff09\u8bcd\u5d4c\u5165\u5411\u91cf\u4e0e\u8bed\u4e49\u6982\u5ff5\u4e4b\u95f4\u7684\u4ea4\u4e92\u8fc7\u7a0b\u3002\uff08b\uff09PCLM\uff08\u53ef\u80fd\u4e3a \u201c\u6982\u5ff5\u5f15\u5bfc\u7684\u63a9\u7801\u8bed\u8a00\u6a21\u578b\u201d \u6216\u7c7b\u4f3c\u7f29\u5199\uff09\uff0c\u5176\u4e2d\u5e26\u63a9\u7801\u7684\u65b9\u5757\u4ee3\u8868\u88ab\u63a9\u76d6\u7684\u6587\u672c\u6807\u8bb0\uff0c\u8fd9\u4e9b\u6807\u8bb0\u53ef\u5173\u6ce8\u6240\u6709\u8bed\u4e49\u6982\u5ff5\u53ca\u524d\u6587\u7684\u6587\u672c\u6807\u8bb0\u3002\uff08\u6ce8\uff1aPCLM \u7684\u5168\u79f0\u9700\u7ed3\u5408\u8bba\u6587\u4e0a\u4e0b\u6587\u786e\u8ba4\uff0c\u6b64\u5904\u57fa\u4e8e\u63a9\u7801\u673a\u5236\u548c\u8bed\u4e49\u6982\u5ff5\u7684\u5173\u8054\u63a8\u6d4b\u5176\u6838\u5fc3\u529f\u80fd\u662f\u901a\u8fc7\u63a9\u7801\u8bad\u7ec3\u8ba9\u6587\u672c\u6807\u8bb0\u4e0e\u8bed\u4e49\u6982\u5ff5\u5efa\u7acb\u5173\u8054\uff0c\u589e\u5f3a\u56fe\u6587\u6982\u5ff5\u7684\u4ea4\u4e92\u80fd\u529b\u3002\uff09<\/p>\n\n\n\n<p><strong>\u8bcd\u5d4c\u5165\u5411\u91cf\u4e0e\u8bed\u4e49\u6982\u5ff5\u7684\u4ea4\u4e92<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u8f93\u5165<\/strong>E\u5373\u8bcd\u5d4c\u5165\u5411\u91cf\uff0c\u6587\u672c\u7aef\u7684\u7279\u5f81\u8868\u793a\uff0c\u4f5c\u4e3a\u67e5\u8be2Q\u8f93\u5165\uff0cL\u4e3a\u8bed\u4e49\u6982\u5ff5\u8f93\u5165\uff0c\u5373\u6765\u81ea\u56fe\u50cf\u7aef\u7684\u6982\u5ff5\u6620\u5c04\u8f93\u51fa\uff0c\u4f5c\u4e3a\u952eK\u548c\u503cV\u8f93\u5165\uff0c\u4e09\u8005\u8f93\u5165\u5230\u6ce8\u610f\u529b\u673a\u5236\u4e2d\u3002\uff08\u5f97\u5230\u878d\u5408\u4e86\u56fe\u50cf\u7279\u5f81\u7684\u6587\u672c\u7279\u5f81\uff09<\/li>\n\n\n\n<li><strong>\u6ce8\u610f\u529b\u673a\u5236<\/strong>\u901a\u8fc7\u7ecf\u5178\u7684\u6ce8\u610f\u529b\u673a\u5236Attention(Q,K,V)\uff0c\u8ba9\u8bcd\u5d4c\u5165\u5411\u91cf\u4e0e\u8bed\u8a00\u6982\u5ff5\u7279\u5f81\u8fdb\u884c\u4ea4\u4e92\uff0c\u6316\u6398\u4e24\u8005\u7684\u8bed\u4e49\u5173\u8054\u3002<\/li>\n\n\n\n<li><strong>\u8f93\u51fa<\/strong>\u751f\u6210\u878d\u5408\u540e\u7684\u7279\u5f81X\uff0c\u5b8c\u6210\u6587\u672c\u4e0e\u56fe\u50cf\u6982\u5ff5\u7684\u521d\u6b65\u5bf9\u9f50\u3002<\/li>\n<\/ul>\n\n\n\n<p><strong>PCLM\uff08\u6982\u5ff5\u5f15\u5bfc\u7684\u63a9\u7801\u8bed\u8a00\u6a21\u578b\uff09\u673a\u5236<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>\u533a\u57df\u5212\u5206<\/strong>\u5c06\u7279\u5f81\u7a7a\u95f4\u5206\u4e3a\u6982\u5ff5\u4f4d\u7f6eC\u548c\u8bcd\u4f4d\u7f6eW\uff0c\u5206\u522b\u5bf9\u5e94\u56fe\u50cf\u6982\u5ff5\u548c\u6587\u672c\u8bcd\u6c47\u7684\u7279\u5f81\u533a\u57df\uff0c<\/li>\n\n\n\n<li><strong>\u63a9\u7801\u89c4\u5219<\/strong>\u88ab\u63a9\u7801\u7684\u6587\u672c\u6807\u8bb0\uff08masked\uff0c\u68d5\u8272\u65b9\u5757\uff09\u53ef\u4ee5\u5173\u6ce8\u6240\u6709\u8bed\u4e49\u6982\u5ff5C\u533a\u57df\u548c\u524d\u6587\u7684\u6587\u672c\u6807\u8bb0\uff08W\u533a\u57df\u4e2d\u672a\u88ab\u63a9\u7801\u7684\u90e8\u5206\uff09\u672a\u88ab\u63a9\u7801\u7684\u6807\u8bb0\uff08unmasked\uff0c\u767d\u8272\u65b9\u5757\uff09\u5219\u6309\u5e38\u89c4\u6ce8\u610f\u529b\u89c4\u5219\u8ba1\u7b97\u3002<\/li>\n\n\n\n<li><strong>\u529f\u80fd<\/strong>\u901a\u8fc7\u8fd9\u79cd\u63a9\u7801\u8bbe\u8ba1\uff0c\u5f3a\u5236\u6a21\u578b\u5728\u9884\u6d4b\u88ab\u63a9\u7801\u7684\u6587\u672c\u65f6\uff0c\u5fc5\u987b\u7ed3\u5408\u56fe\u50cf\u8bed\u4e49\u6982\u5ff5\u548c\u5df2\u6709\u7684\u6587\u672c\u4fe1\u606f\uff0c\u4ece\u800c\u589e\u5f3a\u56fe\u6587\u6982\u5ff5\u7684\u4ea4\u4e92\u548c\u5bf9\u9f50\uff0c\u63d0\u793a\u6587\u672c\u7684\u8bed\u4e49\u51c6\u786e\u6027\u3002<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">3.4.2 Transformer<\/h3>\n\n\n\n<figure class=\"wp-block-image size-full\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-22.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"651\" height=\"223\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-22.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-708\"  sizes=\"auto, (max-width: 651px) 100vw, 651px\" \/><\/div><\/figure>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u5982\u56fe 2 \u6240\u793a\uff0c\u6240\u63d0 Transformer \u9075\u5faa\u6807\u51c6 Transformer [29] \u7684\u7ed3\u6784\uff0c\u7f16\u7801\u5668\u7531 o \u4e2a\u76f8\u540c\u7684\u5c42\u5806\u53e0\u800c\u6210\u3002\u6bcf\u4e2a\u7f16\u7801\u5c42\u4f9d\u6b21\u5806\u53e0\u81ea\u6ce8\u610f\u529b\u5757\u548c\u4f4d\u7f6e &#8211; wise \u5168\u8fde\u63a5\u524d\u9988\u7f51\u7edc\uff08FFN\uff09\uff0c\u5e76\u5728\u7b2c\u4e00\u4e2a\u548c\u7b2c\u4e8c\u4e2a\u5b50\u5c42\u5468\u56f4\u5206\u522b\u52a0\u5165\u6b8b\u5dee\u8fde\u63a5\u548c\u5c42\u5f52\u4e00\u5316\u3002\u6bcf\u4e2a\u7f16\u7801\u5c42\u7684\u8f93\u5165\u5305\u62ec\u524d\u4e00\u4e2a\u7f16\u7801\u5c42\u7684\u8f93\u51fa\uff0c\u5176\u4e2d\u7b2c\u4e00\u5c42\u7684\u8f93\u5165\u91c7\u7528 GFE \u6a21\u5757\u7684\u8f93\u51fa\\(\\tilde{F}_{1}\\)\u3002\u7ecf\u8fc7 o \u4e2a\u7f16\u7801\u5c42\u540e\uff0c\u5f97\u5230\u7f16\u7801\u5668\u7684\u6700\u7ec8\u8f93\u51fa\\(F_{e}\\)\u3002<\/p>\n\n\n\n<p>\u89e3\u7801\u5668\u540c\u6837\u7531 o \u4e2a\u76f8\u540c\u7684\u89e3\u7801\u5c42\u5806\u53e0\u800c\u6210\uff0c\u6bcf\u4e2a\u89e3\u7801\u5c42\u5305\u542b\u63a9\u7801\u81ea\u6ce8\u610f\u529b\u3001\u4ea4\u53c9\u6ce8\u610f\u529b\u548c FFN \u5c42\uff0c\u4e14\u5728\u6bcf\u4e2a\u5b50\u5c42\u5468\u56f4\u5747\u52a0\u5165\u6b8b\u5dee\u8fde\u63a5\u548c\u5c42\u5f52\u4e00\u5316\u3002\u4e0e\u7f16\u7801\u5668\u7c7b\u4f3c\uff0c\u6bcf\u4e2a\u89e3\u7801\u5c42\u7684\u8f93\u5165\u5305\u62ec\u524d\u4e00\u4e2a\u89e3\u7801\u5c42\u7684\u8f93\u51fa\u548c\u7f16\u7801\u5668\u7684\u8f93\u51fa\\(F_{e}\\)\uff0c\u5176\u4e2d\u7b2c\u4e00\u5c42\u7684\u8f93\u5165\u4e3a\u4ea4\u4e92\u5f0f\u6982\u5ff5\u7f51\u7edc\u7684\u8f93\u51fa\u3002\u6700\u540e\uff0c\u5c06\u6700\u540e\u4e00\u4e2a\u89e3\u7801\u5c42\u7684\u8f93\u51fa\\(F_{d}\\)\u4f5c\u4e3a\u89e3\u7801\u5668\u7684\u8f93\u51fa\uff0c\u5e76\u5c06\u5176\u8f93\u5165\u5230\u5168\u8fde\u63a5\u5c42\u548c SoftMax \u4e2d\uff0c\u8ba1\u7b97\u5728\u8bcd\u6c47\u8868\u6240\u6709\u53ef\u80fd\u8bcd\u6c47\u4e0a\u7684\u6982\u7387\u5206\u5e03\u3002<\/p>\n<\/details>\n\n\n\n<p>1\u3001\u6240\u63d0Transformer\u9075\u5faa\u6807\u51c6\u7684Transformer\u7ed3\u6784\u3002<\/p>\n\n\n\n<p>2\u3001\u7f16\u7801\u5668\u5305\u542b\u81ea\u6ce8\u610f\u529b\u673a\u5236\u548c\u524d\u9988\u795e\u7ecf\u7f51\u7edc\u4e24\u4e2a\u6838\u5fc3\u7ec4\u4ef6\uff0c\u8fd9\u4e24\u4e2a\u7ec4\u4ef6\u88ab\u91cd\u590d\u5806\u53e0o\u6b21\uff0c\u8f93\u5165\u6d41\u7a0b\uff1a\u8f93\u5165\u6570\u636e\uff08\u5982\u56fe\u50cf\u7279\u5f81\uff09\u9996\u5148\u8fdb\u5165\u81ea\u6ce8\u610f\u529b\u6a21\u5757\uff0c\u6355\u6349\u5e8f\u5217\u5185\u90e8\u7684\u4f9d\u8d56\u5173\u7cfb\uff1b\u968f\u540e\u8fdb\u5165\u524d\u9988\u7f51\u7edc\uff0c\u5bf9\u7279\u5f81\u8fdb\u884c\u53d8\u6362\u4e0e\u589e\u5f3a\uff0c\u6bcf\u4e00\u5c42\u7684\u8f93\u51fa\u90fd\u4f1a\u901a\u8fc7\u6b8b\u5dee\u8fde\u63a5\u4e0e\u8f93\u51fa\u8fdb\u884c\u7279\u5f81\u878d\u5408\uff08\u7f13\u89e3\u6df1\u5c42\u7f51\u7edc\u7684\u68af\u5ea6\u4e0b\u964d\u95ee\u9898\uff09\uff0c\u6700\u540e\u4f20\u9012\u7ed9\u89e3\u7801\u5668\u7684\u4ea4\u53c9\u6ce8\u610f\u529b\u6a21\u5757\u3002<\/p>\n\n\n\n<p>3\u3001\u89e3\u7801\u5668\u5305\u542b\u81ea\u6ce8\u610f\u529b\u3001\u4ea4\u53c9\u6ce8\u610f\u529b\u3001\u524d\u9988\u795e\u7ecf\u7f51\u7edc\uff0c\u6d41\u7a0b\uff1a\u8f93\u5165\u6570\u636e\u9996\u5148\u7ecf\u8fc7\u6982\u5ff5\u4ea4\u4e92\u6a21\u5757\u8fdb\u884c\u56fe\u6587\u4ea4\u4e92\uff0c\u800c\u540e\u8fdb\u884c\u89e3\u7801\u5668\uff0c\u81ea\u6ce8\u610f\u529b\u6a21\u5757\u6355\u6349\u6587\u672c\u5e8f\u5217\u5185\u90e8\u7684\u4f9d\u8d56\u5173\u7cfb\u3001\u4ea4\u53c9\u6ce8\u610f\u529b\u6a21\u5757\u5efa\u7acb\u6587\u672c\u5e8f\u5217\u4e0e\u7f16\u7801\u5668\u8f93\u51fa\u7279\u5f81\u4e4b\u95f4\u7684\u5173\u8054\u3001\u524d\u9988\u7f51\u7edc\u5bf9\u7279\u5f81\u8fdb\u884c\u6700\u7ec8\u7684\u53d8\u6362\u4e0e\u8f93\u51fa\uff0c\u751f\u6210\u878d\u5408\u56fe\u50cf\u7279\u5f81\u7684\u6587\u672c\u5185\u5bb9\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">3.5 \u8bad\u7ec3\u4e0e\u76ee\u6807\u51fd\u6570<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">3.5.1 \u57fa\u4e8e\u4ea4\u53c9\u71b5\u635f\u5931\u7684\u8bad\u7ec3<\/h3>\n\n\n\n<figure class=\"wp-block-image size-full\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-23.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"663\" height=\"154\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-23.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-711\"  sizes=\"auto, (max-width: 663px) 100vw, 663px\" \/><\/div><\/figure>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u9075\u5faa\u56fe\u50cf captioning \u4efb\u52a1\u7684\u6807\u51c6\u505a\u6cd5 [10]-[11]\uff0c\u9996\u5148\u91c7\u7528\u8bcd\u7ea7\u4ea4\u53c9\u71b5\u635f\u5931\uff08XE\uff09\u5bf9\u6a21\u578b\u8fdb\u884c\u8bad\u7ec3\uff0c\u635f\u5931\u51fd\u6570\u5b9a\u4e49\u5982\u4e0b\uff1a\\(L_{c}(\\theta)=-\\sum_{t=1}^{T} \\log \\left(p_{\\theta}\\left(y_{t} | y_{[1, t-1]}\\right)\\right)\\)\u5176\u4e2d\uff0c\\(y_{[1, t-1]}\\)\u8868\u793a\u771f\u5b9e\u5e8f\u5217\u7684\u4e00\u90e8\u5206\uff0c\u5373\u4ece\u8d77\u59cb\u4f4d\u7f6e\u5230\u7b2c\\((t-1)\\)\u4e2a\u65f6\u95f4\u6b65\u7684\u8bcd\u6c47\u5e8f\u5217\uff1bT \u4e3a\u6700\u5927\u65f6\u95f4\u6b65\uff1b\\(\\theta\\)\u8868\u793a\u6a21\u578b\u7684\u53ef\u5b66\u4e60\u53c2\u6570\u3002<\/p>\n<\/details>\n\n\n\n<h3 class=\"wp-block-heading\">3.5.2 CIDEr\u5206\u6570\u4f18\u5316<\/h3>\n\n\n\n<figure class=\"wp-block-image size-full\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-24.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"664\" height=\"186\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-24.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-712\"  sizes=\"auto, (max-width: 664px) 100vw, 664px\" \/><\/div><\/figure>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u7136\u800c\uff0c\u8fd1\u671f\u7814\u7a76 [23] \u8868\u660e\uff0c\u4ea4\u53c9\u71b5\u8bad\u7ec3\u53ef\u80fd\u5bfc\u81f4\u66b4\u9732\u504f\u5dee\u95ee\u9898\uff0c\u5373\u6a21\u578b\u5728\u8bad\u7ec3\u9636\u6bb5\u4ee5\u771f\u5b9e\u8bcd\u6c47\u4f5c\u4e3a\u8f93\u5165\uff0c\u800c\u5728\u6d4b\u8bd5\u9636\u6bb5\u5374\u4ee5\u81ea\u8eab\u9884\u6d4b\u7ed3\u679c\u4f5c\u4e3a\u8f93\u5165\u3002\u8fd9\u79cd\u5dee\u5f02\u4f1a\u4f7f\u6a21\u578b\u7684\u8bef\u5dee\u9010\u6e10\u7d2f\u79ef\uff0c\u4ece\u800c\u964d\u4f4e\u5176\u5728\u5e8f\u5217\u9884\u6d4b\u4efb\u52a1\u4e2d\u7684\u6027\u80fd\u3002\u4e3a\u6b64\uff0c\u5c06\u81ea\u6279\u5224\u5e8f\u5217\u8bad\u7ec3\u5f3a\u5316\u5b66\u4e60\u65b9\u6cd5 [48] \u5e94\u7528\u4e8e\u56fe\u50cf captioning \u4efb\u52a1\uff0c\u8be5\u65b9\u6cd5\u5c06\u63a8\u7406\u8fc7\u7a0b\u878d\u5165\u8bad\u7ec3\u8fc7\u7a0b\uff0c\u5e76\u5f15\u5165\u5e8f\u5217\u7ea7\u8bc4\u4f30\u6307\u6807 CIDEr \u4f5c\u4e3a\u635f\u5931\u51fd\u6570\u4e2d\u7684\u5956\u52b1\u51fd\u6570\u3002\u8fd9\u79cd\u65b9\u5f0f\u6709\u52a9\u4e8e\u7f29\u5c0f\u8bad\u7ec3\u4e0e\u6d4b\u8bd5\u9636\u6bb5\u7684\u5dee\u8ddd\uff0c\u540c\u65f6\u4f7f\u6a21\u578b\u5b66\u4f1a\u57fa\u4e8e\u4e4b\u524d\u7684\u9884\u6d4b\u7ed3\u679c\u751f\u6210\u8bcd\u6c47\u3002\u5956\u52b1\u51fd\u6570\u548c\u635f\u5931\u51fd\u6570\u6784\u5efa\u5982\u4e0b\uff08\u539f\u6587\u6b64\u5904\u516c\u5f0f\u8868\u8ff0\u4e0d\u5b8c\u6574\uff0c\u672a\u660e\u786e\u5199\u51fa\u5177\u4f53\u8868\u8fbe\u5f0f\uff0c\u6545\u4fdd\u6301\u539f\u6837\uff09\uff1a\u5176\u4e2d\uff0ck \u4e3a\u6837\u672c\u6570\u91cf\uff1b\\(w^{i}\\)\u8868\u793a\u675f\u641c\u7d22\u4e2d\u91c7\u6837\u7684\u7b2c i \u4e2a\u8bed\u53e5\uff1b\\(r(\\cdot)\\)\u8868\u793a\u5956\u52b1\u51fd\u6570\uff1b\\(b=1 \/ k \\sum_{i=1}^{k} r(w^{i})\\)\u4e3a\u57fa\u51c6\u5956\u52b1\uff0c\u7528\u4e8e\u4fdd\u8bc1\u8bad\u7ec3\u7684\u7a33\u5b9a\u6027\u3002<\/p>\n<\/details>\n\n\n\n<h1 class=\"wp-block-heading\">4 \u5b9e\u9a8c\u7814\u7a76<\/h1>\n\n\n\n<h2 class=\"wp-block-heading\">4.1 \u5b9e\u9a8c\u8bbe\u7f6e\u4e0e\u6570\u636e\u96c6<\/h2>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u4e3a\u9a8c\u8bc1\u6240\u63d0 ICNET \u7684\u6027\u80fd\uff0c\u5728\u4e09\u4e2a\u516c\u5f00\u6570\u636e\u96c6\uff08Sydney-captions\u3001UCM-captions \u548c NWPU-captions\uff09\u4e0a\u8fdb\u884c\u5b9e\u9a8c\u3002\u8868 1 \u603b\u7ed3\u4e86\u4e09\u4e2a\u6570\u636e\u96c6\u7684\u8be6\u7ec6\u4fe1\u606f\u3002<\/p>\n<\/details>\n\n\n\n<figure class=\"wp-block-image size-full is-resized\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-25.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"664\" height=\"314\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-25.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-713\" style=\"width:840px;height:auto\"  sizes=\"auto, (max-width: 664px) 100vw, 664px\" \/><\/div><\/figure>\n\n\n\n<p class=\"has-text-align-center\">\u88681.\u6570\u636e\u96c6\u7684\u7edf\u8ba1\u4fe1\u606f<\/p>\n\n\n\n<p>\u4f7f\u7528\u6570\u636e\u96c6\u4e09\u4e2aCaptions\uff1aUCM-Captions\u3001Sydney-Captions\u3001NWPU-Captions<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">4.1.1 Sydney-captions \u6570\u636e\u96c6<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u8be5\u6570\u636e\u96c6\u57fa\u4e8e\u9065\u611f\u573a\u666f\u6570\u636e\u96c6 Sydney \u6570\u636e\u96c6 [67] \u6784\u5efa\uff0c\u5305\u542b 613 \u5e45\u56fe\u50cf\uff0c\u5206\u4e3a 7 \u7c7b\u573a\u666f\uff1a\u5c45\u6c11\u533a\u3001\u673a\u573a\u3001\u8349\u5730\u3001\u6cb3\u6d41\u3001\u6d77\u6d0b\u3001\u5de5\u4e1a\u533a\u548c\u8dd1\u9053\u3002\u6bcf\u5e45\u56fe\u50cf\u5c3a\u5bf8\u4e3a 500\u00d7500 \u50cf\u7d20\uff0c\u50cf\u7d20\u5206\u8fa8\u7387\u7ea6\u4e3a 50 \u5398\u7c73\u3002\u6240\u6709\u56fe\u50cf\u5747\u4ece\u8c37\u6b4c\u5730\u7403\u4e0b\u8f7d\u7684\u6089\u5c3c\u5730\u533a\u539f\u59cb\u56fe\u50cf\uff08\u5c3a\u5bf8\u4e3a 18000\u00d714000 \u50cf\u7d20\uff09\u4e2d\u4eba\u5de5\u63d0\u53d6\u5f97\u5230\u3002<\/p>\n<\/details>\n\n\n\n<h3 class=\"wp-block-heading\">4.1.2 UCM-captions \u6570\u636e\u96c6<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u8be5\u6570\u636e\u96c6\u57fa\u4e8e\u7f8e\u56fd\u52a0\u5dde\u5927\u5b66\u9ed8\u585e\u5fb7\u5206\u6821\u7684\u571f\u5730\u5229\u7528\u9065\u611f\u573a\u666f\u5206\u7c7b\u6570\u636e\u96c6 [68] \u6784\u5efa\uff0c\u5305\u542b 2100 \u5e45\u56fe\u50cf\uff0c\u6db5\u76d6 21 \u7c7b\u5178\u578b\u571f\u5730\u5229\u7528\u573a\u666f\uff0c\u5305\u62ec\u519c\u4e1a\u7528\u5730\u3001\u98de\u673a\u3001\u68d2\u7403\u573a\u3001\u6d77\u6ee9\u3001\u5efa\u7b51\u7269\u3001\u704c\u6728\u4e1b\u3001\u9ad8\u5bc6\u5ea6\u4f4f\u5b85\u533a\u3001\u68ee\u6797\u3001\u9ad8\u901f\u516c\u8def\u3001\u9ad8\u5c14\u592b\u7403\u573a\u3001\u6e2f\u53e3\u3001\u5341\u5b57\u8def\u53e3\u3001\u4e2d\u5bc6\u5ea6\u4f4f\u5b85\u533a\u3001\u79fb\u52a8\u623f\u5c4b\u516c\u56ed\u3001\u7acb\u4ea4\u6865\u3001\u505c\u8f66\u573a\u3001\u6cb3\u6d41\u3001\u8dd1\u9053\u3001\u4f4e\u5bc6\u5ea6\u4f4f\u5b85\u533a\u3001\u50a8\u6cb9\u7f50\u548c\u7f51\u7403\u573a\u3002\u6bcf\u4e2a\u7c7b\u522b\u5305\u542b 100 \u5e45\u56fe\u50cf\uff0c\u5c3a\u5bf8\u5747\u4e3a 256\u00d7256 \u50cf\u7d20\uff0cRGB \u7a7a\u95f4\u50cf\u7d20\u5206\u8fa8\u7387\u4e3a 30 \u5398\u7c73\u3002UCM \u6570\u636e\u96c6\u4ece\u7f8e\u56fd\u5730\u8d28\u8c03\u67e5\u5c40\uff08USGS\uff09\u4e0b\u8f7d\u7684\u822a\u7a7a\u6b63\u5c04\u5f71\u50cf\u4e2d\u63d0\u53d6\u5f97\u5230\u3002<\/p>\n<\/details>\n\n\n\n<h3 class=\"wp-block-heading\">4.1.3 NWPU-captions \u6570\u636e\u96c6<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u8be5\u6570\u636e\u96c6\u57fa\u4e8e\u897f\u5317\u5de5\u4e1a\u5927\u5b66\u521b\u5efa\u7684\u9065\u611f\u56fe\u50cf\u5206\u7c7b\u6570\u636e\u96c6\u6784\u5efa\uff0c\u5305\u542b 31500 \u5e45\u822a\u7a7a\u9065\u611f RGB \u4e09\u901a\u9053\u56fe\u50cf\u548c 157500 \u6761\u56fe\u50cf\u63cf\u8ff0\u8bed\u53e5\u3002\u6bcf\u5e45\u56fe\u50cf\u5c3a\u5bf8\u4e3a 256\u00d7256 \u50cf\u7d20\uff0c\u5730\u9762\u91c7\u6837\u8ddd\u79bb\u8303\u56f4\u4e3a 0.2-30 \u7c73\u3002\u6b64\u5916\uff0c\u4e3a\u63d0\u5347\u8bed\u53e5\u7684\u8bed\u4e49\u4e30\u5bcc\u5ea6\uff0c\u6bcf\u5e45\u56fe\u50cf\u90fd\u914d\u6709 5 \u6761\u4e0d\u540c\u7684\u63cf\u8ff0\u6587\u672c\u3002<\/p>\n<\/details>\n\n\n\n<figure class=\"wp-block-image size-full is-resized\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-26.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"644\" height=\"209\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-26.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-714\" style=\"width:840px;height:auto\"  sizes=\"auto, (max-width: 644px) 100vw, 644px\" \/><\/div><\/figure>\n\n\n\n<p class=\"has-text-align-center\">\u8868 2 \u8bad\u7ec3\u96c6 &#8211; \u9a8c\u8bc1\u96c6 &#8211; \u6d4b\u8bd5\u96c6\u7684\u5212\u5206\u7b56\u7565\u3002<\/p>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u5728\u8fd9\u4e09\u4e2a\u6570\u636e\u96c6\u7684\u6bcf\u5e45\u56fe\u50cf\u90fd\u7531\u4e0d\u540c\u89c2\u6d4b\u8005\u63d0\u4f9b 5 \u6761\u63cf\u8ff0\u8bed\u53e5\u3002\u6570\u636e\u96c6\u7684\u8bad\u7ec3\u96c6\u3001\u9a8c\u8bc1\u96c6\u548c\u6d4b\u8bd5\u96c6\u5212\u5206\u65b9\u5f0f\u4e0e\u539f\u59cb\u6587\u732e\u4fdd\u6301\u4e00\u81f4\uff0c\u5747\u6309\u7167 80%\/10%\/10% \u7684\u6bd4\u4f8b\u5212\u5206\uff0c\u5177\u4f53\u6837\u672c\u6570\u91cf\u5982\u8868 2 \u6240\u793a\u3002\u9700\u8981\u6ce8\u610f\u7684\u662f\uff0c\u90e8\u5206\u5bf9\u6bd4\u65b9\u6cd5\u672a\u91c7\u7528\u9ed8\u8ba4\u7684\u6570\u636e\u96c6\u5212\u5206\u65b9\u5f0f\uff0c\u800c\u662f\u91c7\u7528\u968f\u673a\u5212\u5206\u3002\u56e0\u6b64\uff0c\u672c\u6587\u7edf\u8ba1\u4e86\u5bf9\u6bd4\u65b9\u6cd5\u7684\u5212\u5206\u65b9\u5f0f\uff1a\u6587\u732e [18]\u3001[28]\u3001[23]\u3001[74]\u3001[75] \u91c7\u7528\u968f\u673a\u5212\u5206\uff1b\u6587\u732e [19]\u3001[27]\u3001[62]\u3001[69] \u91c7\u7528\u6807\u51c6\u5212\u5206\uff1b\u6587\u732e [22]\u3001[49]\u3001[76]-[80] \u672a\u660e\u786e\u8bf4\u660e\u5212\u5206\u65b9\u5f0f\u3002<\/p>\n\n\n\n<p>\u4e3a\u4fdd\u8bc1\u4e0e\u5bf9\u6bd4\u65b9\u6cd5\u7684\u516c\u5e73\u6027\uff0c\u53d7 [23] \u548c [27] \u542f\u53d1\uff0c\u5bf9\u6bcf\u4e2a\u6570\u636e\u96c6\u6309\u7167\u76f8\u540c\u6bd4\u4f8b\u8fdb\u884c 5 \u6b21\u5b9e\u9a8c\uff0c\u6bcf\u6b21\u5b9e\u9a8c\u91c7\u7528\u4e0d\u540c\u7684\u968f\u673a\u5212\u5206\u3002\u4e3a\u83b7\u5f97\u66f4\u53ef\u9760\u7684\u7ed3\u679c\uff0c\u5254\u9664\u6bcf\u6b21\u5b9e\u9a8c\u4e2d\u7684\u6700\u4f18\u548c\u6700\u5dee\u7ed3\u679c\uff0c\u5bf9\u5269\u4f59\u7ed3\u679c\u53d6\u5e73\u5747\u503c\u5e76\u8ba1\u7b97\u6807\u51c6\u5dee\u3002<\/p>\n\n\n\n<p>\u6240\u63d0 ICNET \u9075\u5faa\u9065\u611f\u56fe\u50cf captioning \u4efb\u52a1\u7684\u6807\u51c6\u5b9e\u8df5\uff0c\u91c7\u7528\u5728 ImageNet \u4e0a\u9884\u8bad\u7ec3\u7684 ResNet-152 \u63d0\u53d6\u533a\u57df\u7279\u5f81\uff0c\u91c7\u7528\u4ee5 ViT-b\/32 \u4e3a\u9aa8\u5e72\u7f51\u7edc\u7684\u9884\u8bad\u7ec3 CLIP \u56fe\u50cf\u7f16\u7801\u5668\u63d0\u53d6\u5168\u5c40\u7279\u5f81\u3002\u501f\u52a9 ViT\uff0cCLIP \u56fe\u50cf\u5d4c\u5165\u80fd\u591f\u5bf9\u6240\u6709\u56fe\u50cf\u5757\u7684\u89c6\u89c9\u6807\u8bb0\u8fdb\u884c\u7f16\u7801\uff1b\u6b64\u5916\uff0c\u89c6\u89c9\u4e0e\u6587\u672c\u7684\u878d\u5408\u8fdb\u4e00\u6b65\u4e30\u5bcc\u4e86\u4e24\u79cd\u6a21\u6001\u7684\u4fe1\u606f\u5d4c\u5165\u3002<\/p>\n\n\n\n<p>\u968f\u540e\uff0c\u91c7\u7528\u5305\u542b\\(N=3\\)\u4e2a\u76f8\u540c\u5c42\u7684\u6807\u51c6 Transformer \u6846\u67b6\uff0c\u5728\u591a\u5934\u6ce8\u610f\u529b\u673a\u5236\u4e2d\u4f7f\u7528 8 \u4e2a\u5e76\u884c\u6ce8\u610f\u529b\u5c42\u3002\u5bf9\u4e8e\u6587\u672c\u8868\u5f81\uff0c\u5728\u6bcf\u4e2a\u771f\u5b9e\u5e8f\u5217\u7684\u524d\u540e\u5206\u522b\u6dfb\u52a0\u8d77\u59cb\u6807\u8bb0\u27e8bos\u27e9\u548c\u7ed3\u675f\u6807\u8bb0\u27e8eos\u27e9\u3002\u8bad\u7ec3\u9636\u6bb5\uff0c\u89e3\u7801\u5668\u7684\u8f93\u5165\u4e3a\u771f\u5b9e\u5e8f\u5217\uff1b\u800c\u5728\u63a8\u7406\u9636\u6bb5\uff0c\u89e3\u7801\u5668\u53ea\u9700\u8f93\u5165\u8d77\u59cb\u6807\u8bb0\u27e8bos\u27e9\u3002\u8bcd\u5d4c\u5165\u7ef4\u5ea6\u8bbe\u7f6e\u4e3a 512\uff0c\u4f4d\u7f6e\u7f16\u7801\u7684\u6700\u5927\u5e8f\u5217\u957f\u5ea6\u4e3a 128\uff0c\u8f93\u51fa\u5e8f\u5217\u7684\u6700\u5927\u957f\u5ea6\u4e3a 25\u3002<\/p>\n\n\n\n<p>\u5b9e\u9a8c\u4e2d\uff0c\u6279\u5904\u7406\u5927\u5c0f\u8bbe\u7f6e\u4e3a 50\uff0c\u91c7\u7528 Adam \u4f18\u5316\u5668\u4f18\u5316\u6a21\u578b\u53c2\u6570\uff1b\u4e3a\u907f\u514d\u8fc7\u62df\u5408\uff0c\u5728\u6bcf\u4e2a\u6ce8\u610f\u529b\u5c42\u548c FFN \u5c42\u540e\u91c7\u7528\u6982\u7387\u4e3a 0.1 \u7684\u4e22\u5f03\u7b56\u7565\uff08dropout\uff09\uff1b\u6bcf\u4e2a FFN \u5c42\u7684\u9690\u85cf\u7ef4\u5ea6\u8bbe\u7f6e\u4e3a 2048\u3002\u6b64\u5916\uff0c\u901a\u8fc7\u8bbe\u7f6e \u201c\u8010\u5fc3\u503c\uff08patience\uff09\u201d \u76d1\u63a7\u6a21\u578b\u5728\u9a8c\u8bc1\u96c6\u4e0a\u7684\u6027\u80fd\uff0c\u4ee5\u51b3\u5b9a\u662f\u5426\u9700\u8981\u5bf9\u6a21\u578b\u8fdb\u884c\u6539\u8fdb\u3002\u6240\u6709\u5b9e\u9a8c\u5728\u914d\u5907 NVIDIA Tesla V100-32GB \u663e\u5361\u7684\u8bbe\u5907\u4e0a\u8fdb\u884c\uff0c\u57fa\u4e8e Pytorch 1.12.1 \u6846\u67b6\u5b9e\u73b0\u3002<\/p>\n\n\n\n<p>\u5728\u4ea4\u53c9\u71b5\u8bad\u7ec3\u8fc7\u7a0b\u4e2d\uff0c\u91c7\u7528\u5b66\u4e60\u7387\u8c03\u5ea6\u7b56\u7565 [29]\uff0c\u8be5\u7b56\u7565\u5305\u542b 10000 \u6b21\u8fed\u4ee3\u7684\u9884\u70ed\u64cd\u4f5c\u3002\u5c06\u5728\u9a8c\u8bc1\u96c6\u4e0a\u53d6\u5f97\u6700\u9ad8 CIDEr \u5206\u6570\u7684\u6a21\u578b\u4fdd\u5b58\u4e3a\u540e\u7eed\u8bad\u7ec3\u9636\u6bb5\u6216\u63a8\u7406\u8fc7\u7a0b\u7684\u521d\u59cb\u5316\u6a21\u578b\u3002\u5f53\u8010\u5fc3\u503c\u8fbe\u5230 5 \u65f6\uff0c\u8bad\u7ec3\u91cd\u70b9\u8f6c\u5411 CIDEr \u4f18\u5316\uff0c\u5e76\u5c06\u5b66\u4e60\u7387\u56fa\u5b9a\u4e3a5\u00d710\u7684-6\u6b21\u65b9\u5728\u4f18\u5316\u548c\u89e3\u7801\u9636\u6bb5\uff0c\u5747\u91c7\u7528\u675f\u5927\u5c0f\u4e3a 5 \u7684\u675f\u641c\u7d22\u751f\u6210\u5019\u9009\u5e8f\u5217\u3002<\/p>\n<\/details>\n\n\n\n<h2 class=\"wp-block-heading\">4.2 \u8bc4\u4f30\u6307\u6807<\/h2>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u4e3a\u4ece\u591a\u4e2a\u89d2\u5ea6\u8bc4\u4f30\u6a21\u578b\u6027\u80fd\uff0c\u9009\u53d6\u56fe\u50cf captioning \u4efb\u52a1\u4e2d\u5e38\u7528\u7684 6 \u79cd\u8bc4\u4f30\u6307\u6807\uff1aBLEU [70]\u3001METEOR [71]\u3001ROUGE-L [72]\u3001CIDEr [73]\u3001SPICE \u548cSm<\/p>\n<\/details>\n\n\n\n<p>1\u3001\u9009\u53d6\u56fe\u50cf captioning \u4efb\u52a1\u4e2d\u5e38\u7528\u7684 6 \u79cd\u8bc4\u4f30\u6307\u6807\uff1aBLEU\u3001METEOR\u3001ROUGE-L\u3001CIDEr\u3001SPICE \u548cSm<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">4.2.1 BLEU<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>BLEU \u662f\u8861\u91cf\u751f\u6210\u6587\u672c\u4e0e\u53c2\u8003\u6587\u672c\u76f8\u4f3c\u5ea6\u7684\u5e38\u7528\u81ea\u52a8\u8bc4\u4f30\u6307\u6807\uff0c\u66f4\u4fa7\u91cd\u4e8e\u7cbe\u786e\u7387\u3002\u5b83\u901a\u8fc7\u8ba1\u7b97\u751f\u6210\u6587\u672c\u4e2d\u51fa\u73b0\u5728\u53c2\u8003\u6587\u672c\u4e2d\u7684 n \u5143\u8bed\u6cd5\uff08\u4ece 1 \u5143\u5230 4 \u5143\uff09\u7684\u6bd4\u4f8b\uff0c\u7ed3\u5408\u7cbe\u786e\u5339\u914d\u548c\u90e8\u5206\u5339\u914d\u6765\u8ba1\u7b97\u5206\u6570\u3002\u5176\u4e2d\uff0cn \u5143\u8bed\u6cd5\u6307\u7531 n \u4e2a\u8fde\u7eed\u8bcd\u6c47\u7ec4\u6210\u7684\u5e8f\u5217\uff0cn \u7684\u53d6\u503c\u8303\u56f4\u4e3a 1-4\u3002<\/p>\n<\/details>\n\n\n\n<h3 class=\"wp-block-heading\">4.2.2 METEOR<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>METEOR \u7ed3\u5408\u4e86\u7cbe\u786e\u5339\u914d\u548c\u975e\u7cbe\u786e\u5339\u914d\u51c6\u5219\uff08\u5982\u540c\u4e49\u8bcd\u548c\u8bcd\u5e72\u5339\u914d\uff09\uff0c\u901a\u8fc7\u8c03\u7528\u5916\u90e8\u8bcd\u5178\u8bc6\u522b\u540c\u4e49\u8bcd\u548c\u8bcd\u5e72\uff0c\u80fd\u591f\u66f4\u5168\u9762\u5730\u8bc4\u4f30\u751f\u6210\u6587\u672c\u7684\u8d28\u91cf\u3002<\/p>\n<\/details>\n\n\n\n<h3 class=\"wp-block-heading\">4.2.3 ROUGE<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>ROUGE \u662f\u8861\u91cf\u751f\u6210\u6587\u672c\u4e0e\u53c2\u8003\u6587\u672c\u53ec\u56de\u7387\u7684\u8bc4\u4f30\u6307\u6807\uff0c\u901a\u8fc7\u6bd4\u8f83\u6700\u957f\u516c\u5171\u5b50\u5e8f\u5217\u7684\u957f\u5ea6\u6765\u8bc4\u4f30\u751f\u6210\u6587\u672c\u4e0e\u53c2\u8003\u6587\u672c\u7684\u76f8\u4f3c\u5ea6\u3002<\/p>\n<\/details>\n\n\n\n<h3 class=\"wp-block-heading\">4.2.4 CIDEr<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>CIDEr \u662f\u4e13\u95e8\u7528\u4e8e\u56fe\u50cf captioning \u4efb\u52a1\u7684\u8bc4\u4f30\u6307\u6807\uff0c\u5173\u6ce8\u751f\u6210\u63cf\u8ff0\u7684\u591a\u6837\u6027\u548c\u4e00\u81f4\u6027\u3002\u5b83\u901a\u8fc7\u6bd4\u8f83\u751f\u6210\u6587\u672c\u4e0e\u591a\u4e2a\u53c2\u8003\u6587\u672c\u4e4b\u95f4\u7684\u8bcd\u9891\u7edf\u8ba1\u4fe1\u606f\u6765\u8ba1\u7b97\u5206\u6570\uff0c\u5f3a\u8c03\u5728\u591a\u4e2a\u53c2\u8003\u6587\u672c\u95f4\u7684\u4e00\u81f4\u6027\u3002<\/p>\n<\/details>\n\n\n\n<h3 class=\"wp-block-heading\">4.2.5 SPICE<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>SPICE \u662f\u9488\u5bf9\u56fe\u50cf\u63cf\u8ff0\u8bc4\u4f30\u63d0\u51fa\u7684\u6307\u6807\u3002\u4e0e\u4f20\u7edf\u57fa\u4e8e n \u5143\u8bed\u6cd5\u3001\u4e3b\u8981\u901a\u8fc7\u8ba1\u7b97\u53c2\u8003\u6587\u672c\u4e0e\u751f\u6210\u6587\u672c\u95f4\u8bcd\u6c47\u91cd\u53e0\u5ea6\u7684\u8bc4\u4f30\u65b9\u6cd5\u4e0d\u540c\uff0cSPICE \u5229\u7528\u8bed\u4e49\u89e3\u6790\u6811\u8861\u91cf\u751f\u6210\u6587\u672c\u4e0e\u53c2\u8003\u6587\u672c\u7684\u76f8\u4f3c\u5ea6\uff0c\u80fd\u591f\u5b9e\u73b0\u66f4\u4f18\u7684\u8bed\u4e49\u5339\u914d\u3002<\/p>\n<\/details>\n\n\n\n<h3 class=\"wp-block-heading\">4.2.6 Sm<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>Sm\u4e3a BLEU-4\u3001METEOR\u3001ROUGE \u548c CIDEr \u56db\u79cd\u6307\u6807\u7684\u7b97\u672f\u5e73\u5747\u503c\uff0c\u53ef\u5168\u9762\u8bc4\u4f30\u751f\u6210\u63cf\u8ff0\u7684\u8d28\u91cf\u3002<\/p>\n<\/details>\n\n\n\n<h2 class=\"wp-block-heading\">4.3 \u5bf9\u6bd4\u65b9\u6cd5<\/h2>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u4e3a\u8bc4\u4f30\u6240\u63d0 ICNET \u7684\u6709\u6548\u6027\uff0c\u5c06\u5176\u4e0e\u4ee5\u4e0b\u51e0\u79cd\u5f53\u524d\u4e3b\u6d41\u65b9\u6cd5\u8fdb\u884c\u5bf9\u6bd4\uff1a<\/p>\n\n\n\n<p>MLAT [28]\uff1a\u5229\u7528 LSTM \u878d\u5408 ResNet-50 \u4e0d\u540c\u5c42\u7684\u7279\u5f81\u4ee5\u63d0\u53d6\u591a\u5c3a\u5ea6\u4fe1\u606f\uff0c\u968f\u540e\u901a\u8fc7\u591a\u5c42\u805a\u5408 Transformer \u751f\u6210\u63cf\u8ff0\u6587\u672c\u3002<\/p>\n\n\n\n<p>Soft-attention \u548c Hard-attention [74]\uff1a\u5206\u522b\u5728 CNN-RNN \u6846\u67b6\u4e2d\u5f15\u5165\u786c\u6ce8\u610f\u529b\u548c\u8f6f\u6ce8\u610f\u529b\u673a\u5236\u3002<\/p>\n\n\n\n<p>FC-Att + LSTM \u548c SM-Att + LSTM [18]\uff1a\u57fa\u4e8e\u5c5e\u6027\u6ce8\u610f\u529b\u673a\u5236\uff0c\u878d\u5408\u4f4e\u5c42\u6b21\u7279\u5f81\u548c\u9ad8\u5c42\u6b21\u5c5e\u6027\u7279\u5f81\uff0c\u7ed3\u679c\u6765\u6e90\u4e8e [76]\u3002<\/p>\n\n\n\n<p>Trans + SCST [75]\uff1a\u63d0\u51fa\u4e00\u79cd\u65b0\u7684 Transformer \u7ed3\u6784\uff0c\u5728 Transformer \u4e2d\u6dfb\u52a0\u4e22\u5f03\u5c42\u3001\u6b8b\u5dee\u8fde\u63a5\u548c\u81ea\u9002\u5e94\u7279\u5f81\u878d\u5408\uff0c\u5e76\u91c7\u7528\u5f3a\u5316\u5b66\u4e60\u63d0\u5347\u6a21\u578b\u7684\u5b9e\u9645\u6027\u80fd\u3002<\/p>\n\n\n\n<p>Word-Sentence [62]\uff1a\u5178\u578b\u7684\u7f16\u7801\u5668 &#8211; \u89e3\u7801\u5668\u6846\u67b6\uff0c\u7531\u8bcd\u63d0\u53d6\u5668\u548c\u8bed\u53e5\u751f\u6210\u5668\u7ec4\u6210\uff0c\u65e8\u5728\u4ece\u9065\u611f\u56fe\u50cf\u4e2d\u63d0\u53d6\u8bcd\u6c47\u5e76\u7ec4\u6210\u8fde\u8d2f\u8bed\u53e5\u3002<\/p>\n\n\n\n<p>MLCA-Net [69]\uff1a\u901a\u8fc7\u591a\u7ea7\u4e0a\u4e0b\u6587\u6ce8\u610f\u529b\u6a21\u5757\u805a\u5408\u4e0d\u540c\u5c3a\u5ea6\u7684\u7279\u5f81\u548c\u6f5c\u5728\u4e0a\u4e0b\u6587\u4fe1\u606f\uff0c\u5e76\u901a\u8fc7 LSTM \u751f\u6210\u63cf\u8ff0\u6587\u672c\u3002<\/p>\n\n\n\n<p>MC-Net [76]\uff1a\u5305\u542b\u4e00\u4e2a\u589e\u5f3a\u578b\u7f16\u7801\u5668\uff08\u914d\u5907\u591a\u5c3a\u5ea6\u63d0\u53d6\u6a21\u5757\uff09\u3001\u4e00\u4e2a\u7279\u5f81\u878d\u5408\u6a21\u5757\u4ee5\u53ca\u4e00\u4e2a\u5fae\u8c03\u540e\u7684 LSTM\uff08\u914d\u5907\u89c6\u89c9 &#8211; \u6587\u672c\u5bf9\u9f50\u6a21\u5757\uff09\u3002<\/p>\n\n\n\n<p>GLCM [27]\uff1a\u57fa\u4e8e\u6ce8\u610f\u529b\u7684\u5168\u5c40 &#8211; \u5c40\u90e8\u63cf\u8ff0\u751f\u6210\u6a21\u578b\uff0c\u901a\u8fc7 CNN \u63d0\u53d6\u5168\u5c40\u548c\u5c40\u90e8\u7279\u5f81\u7528\u4e8e\u8bed\u53e5\u751f\u6210\u3002<\/p>\n\n\n\n<p>PKG-Transformer [23]\uff1a\u9996\u5148\u5229\u7528\u76ee\u6807 &#8211; \u76ee\u6807\u548c\u573a\u666f &#8211; \u573a\u666f\u5173\u7cfb\u4e30\u5bcc\u76ee\u6807\u548c\u573a\u666f\u7279\u5f81\uff0c\u7136\u540e\u5c06\u573a\u666f &#8211; \u76ee\u6807\u5173\u7cfb\u4f5c\u4e3a\u5148\u9a8c\u77e5\u8bc6\u878d\u5165 Transformer \u7f16\u7801\u5668\u3002<\/p>\n\n\n\n<p>FPHD [77]\uff1a\u57fa\u4e8e\u5206\u5c42\u56fe\u50cf\u4fe1\u606f\u533a\u5206\u524d\u666f\u548c\u80cc\u666f\uff0c\u8bbe\u8ba1\u53ef\u53d8\u5f62 Transformer\uff0c\u5e76\u901a\u8fc7\u4ea4\u4e92\u4ece\u524d\u666f\u548c\u80cc\u666f\u4e2d\u5b66\u4e60\u591a\u5c3a\u5ea6\u7279\u5f81\u3002<\/p>\n\n\n\n<p>BITA [78]\uff1a\u5229\u7528\u8f7b\u91cf\u7ea7\u4ea4\u4e92\u5f0f\u5085\u91cc\u53f6\u53d8\u6362\u5b9e\u73b0\u9065\u611f\u56fe\u50cf &#8211; \u6587\u672c\u7279\u5f81\u5bf9\u9f50\uff0c\u5e76\u91c7\u7528\u4e24\u7ea7\u89c6\u89c9 &#8211; \u8bed\u8a00\u9884\u8bad\u7ec3\u65b9\u6cd5\u6307\u5bfc\u4ea4\u4e92\u5f0f\u56fe\u50cf &#8211; \u6587\u672c\u5bf9\u9f50\u3002<\/p>\n\n\n\n<p>CASK [79]\uff1a\u901a\u8fc7\u8bed\u4e49\u6982\u5ff5\u63d0\u53d6\u5668\u6355\u6349\u8bed\u4e49\u6982\u5ff5\uff0c\u5c06\u8054\u5408\u89c6\u89c9 &#8211; \u8bed\u4e49\u534f\u540c\u6ce8\u610f\u529b\u878d\u5165\u5171\u8bc6\u6784\u5efa\u6a21\u5757\uff0c\u4ee5\u5b66\u4e60\u8de8\u6a21\u6001\u5171\u8bc6\u611f\u77e5\u77e5\u8bc6\u3002<\/p>\n\n\n\n<p>PureT [49]\uff1a\u91c7\u7528 Swin Transformer \u63d0\u53d6\u56fe\u50cf\u7279\u5f81\uff0c\u901a\u8fc7\u4f18\u5316\u7f16\u7801\u5668\u6355\u6349\u7279\u5f81\u95f4\u7684\u5173\u7cfb\uff0c\u89e3\u7801\u5668\u5c06\u4f18\u5316\u540e\u7684\u7279\u5f81\u9010\u5b57\u89e3\u7801\u4e3a\u63cf\u8ff0\u6587\u672c\uff0c\u7ed3\u679c\u6765\u6e90\u4e8e [19]\u3002<\/p>\n\n\n\n<p>RS-CapRet [22]\uff1a\u91c7\u7528 CLIP \u5bf9\u63cf\u8ff0\u6587\u672c\u8fdb\u884c\u7f16\u7801\uff0c\u901a\u8fc7\u7b80\u5355\u7684\u878d\u5408\u7f51\u7edc\u751f\u6210\u63cf\u8ff0\u6587\u672c\u5e76\u5b9e\u73b0\u6587\u672c &#8211; \u56fe\u50cf\u68c0\u7d22\uff0c\u968f\u540e\u5bf9 LLamaV2-7B \u6a21\u578b\u8fdb\u884c\u5fae\u8c03\u3002<\/p>\n\n\n\n<p>BLIP-2 [80]\uff1a\u4e00\u79cd\u89c6\u89c9 &#8211; \u8bed\u8a00\u9884\u8bad\u7ec3\u65b9\u6cd5\uff0c\u501f\u9274 BERT \u57fa\u7840\u6a21\u578b\u7684\u8bbe\u8ba1\u601d\u8def\uff0c\u91c7\u7528 Q-Former \u7ed3\u6784\u5b9e\u73b0\u591a\u6a21\u6001\u878d\u5408\u3002<\/p>\n\n\n\n<p>HCNet [19]\uff1a\u91c7\u7528 CNN \u8fdb\u884c\u5206\u5c42\u89c6\u89c9\u7279\u5f81\u63d0\u53d6\uff0c\u901a\u8fc7 LSTM \u5b9e\u73b0\u8bed\u8a00\u751f\u6210\u3002<\/p>\n<\/details>\n\n\n\n<p>1\u3001\u5bf9\u6bd4\u6a21\u578b\uff1aMLAT\u3001Soft-attention \u548c Hard-attention\u3001FC-Att + LSTM \u548c SM-Att + LSTM\u3001Trans + SCST\u3001Word-Sentence\u3001MLCA-Net\u3001MC-Net\u3001GLCM\u3001PKG-Transformer\u3001FPHD\u3001BITA\u3001CASK\u3001PureT\u3001RS-CapRet\u3001BLIP-2<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">4.4 \u4e0e\u73b0\u6709\u65b9\u6cd5\u7684\u5bf9\u6bd4<\/h2>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u672c\u8282\u5c06\u6240\u63d0 ICNET \u4e0e\u4e0a\u8ff0\u591a\u79cd\u4e3b\u6d41\u65b9\u6cd5\uff08\u5305\u62ec\u57fa\u4e8e\u6ce8\u610f\u529b\u7684\u65b9\u6cd5\u548c\u57fa\u4e8e Transformer \u7684\u65b9\u6cd5\uff09\u8fdb\u884c\u9009\u62e9\u6027\u5bf9\u6bd4\uff0c\u7ed3\u679c\u5982\u8868 3 &#8211; \u8868 5 \u6240\u793a\u3002\u5b9e\u9a8c\u7ed3\u679c\u5982\u56fe 4 \u6240\u793a\uff0c\u57fa\u4e8e Transformer \u7684\u65b9\u6cd5\u5728\u6027\u80fd\u4e0a\u4e00\u5b9a\u7a0b\u5ea6\u4f18\u4e8e\u5176\u4ed6\u65b9\u6cd5\uff0c\u800c\u6240\u63d0\u6a21\u578b\u5728\u5927\u591a\u6570\u6307\u6807\u4e0a\u5747\u53d6\u5f97\u4e86\u6700\u4f18\u7684\u5b9e\u9a8c\u7ed3\u679c\u3002<\/p>\n\n\n\n<p>\u8fd9\u4e00\u4f18\u52bf\u7684\u539f\u56e0\u5728\u4e8e\uff1a\u6240\u63d0\u6a21\u578b\u4e0d\u4ec5\u5229\u7528\u8bed\u4e49\u6982\u5ff5\u5e76\u4fdd\u7559\u5bf9\u540e\u7eed\u6587\u672c\u751f\u6210\u6709\u5e2e\u52a9\u7684\u5168\u5c40\u4fe1\u606f\uff0c\u8fd8\u4fdd\u7559\u4e86\u56fe\u50cf\u4e2d\u663e\u8457\u533a\u57df\u6216\u76ee\u6807\u7684\u7279\u5f81\uff0c\u901a\u8fc7\u5bf9\u5c40\u90e8\u7279\u5f81\u7684\u6620\u5c04\u8fdb\u4e00\u6b65\u63d0\u5347\u4e86\u8fd9\u4e9b\u7279\u5f81\u7684\u8868\u5f81\u80fd\u529b\uff0c\u4ece\u800c\u589e\u5f3a\u4e86\u6a21\u578b\u5bf9\u76ee\u6807\u90bb\u57df\u7279\u5f81\u7684\u611f\u77e5\u80fd\u529b\u3002\u5728\u6587\u672c\u751f\u6210\u8fc7\u7a0b\u4e2d\uff0c\u89e3\u7801\u5668\u76f4\u63a5\u4e0e\u6240\u6709\u7c7b\u578b\u7684\u6982\u5ff5\u76f8\u8fde\uff0c\u800c\u4f20\u7edf Transformer \u7684\u89e3\u7801\u5668\u4ec5\u5173\u6ce8\u7f16\u7801\u5668\u6700\u540e\u4e00\u5c42\u7684\u8f93\u51fa\u3002\u6240\u63d0\u65b9\u6cd5\u7684\u8fd9\u4e00\u8bbe\u8ba1\u4f7f\u5f97\u4e0d\u540c\u7c7b\u578b\u7684\u7279\u5f81\u90fd\u80fd\u53c2\u4e0e\u6587\u672c\u751f\u6210\uff0c\u540c\u65f6\u63d0\u4f9b\u5168\u5c40\u4fe1\u606f\u548c\u5c40\u90e8\u663e\u8457\u533a\u57df\u4fe1\u606f\uff0c\u5145\u5206\u4f53\u73b0\u4e86 ICNET \u7684\u4f18\u8d8a\u6027\u3002<\/p>\n\n\n\n<p><\/p>\n<\/details>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-27-1024x367.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"367\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-27-1024x367.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-717\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<p>\u88683.\u90e8\u5206\u5148\u8fdb\u65b9\u6cd5\u5728 SYDNEY-CAPTIONS \u6570\u636e\u96c6\u4e0a\u7684\u5bf9\u6bd4\u3002\u6240\u6709\u7ed3\u679c\u5747\u4ee5\u767e\u5206\u6bd4\uff08%\uff09\u5f62\u5f0f\u5448\u73b0\u3002\u7b26\u53f7 \u201c-\u201d \u8868\u793a\u8be5\u6587\u732e\u672a\u62a5\u544a\u6b64\u7ed3\u679c\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-28-1024x405.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"405\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-28-1024x405.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-718\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<p>\u88684.\u90e8\u5206\u5148\u8fdb\u65b9\u6cd5\u5728 UCM-CAPTIONS \u6570\u636e\u96c6\u4e0a\u7684\u5bf9\u6bd4\u3002\u6240\u6709\u7ed3\u679c\u5747\u4ee5\u767e\u5206\u6bd4\uff08%\uff09\u5f62\u5f0f\u5448\u73b0\u3002\u7b26\u53f7 \u201c-\u201d \u8868\u793a\u8be5\u6587\u732e\u672a\u62a5\u544a\u6b64\u7ed3\u679c\u3002<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-29-1024x337.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"337\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-29-1024x337.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-719\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<p>\u88685.