{"id":94840,"date":"2026-08-15T12:32:56","date_gmt":"2026-08-15T04:32:56","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/94840.html"},"modified":"2026-08-15T12:32:56","modified_gmt":"2026-08-15T04:32:56","slug":"yolo11%e6%94%b9%e8%bf%9b-c3k2%e8%9e%8d%e5%90%88-ira%e5%80%92%e6%ae%8b%e5%b7%ae%e6%b3%a8%e6%84%8f%e5%8a%9b%e6%a8%a1%e5%9d%97%ef%bc%8c%e8%bd%bb%e9%87%8f%e8%9e%8d%e5%90%88%e5%8d%b7%e7%a7%af%e8%a1%a8","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/94840.html","title":{"rendered":"YOLO11\u6539\u8fdb - C3k2\u878d\u5408 | IRA\u5012\u6b8b\u5dee\u6ce8\u610f\u529b\u6a21\u5757\uff0c\u8f7b\u91cf\u878d\u5408\u5377\u79ef\u8868\u5f81\u4e0e\u901a\u9053\u7a7a\u95f4\u6ce8\u610f\uff0c\u52a9\u529b\u79fb\u52a8\u7aef\u591a\u5149\u8c31\u989c\u8272\u6821\u6b63\u8bef\u5dee\u5927\u5e45\u964d\u4f4e | CVPR 2026"},"content":{"rendered":"

\u524d\u8a00<\/h2>\n

\u672c\u6587\u4ecb\u7ecd\u4e86\u7528\u4e8e\u8f7b\u91cf\u7aef\u5230\u7aef\u989c\u8272\u6821\u6b63\u7684\u5012\u6b8b\u5dee\u6ce8\u610f\u529b\u6a21\u5757 IRA，\u878d\u5408\u4e86 MobileNet \u98ce\u683c\u5012\u6b8b\u5dee\u7ed3\u6784\u4e0e\u901a\u9053\u3001\u7a7a\u95f4\u6ce8\u610f\u529b\u673a\u5236\u3002\u8be5\u6a21\u5757\u901a\u8fc7\u6df1\u5ea6\u53ef\u5206\u79bb\u5377\u79ef\u9ad8\u6548\u63d0\u53d6\u5c40\u90e8\u7279\u5f81，\u5e76\u5229\u7528\u6ce8\u610f\u529b\u81ea\u9002\u5e94\u5f3a\u5316\u4e0e\u989c\u8272\u3001\u7eb9\u7406\u3001\u8fb9\u7f18\u76f8\u5173\u7684\u5173\u952e\u4fe1\u606f，\u5728\u8f83\u4f4e\u8ba1\u7b97\u6210\u672c\u4e0b\u63d0\u5347\u7279\u5f81\u9009\u62e9\u80fd\u529b\u4e0e\u8868\u8fbe\u7a33\u5b9a\u6027\u3002\u6211\u4eec\u5c06 IRA \u6210\u529f\u96c6\u6210\u8fdb YOLO11 \u7684 C3k2 \u6a21\u5757，\u6784\u5efa C3k2_IRA，\u7528\u4e8e\u589e\u5f3a\u4e3b\u5e72\u4e0e\u9888\u90e8\u7f51\u7edc\u7684\u5c40\u90e8\u7ec6\u8282\u5efa\u6a21\u548c\u6709\u6548\u7279\u5f81\u7b5b\u9009\u80fd\u529b，\u4ece\u800c\u63d0\u5347\u76ee\u6807\u68c0\u6d4b\u4e2d\u7684\u7279\u5f81\u8868\u8fbe\u6548\u679c\u3002<\/p>\n

\u6587\u7ae0\u76ee\u5f55： YOLO11\u6539\u8fdb\u5927\u5168：\u5377\u79ef\u5c42\u3001\u8f7b\u91cf\u5316\u3001\u6ce8\u610f\u529b\u673a\u5236\u3001\u635f\u5931\u51fd\u6570\u3001Backbone\u3001SPPF\u3001Neck\u3001\u68c0\u6d4b\u5934\u5168\u65b9\u4f4d\u4f18\u5316\u6c47\u603b<\/p>\n

\u4e13\u680f\u94fe\u63a5: YOLO11\u6539\u8fdb\u4e13\u680f<\/p>\n

\u4ecb\u7ecd<\/h2>\n

\"image-20260812215334303\"<\/p>\n

\u8fd1\u5e74\u6765，\u5feb\u7167\u5f0f\u591a\u5149\u8c31（MS）\u6210\u50cf\u7684\u53d1\u5c55\u4f7f\u7d27\u51d1\u3001\u4f4e\u6210\u672c\u7684\u5149\u8c31\u4f20\u611f\u5668\u80fd\u591f\u5e94\u7528\u4e8e\u6d88\u8d39\u7ea7\u548c\u79fb\u52a8\u8bbe\u5907\u3002\u76f8\u6bd4\u4f20\u7edf RGB \u4f20\u611f\u5668，\u8fd9\u7c7b\u7cfb\u7edf\u80fd\u591f\u6355\u83b7\u66f4\u4e30\u5bcc\u7684\u5149\u8c31\u4fe1\u606f，\u4ece\u800c\u589e\u5f3a\u5305\u62ec\u989c\u8272\u6821\u6b63\u5728\u5185\u7684\u5173\u952e\u6210\u50cf\u4efb\u52a1\u3002\u7136\u800c，\u5927\u591a\u6570\u73b0\u6709\u65b9\u6cd5\u5c06\u989c\u8272\u6821\u6b63\u6d41\u7a0b\u62c6\u5206\u4e3a\u591a\u4e2a\u72ec\u7acb\u9636\u6bb5，\u5e76\u4e14\u5f80\u5f80\u5728\u6d41\u7a0b\u65e9\u671f\u5c31\u4e22\u5f03 MS \u6570\u636e\u3002\u672c\u6587\u63d0\u51fa\u4e00\u79cd\u7edf\u4e00\u7684\u3001\u57fa\u4e8e\u5b66\u4e60\u7684\u6846\u67b6，\u7528\u4e8e\u6267\u884c\u7aef\u5230\u7aef\u989c\u8272\u6821\u6b63，\u5e76\u8054\u5408\u5229\u7528\u6765\u81ea\u9ad8\u5206\u8