\u90e8\u5206\u5148\u8fdb\u65b9\u6cd5\u5728 NWPU-CAPTIONS \u6570\u636e\u96c6\u4e0a\u7684\u5bf9\u6bd4\u3002\u6240\u6709\u7ed3\u679c\u5747\u4ee5\u767e\u5206\u6bd4\uff08%\uff09\u5f62\u5f0f\u5448\u73b0\u3002\u7b26\u53f7 \u201c-\u201d \u8868\u793a\u8be5\u6587\u732e\u672a\u62a5\u544a\u6b64\u7ed3\u679c\u3002<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">4.5 \u6d88\u878d\u5b9e\u9a8c<\/h2>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u901a\u8fc7\u6d88\u878d\u5b9e\u9a8c\u8bc4\u4f30 ICNET \u4e2d\u5404\u6a21\u5757\u5bf9\u6574\u4f53\u6027\u80fd\u7684\u8d21\u732e\u3002\u5c06\u4ec5\u5305\u542b ViT \u548c Transformer \u7684\u6a21\u578b\u4f5c\u4e3a\u57fa\u51c6\u6a21\u578b\uff08Baseline\uff09\uff0c\u901a\u8fc7\u9010\u4e00\u6dfb\u52a0\u4e09\u4e2a\u6a21\u5757\u6784\u5efa\u6d88\u878d\u6a21\u578b\uff1a\u6a21\u578b 1\uff08Model-1\uff09\u4e3a\u57fa\u51c6\u6a21\u578b + GFE \u6a21\u5757\uff1b\u6a21\u578b 2\uff08Model-2\uff09\u4e3a\u57fa\u51c6\u6a21\u578b + CNN+CIM \u6a21\u5757\uff1b\u6a21\u578b 3\uff08Model-3\uff09\u4e3a\u57fa\u51c6\u6a21\u578b + CMN+CIM \u6a21\u5757\uff1bICNET \u4e3a\u57fa\u51c6\u6a21\u578b + GFE+CMN\u3002\u91c7\u7528 8 \u79cd\u7efc\u5408\u8bc4\u4f30\u6307\u6807\uff08BLEU1\u3001BLEU2\u3001BLEU3\u3001BLEU4\u3001METEOR\u3001ROUGE\u3001CIDEr \u548cSm\u5bf9\u4e09\u4e2a\u6d4b\u8bd5\u6570\u636e\u96c6\u4e0a\u7684\u6a21\u578b\u6027\u80fd\u8fdb\u884c\u8bc4\u4f30\uff0c\u7ed3\u679c\u5982\u8868 6 &#8211; \u8868 8 \u6240\u793a\u3002<\/p>\n<\/details>\n\n\n\n<h3 class=\"wp-block-heading\">4.5.1 CMN\u7684\u4f5c\u7528<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u901a\u8fc7\u57fa\u51c6\u6a21\u578b\u3001\u6a21\u578b 2 \u548c\u6a21\u578b 3 \u63a2\u7a76 CMN \u5728\u6a21\u578b\u4e2d\u7684\u4f5c\u7528\u3002CMN \u6a21\u5757\u7528\u4e8e\u5c06 CNN \u63d0\u53d6\u7684\u5c40\u90e8\u7279\u5f81\u8f6c\u6362\u4e3a\u8bed\u4e49\u6982\u5ff5\uff0c\u4ee5\u4fbf\u4e0e\u8bcd\u5d4c\u5165\u8fdb\u884c\u4ea4\u4e92\u3002\u5bf9\u4e8e\u6a21\u578b 2\uff0c\u76f4\u63a5\u5bf9 CNN \u63d0\u53d6\u7684\u7279\u5f81\u8fdb\u884c\u5168\u5c40\u5e73\u5747\u6c60\u5316\u540e\u7528\u4e8e\u4ea4\u4e92\u3002\u4e09\u4e2a\u6570\u636e\u96c6\u6d4b\u8bd5\u96c6\u7684\u7ed3\u679c\u663e\u793a\uff0c\u6a21\u578b 3 \u7684\u6027\u80fd\u4f18\u4e8e\u6a21\u578b 2\uff0c\u800c\u6a21\u578b 2 \u7531\u4e8e\u5f15\u5165\u5c40\u90e8\u7279\u5f81\uff0c\u76f8\u8f83\u4e8e\u57fa\u51c6\u6a21\u578b\u4e5f\u6709\u663e\u8457\u63d0\u5347\u3002\u8fd9\u4e00\u7ed3\u679c\u9a8c\u8bc1\u4e86\u5c06\u5c40\u90e8\u7279\u5f81\u4e0e\u5168\u5c40\u7279\u5f81\u7ed3\u5408\u4f7f\u7528\u6709\u52a9\u4e8e\u63d0\u5347\u6a21\u578b\u6027\u80fd\u3002<\/p>\n<\/details>\n\n\n\n<h3 class=\"wp-block-heading\">4.5.2 GFE\u7684\u4f5c\u7528<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u7c7b\u4f3c\u5730\uff0c\u5728\u4e09\u4e2a\u6570\u636e\u96c6\u4e0a\u63a2\u7a76 GFE \u6a21\u5757\u7684\u5f71\u54cd\u3002\u901a\u8fc7\u5bf9\u6bd4\u57fa\u51c6\u6a21\u578b\u4e0e\u6a21\u578b 1 \u7684\u6027\u80fd\uff08\u5982\u56fe 5 \u6240\u793a\uff09\u53ef\u4ee5\u53d1\u73b0\uff0cGFE \u6a21\u5757\u901a\u8fc7\u8fc7\u6ee4\u5168\u5c40\u7279\u5f81\u4e2d\u7684\u65e0\u5173\u6216\u5f31\u76f8\u5173\u5185\u5bb9\uff0c\u6709\u6548\u63d0\u5347\u4e86\u6a21\u578b\u6027\u80fd\u3002\u4f8b\u5982\uff0c\u5728 Sydney-captions \u6570\u636e\u96c6\u7684\u5de5\u4e1a\u533a\u573a\u666f\u63cf\u8ff0\u7ed3\u679c\u4e2d\uff0c\u201c\u9053\u8def\u65c1\u7684\u8349\u576a\u201d \u8fd9\u4e00\u65e0\u5173\u8868\u8ff0\u88ab\u6210\u529f\u8fc7\u6ee4\u3002\u6b64\u5916\uff0c\u6ce8\u610f\u529b\u95e8\u63a7\u80fd\u591f\u9009\u62e9\u6027\u5730\u7a81\u51fa\u7279\u5f81\u4e2d\u7684\u91cd\u8981\u4fe1\u606f\uff0c\u4e3a\u6a21\u578b\u63d0\u4f9b\u66f4\u9c81\u68d2\u7684\u5168\u5c40\u8868\u5f81\u548c\u66f4\u4f18\u7684\u533a\u5206\u80fd\u529b\u3002<\/p>\n<\/details>\n\n\n\n<h3 class=\"wp-block-heading\">4.5.3 \u914d\u5907CIM\u7684Transformer<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u5728\u4e09\u4e2a\u6570\u636e\u96c6\u4e0a\u9a8c\u8bc1 CIM \u6a21\u5757\u7684\u6709\u6548\u6027\uff0c\u8868 9 &#8211; \u8868 11 \u5bf9\u6bd4\u4e86\u4e0d\u540c\u4ea4\u4e92\u65b9\u5f0f\u7684\u6027\u80fd\u3002\u7ed3\u679c\u663e\u793a\uff0c\u4e0e\u4f20\u7edf\u7684\u7279\u5f81\u52a0\u6cd5\u548c\u4e58\u6cd5\u4ea4\u4e92\u65b9\u5f0f\u76f8\u6bd4\uff0cCIM \u6a21\u5757\u7684\u6027\u80fd\u66f4\u4f18\uff1b\u540c\u65f6\uff0c\u56fe 6 \u4e2d\u4e09\u79cd\u4ea4\u4e92\u65b9\u5f0f\u5728 8 \u4e2a\u91cf\u5316\u6307\u6807\u4e0a\u7684\u5206\u5e03\u5bf9\u6bd4\u4e5f\u8868\u660e\uff0c\u6240\u63d0\u65b9\u6cd5\u7684\u5404\u9879\u6307\u6807\u5747\u66f4\u4f18\u3002\u5728 Sydney-captions \u6570\u636e\u96c6\u4e0a\uff0cSm\u5206\u6570\u76f8\u8f83\u4e8e\u52a0\u6cd5\u4ea4\u4e92\u63d0\u5347\u4e86 9.58%\uff0c\u76f8\u8f83\u4e8e\u70b9\u79ef\u4ea4\u4e92\u63d0\u5347\u4e86 5.57%\uff1b\u5728 UCM-captions \u6570\u636e\u96c6\u4e0a\uff0cSm\u5206\u6570\u76f8\u8f83\u4e8e\u52a0\u6cd5\u4ea4\u4e92\u63d0\u5347\u4e86 7.55%\uff0c\u76f8\u8f83\u4e8e\u70b9\u79ef\u4ea4\u4e92\u63d0\u5347\u4e86 3.95%\uff1b\u5728 NWPU-captions \u6570\u636e\u96c6\u4e0a\uff0cSm\u5206\u6570\u76f8\u8f83\u4e8e\u52a0\u6cd5\u4ea4\u4e92\u63d0\u5347\u4e86 3.79%\uff0c\u76f8\u8f83\u4e8e\u70b9\u79ef\u4ea4\u4e92\u63d0\u5347\u4e86 2.68%\u3002<\/p>\n<\/details>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-30-1024x200.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"200\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-30-1024x200.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-722\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<p class=\"has-text-align-center\">\u88686.Sydney-captions \u6570\u636e\u96c6\u4e0a\u7684\u6d88\u878d\u5b9e\u9a8c\u7ed3\u679c\uff08\u7ed3\u679c\u5747\u4ee5\u767e\u5206\u6bd4\uff08%\uff09\u8868\u793a\uff09<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-31-1024x192.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"192\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-31-1024x192.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-723\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<p class=\"has-text-align-center\">\u8868 7 UCM-captions \u6570\u636e\u96c6\u4e0a\u7684\u6d88\u878d\u5b9e\u9a8c\u7ed3\u679c\uff08\u7ed3\u679c\u5747\u4ee5\u767e\u5206\u6bd4\uff08%\uff09\u8868\u793a\uff09<\/p>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-32-1024x188.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"188\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-32-1024x188.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-724\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<p class=\"has-text-align-center\">\u8868 8 NWPU-captions \u6570\u636e\u96c6\u4e0a\u7684\u6d88\u878d\u5b9e\u9a8c\u7ed3\u679c\uff08\u7ed3\u679c\u5747\u4ee5\u767e\u5206\u6bd4\uff08%\uff09\u8868\u793a\uff09<\/p>\n\n\n\n<h2 class=\"wp-block-heading\">4.6 \u5b9a\u6027\u5206\u6790<\/h2>\n\n\n\n<h3 class=\"wp-block-heading\">4.6.1 \u63cf\u8ff0\u751f\u6210\u7ed3\u679c<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u5982\u56fe 5 \u6240\u793a\uff0c\u4ece\u4e09\u4e2a\u6570\u636e\u96c6\uff08Sydney-captions\u3001UCM-captions \u548c NWPU-captions\uff09\u4e2d\u63d0\u53d6\u6240\u63d0\u6a21\u578b\u751f\u6210\u7684\u90e8\u5206\u63cf\u8ff0\u7ed3\u679c\u3002\u4e3a\u8fdb\u884c\u5bf9\u6bd4\uff0c\u5c55\u793a\u4e86\u4e5d\u79cd\u63cf\u8ff0\u7ed3\u679c\uff1a\u6240\u63d0\u6a21\u578b\u7684\u7ed3\u679c\u3001\u5f53\u524d\u4e3b\u6d41\u65b9\u6cd5 GLCM \u7684\u7ed3\u679c\u3001\u57fa\u51c6\u6a21\u578b\u7684\u7ed3\u679c\u3001\u6240\u63d0\u6a21\u578b\u7684\u57fa\u7840\u7ec4\u4ef6\u6a21\u578b 1 \u7684\u7ed3\u679c\uff0c\u4ee5\u53ca GT1-GT5 \u4e94\u79cd\u771f\u5b9e\u6807\u6ce8\u7ed3\u679c\u3002\u4ece\u56fe 5 \u4e2d\u53ef\u4ee5\u770b\u51fa\uff0c\u4e0e GLCM \u65b9\u6cd5\u76f8\u6bd4\uff0c\u6240\u63d0\u65b9\u6cd5\u751f\u6210\u7684\u63cf\u8ff0\u5728\u76ee\u6807\u63cf\u8ff0\u51c6\u786e\u6027\u4e0a\u66f4\u4f18\uff0c\u9519\u8bef\u7387\u66f4\u4f4e\u3002<\/p>\n\n\n\n<p>\u5177\u4f53\u800c\u8a00\uff0c\u5728\u56fe 5 \u4e2d Sydney-captions \u6570\u636e\u96c6\u7684\u7b2c\u4e00\u5e45\u56fe\u50cf\u4e2d\uff0c\u524d\u666f\uff08\u9053\u8def\uff09\u9690\u85cf\u5728\u80cc\u666f\uff08\u5de5\u4e1a\u533a\uff09\u4e2d\u3002\u5728\u8fd9\u79cd\u60c5\u51b5\u4e0b\uff0c\u57fa\u51c6\u65b9\u6cd5\u5f88\u96be\u751f\u6210\u5305\u542b\u9053\u8def\u4fe1\u606f\u7684\u8bed\u53e5\uff0c\u4e14\u751f\u6210\u7ed3\u679c\u7f3a\u4e4f\u53ef\u89e3\u91ca\u6027\uff1b\u5b9e\u9645\u4e0a\uff0c\u57fa\u51c6\u65b9\u6cd5\u751f\u6210\u7684\u63cf\u8ff0\u4e2d\u786e\u5b9e\u672a\u5305\u542b\u9053\u8def\u4fe1\u606f\u3002\u7136\u800c\uff0c\u6240\u63d0\u6a21\u578b\u80fd\u591f\u6709\u6548\u63d0\u53d6\u8fd9\u4e00\u9690\u85cf\u7684\u524d\u666f\u4fe1\u606f\uff0c\u5e76\u5728\u751f\u6210\u7684\u8bed\u53e5\u4e2d\u4f53\u73b0\u51fa\u6765\uff0c\u540c\u65f6\u8bed\u53e5\u7684\u5b8c\u6574\u6027\u4e5f\u6ee1\u8db3\u8981\u6c42\u3002\u8fd9\u4e9b\u8bed\u53e5\u8d28\u91cf\u7684\u63d0\u5347\u5f97\u76ca\u4e8e\u6240\u63d0\u7684 CMN \u548c\u5168\u5c40\u7279\u5f81\u589e\u5f3a\u6a21\u5757\uff0c\u5b83\u4eec\u80fd\u591f\u5728\u7279\u5f81\u4ea4\u4e92\u5b66\u4e60\u8fc7\u7a0b\u4e2d\u51cf\u5c11\u5916\u90e8\u566a\u58f0\u7684\u5e72\u6270\u3002<\/p>\n\n\n\n<p>\u7c7b\u4f3c\u7684\u60c5\u51b5\u4e5f\u51fa\u73b0\u5728\u56fe 5 \u4e2d UCM \u6570\u636e\u96c6\u7684\u6837\u672c\u4e2d\u3002\u5177\u4f53\u800c\u8a00\uff0c\u7b2c\u4e09\u4e2a \u201c\u6e2f\u53e3\u201d \u6837\u672c\u5468\u56f4\u6709\u8bb8\u591a\u8239\u53ea\uff0c\u82e5\u6ca1\u6709\u5148\u9a8c\u77e5\u8bc6\uff0c\u5f88\u96be\u5bf9\u5176\u8fdb\u884c\u8bc6\u522b\uff0c\u56e0\u6b64\u57fa\u51c6\u65b9\u6cd5\u751f\u6210\u7684\u8bed\u53e5\u4e2d\u672a\u5305\u542b\u8fd9\u4e00\u76ee\u6807\u3002\u5728\u56fe 5 \u4e2d NWPU-captions \u6570\u636e\u96c6\u7684\u7b2c\u4e00\u4e2a\u6837\u672c\u4e2d\uff0c\u7531\u4e8e\u8239\u53ea\u6570\u91cf\u4e0e \u201c\u5cb8\u8fb9\u201d \u76ee\u6807\u8ddd\u79bb\u8f83\u8fd1\u4e14\u81ea\u8eab\u5c3a\u5bf8\u8f83\u5c0f\uff0c\u8bc6\u522b\u96be\u5ea6\u8f83\u5927\uff1b\u800c\u6240\u63d0 CMN \u878d\u5165\u4e86\u591a\u5c3a\u5ea6\u601d\u60f3\uff0c\u80fd\u591f\u5f88\u597d\u5730\u8bc6\u522b\u51fa\u8239\u53ea\u6570\u91cf\uff0c\u8fd9\u4e5f\u4f53\u73b0\u4e86\u6240\u63d0\u65b9\u6cd5\u5f3a\u5927\u7684\u6cdb\u5316\u80fd\u529b\u3002\u5176\u4ed6\u6837\u672c\u540c\u6837\u8bc1\u660e\u4e86\u6240\u63d0\u6a21\u578b\u7684\u4f18\u8d8a\u6027\uff1a\u5728\u7b2c\u4e8c\u4e2a\u6837\u672c\u4e2d\uff0c\u6240\u63d0\u65b9\u6cd5\u68c0\u6d4b\u5230\u4e86\u57fa\u51c6\u65b9\u6cd5\u9057\u6f0f\u7684 \u201c\u706b\u529b\u53d1\u7535\u5382\u201d\uff1b\u5bf9\u4e8e\u7b2c\u4e09\u4e2a\u6837\u672c\u4e2d\u7531\u7eff\u8272\u690d\u7269\u548c\u6c34\u4f53\u6784\u6210\u7684\u6e7f\u5730\uff0c\u6240\u63d0\u6a21\u578b\u5728\u63cf\u8ff0\u751f\u6210\u4e0a\u4e5f\u53d6\u5f97\u4e86\u4ee4\u4eba\u6ee1\u610f\u7684\u7ed3\u679c\u3002<\/p>\n<\/details>\n\n\n\n<figure class=\"wp-block-image size-full\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-33.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"922\" height=\"449\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-33.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-725\"  sizes=\"auto, (max-width: 922px) 100vw, 922px\" \/><\/div><\/figure>\n\n\n\n<figure class=\"wp-block-image size-full\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-34.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"934\" height=\"415\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-34.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-726\"  sizes=\"auto, (max-width: 934px) 100vw, 934px\" \/><\/div><\/figure>\n\n\n\n<figure class=\"wp-block-image size-full\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-35.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"935\" height=\"446\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-35.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-727\"  sizes=\"auto, (max-width: 935px) 100vw, 935px\" \/><\/div><\/figure>\n\n\n\n<p>\u56fe 5. \u57fa\u51c6\u65b9\u6cd5\u3001Model-1\u3001GLCM \u53ca ICNET \u65b9\u6cd5\u5728\u4e09\u4e2a\u6570\u636e\u96c6\u4e0a\u751f\u6210\u7684\u793a\u4f8b\u53e5\u5b50\u3002GT \u4ee3\u8868\u771f\u5b9e\u6807\u6ce8\u5b57\u5e55\uff08\u6216 \u201c\u53c2\u8003\u5b57\u5e55\u201d\uff09\u3002\u7ea2\u8272\u6587\u5b57\u4e0e\u5bf9\u5e94\u56fe\u50cf\u5339\u914d\u9519\u8bef\uff0c\u800c\u90e8\u5206\u5173\u952e\u7269\u4f53\u53ca\u5173\u7cfb\u7c7b\u8bcd\u8bed\u5219\u4ee5\u7eff\u8272\u7a81\u51fa\u663e\u793a<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">4.6.2 \u6ce8\u610f\u529b\u53ef\u89c6\u5316<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u501f\u52a9 CMN \u4e2d\u7684\u6ce8\u610f\u529b\u5c42\uff0c\u53ef\u5bf9\u4e0e\u8bcd\u5411\u91cf\u6700\u5339\u914d\u7684\u8bed\u4e49\u6982\u5ff5\u8fdb\u884c\u6ce8\u610f\u529b\u53ef\u89c6\u5316\u3002\u56fe 7 \u5bf9\u6bd4\u4e86\u6240\u63d0 ICNET \u4e0e GLCM \u65b9\u6cd5\u7684\u6ce8\u610f\u529b\u53ef\u89c6\u5316\u7ed3\u679c\u3002\u4e0e GLCM \u65b9\u6cd5\u76f8\u6bd4\uff0c\u5c3d\u7ba1\u5e76\u975e\u6240\u6709\u8bcd\u6c47\u90fd\u80fd\u5728\u56fe\u50cf\u4e2d\u51c6\u786e\u5b9a\u4f4d\uff0c\u4f46\u6240\u63d0\u65b9\u6cd5\u7684\u5927\u591a\u6570\u6fc0\u6d3b\u533a\u57df\u66f4\u5408\u7406\u3002\u6b64\u5916\uff0c\u6709\u8da3\u7684\u662f\uff0c\u4e00\u4e9b\u8fde\u7eed\u8bcd\u6c47\u4f1a\u5bf9\u76f8\u540c\u7684\u5c40\u90e8\u7279\u5f81\u4ea7\u751f\u54cd\u5e94\u3002\u4f8b\u5982\uff0c\u5728\u7f51\u7403\u573a\u666f\u548c\u5c45\u6c11\u533a\u573a\u666f\u4e2d\uff0c\u201c\u4e24\u4e2a\u7f51\u7403\u573a\uff08two tennis courts\uff09\u201d \u548c \u201c\u4e00\u7247\u5c45\u6c11\u533a\uff08a residential area\uff09\u201d \u5206\u522b\u805a\u96c6\u5230\u4e24\u4e2a\u5bf9\u5e94\u7684\u5c40\u90e8\u7279\u5f81\u4e0a\uff0c\u8fd9\u4e00\u73b0\u8c61\u53cd\u6620\u4e86\u81ea\u7136\u8bed\u8a00\u4e2d\u4ece\u8bcd\u6c47\u5230\u77ed\u8bed\u518d\u5230\u8bed\u53e5\u7684\u5c42\u7ea7\u7ed3\u6784\u3002<\/p>\n<\/details>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-36-1024x382.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"382\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-36-1024x382.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-728\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<figure class=\"wp-block-image size-large\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-37-1024x493.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"1024\" height=\"493\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-37-1024x493.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-729\"  sizes=\"auto, (max-width: 1024px) 100vw, 1024px\" \/><\/div><\/figure>\n\n\n\n<p class=\"has-text-align-center\">\u56fe 7. ICNET \u4e0e GLCM \u65b9\u6cd5\u5728\u5b57\u5e55\u751f\u6210\u8fc7\u7a0b\u4e2d\uff0c\u6240\u5173\u6ce8\u56fe\u50cf\u533a\u57df\u7684\u53ef\u89c6\u5316\u7ed3\u679c\u3002\u7ea2\u8272\u8868\u793a\u9ad8\u5173\u6ce8\u5ea6\u533a\u57df<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">4.6.3 \u7b97\u6cd5\u590d\u6742\u5ea6\u5bf9\u6bd4<\/h3>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u672c\u8282\u5bf9\u6bd4\u6240\u63d0\u65b9\u6cd5\u4e0e\u51e0\u79cd\u76f8\u5173\u65b9\u6cd5\u7684\u8ba1\u7b97\u590d\u6742\u5ea6\uff0c\u91cd\u70b9\u5173\u6ce8\u53c2\u6570\u6570\u91cf\u548c\u6d6e\u70b9\u8fd0\u7b97\u6b21\u6570\uff08FLOPs\uff09\u3002\u8868 12 \u5217\u51fa\u4e86\u5177\u4f53\u7684\u53c2\u6570\u6570\u91cf\u548c FLOPs \u5bf9\u6bd4\u7ed3\u679c\u3002\u4e0e GLCM \u76f8\u6bd4\uff0c\u6240\u63d0 ICNET \u7684\u53c2\u6570\u6570\u91cf\u66f4\u591a\uff0c\u8fd9\u662f\u56e0\u4e3a ICNET \u57fa\u4e8e CNN \u548c ViT \u63d0\u53d6\u66f4\u5168\u9762\u7684\u7279\u5f81\uff0c\u4f46\u540c\u65f6\u4e5f\u80fd\u751f\u6210\u66f4\u4f18\u7684\u63cf\u8ff0\u6587\u672c\u3002\u4e0e MLAT \u548c PKG-Transformer \u65b9\u6cd5\u76f8\u6bd4\uff0c\u6240\u63d0 ICNET \u7684\u53c2\u6570\u6570\u91cf\u6700\u5c11\uff0c\u4e14\u4ecd\u80fd\u53d6\u5f97\u826f\u597d\u7684\u6027\u80fd\u3002\u6700\u7ec8\uff0c\u4e0e\u57fa\u51c6\u6a21\u578b\u76f8\u6bd4\uff0cICNET \u5728\u53c2\u6570\u6570\u91cf\u76f8\u8fd1\u7684\u60c5\u51b5\u4e0b\uff0c\u6027\u80fd\u5f97\u5230\u4e86\u663e\u8457\u63d0\u5347\u3002<\/p>\n<\/details>\n\n\n\n<figure class=\"wp-block-image size-full is-resized\"><div class='fancybox-wrapper lazyload-container-unload' data-fancybox='post-images' href='http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-38.png'><img class=\"lazyload lazyload-style-1\" src=\"data:image\/svg+xml;base64,PCEtLUFyZ29uTG9hZGluZy0tPgo8c3ZnIHdpZHRoPSIxIiBoZWlnaHQ9IjEiIHhtbG5zPSJodHRwOi8vd3d3LnczLm9yZy8yMDAwL3N2ZyIgc3Ryb2tlPSIjZmZmZmZmMDAiPjxnPjwvZz4KPC9zdmc+\"  loading=\"lazy\" decoding=\"async\" width=\"536\" height=\"254\" data-original=\"http:\/\/anyone.anuoblog.cn\/wp-content\/uploads\/2025\/10\/image-38.png\" src=\"data:image\/png;base64,iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAAAXNSR0IArs4c6QAAAARnQU1BAACxjwv8YQUAAAAJcEhZcwAADsQAAA7EAZUrDhsAAAANSURBVBhXYzh8+PB\/AAffA0nNPuCLAAAAAElFTkSuQmCC\" alt=\"\" class=\"wp-image-730\" style=\"width:840px;height:auto\"  sizes=\"auto, (max-width: 536px) 100vw, 536px\" \/><\/div><\/figure>\n\n\n\n<p>\u886812.MLAT\u3001PKG-Transformer\u3001GLCM\u3001\u57fa\u51c6\u6a21\u578b\u4e0e\u6240\u63d0 ICNET \u7684\u53c2\u6570\u6570\u91cf\u548c FLOPs \u5bf9\u6bd4\uff08\u201c-\u201d \u8868\u793a\u8be5\u6587\u732e\u672a\u62a5\u544a\u6b64\u7ed3\u679c\uff09<\/p>\n\n\n\n<h1 class=\"wp-block-heading\">5 \u7ed3\u8bba<\/h1>\n\n\n\n<details class=\"wp-block-details is-layout-flow wp-block-details-is-layout-flow\"><summary>\u7ffb\u8bd1<\/summary>\n<p>\u672c\u6587\u63d0\u51fa\u4e00\u79cd\u57fa\u4e8e\u591a\u7ea7\u7279\u5f81\u805a\u5408\u4e0e\u8de8\u6a21\u6001\u7279\u5f81\u4ea4\u4e92\u7684\u65b0\u578b\u9065\u611f\u56fe\u50cf captioning \u65b9\u6cd5 \u2014\u2014ICNET\u3002\u8be5\u65b9\u6cd5\u65e8\u5728\u5229\u7528\u591a\u5c3a\u5ea6\u5c40\u90e8\u7279\u5f81\u4e0e\u5168\u5c40\u7279\u5f81\uff0c\u5f15\u5bfc\u6a21\u578b\u751f\u6210\u76ee\u6807\u4e0e\u5173\u7cfb\u63cf\u8ff0\u51c6\u786e\u7684\u6587\u672c\u3002\u5728 Transformer \u67b6\u6784\u57fa\u7840\u4e0a\uff0c\u5f15\u5165\u4e86\u6982\u5ff5\u6620\u5c04\u7f51\u7edc\uff08CMN\uff09\u3001\u6982\u5ff5\u4ea4\u4e92\u6a21\u5757\uff08CIM\uff09\u548c\u5168\u5c40\u7279\u5f81\u589e\u5f3a\uff08GFE\uff09\u6a21\u5757\uff1a\u4e3a\u63d0\u5347\u5c40\u90e8\u7279\u5f81\u8868\u5f81\u80fd\u529b\uff0c\u63d0\u51fa CMN \u6a21\u5757\uff0c\u5bf9\u4e0d\u540c\u5c3a\u5ea6\u7684\u5c40\u90e8\u7279\u5f81\u8fdb\u884c\u6709\u6548\u805a\u5408\uff0c\u751f\u6210\u9ad8\u5c42\u8bed\u4e49\u6982\u5ff5\uff1b\u901a\u8fc7 CIM \u6a21\u5757\u5bf9\u8bed\u4e49\u6982\u5ff5\u4e0e\u8bcd\u5411\u91cf\u7684\u4ea4\u4e92\u8fc7\u7a0b\u8fdb\u884c\u5efa\u6a21\uff0c\u5c06\u89c6\u89c9\u7279\u5f81\u6620\u5c04\u4e2d\u7684\u76f8\u5173\u7279\u5f81\u5411\u91cf\u6574\u5408\u4e3a\u591a\u4e2a\u76f8\u5bf9\u72ec\u7acb\u7684\u8bcd\u7279\u5f81\uff0c\u4ece\u800c\u5f25\u5408\u89c6\u89c9\u7279\u5f81\u4e0e\u8bed\u4e49\u6982\u5ff5\u4e4b\u95f4\u7684\u9e3f\u6c9f\uff1b\u4e3a\u589e\u5f3a\u6a21\u578b\u7684\u5168\u5c40\u533a\u5206\u80fd\u529b\uff0c\u901a\u8fc7 GFE \u6a21\u5757\u83b7\u53d6\u66f4\u5168\u9762\u3001\u66f4\u5177\u533a\u5206\u6027\u7684\u7279\u5f81\u8868\u5f81\uff0c\u5e76\u4ee5\u4ea4\u53c9\u6ce8\u610f\u529b\u7684\u65b9\u5f0f\u5c06\u5176\u6ce8\u5165\u6a21\u578b\uff0c\u6700\u7ec8\u751f\u6210\u9c81\u68d2\u6027\u5f3a\u3001\u6027\u80fd\u4f18\u5f02\u7684\u63cf\u8ff0\u6587\u672c\u3002<\/p>\n\n\n\n<p>\u4e0e\u5f53\u524d\u4e3b\u6d41\u6a21\u578b\u7684\u5bf9\u6bd4\u5206\u6790\u8868\u660e\uff0c\u6240\u63d0 ICNET \u5728\u4e09\u4e2a\u9065\u611f\u56fe\u50cf captioning \u6570\u636e\u96c6\u4e0a\u5747\u53d6\u5f97\u4e86\u66f4\u4f18\u7684\u63cf\u8ff0\u751f\u6210\u7ed3\u679c\uff1b\u6b64\u5916\uff0c\u901a\u8fc7\u6d88\u878d\u5b9e\u9a8c\u5206\u6790\u4e86\u5404\u6a21\u5757\u5bf9\u6a21\u578b\u6027\u80fd\u7684\u8d21\u732e\u3002<\/p>\n<\/details>\n\n\n\n<p>1\u3001\u672c\u6587\u63d0\u51fa\u7684\u4e00\u79cd\u57fa\u4e8e\u591a\u7ea7\u7279\u5f81\u805a\u5408\u4e0e\u8de8\u6a21\u6001\u7279\u5f81\u4ea4\u4e92\u7684\u65b0\u578b\u9065\u611f\u56fe\u50cfcaptioning\u65b9\u6cd5\uff0c\u8be5\u65b9\u6cd5\u65e8\u5728\u5229\u7528\u591a\u5c3a\u5ea6\u5c40\u90e8\u7279\u5f81\u4e0e\u5168\u5c40\u7279\u5f81\uff0c\u5f15\u5bfc\u6a21\u578b\u751f\u6210\u76ee\u6807\u4e0e\u5173\u7cfb\u63cf\u8ff0\u51c6\u786e\u7684\u6587\u672c\u3002<\/p>\n\n\n\n<p>2\u3001\u521b\u65b0\u6a21\u5757\uff1a\u5f15\u5165\u4e86\u6982\u5ff5\u6620\u5c04\u7f51\u7edc\uff08CMN\uff09\u3001\u6982\u5ff5\u4ea4\u4e92\u6a21\u5757\uff08CIM\uff09\u548c\u5168\u5c40\u7279\u5f81\u589e\u5f3a\uff08GFE\uff09\u6a21\u5757<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Interactive Concept Network Enhanced Transformer for Re [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[],"class_list":["post-696","post","type-post","status-publish","format-standard","hentry","category-message"],"_links":{"self":[{"href":"http:\/\/anyone.anuoblog.cn\/index.php\/wp-json\/wp\/v2\/posts\/696","targetHints":{"allow":["GET"]}}],"collection":[{"href":"http:\/\/anyone.anuoblog.cn\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/anyone.anuoblog.cn\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/anyone.anuoblog.cn\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/anyone.anuoblog.cn\/index.php\/wp-json\/wp\/v2\/comments?post=696"}],"version-history":[{"count":14,"href":"http:\/\/anyone.anuoblog.cn\/index.php\/wp-json\/wp\/v2\/posts\/696\/revisions"}],"predecessor-version":[{"id":808,"href":"http:\/\/anyone.anuoblog.cn\/index.php\/wp-json\/wp\/v2\/posts\/696\/revisions\/808"}],"wp:attachment":[{"href":"http:\/\/anyone.anuoblog.cn\/index.php\/wp-json\/wp\/v2\/media?parent=696"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/anyone.anuoblog.cn\/index.php\/wp-json\/wp\/v2\/categories?post=696"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/anyone.anuoblog.cn\/index.php\/wp-json\/wp\/v2\/tags?post=696"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}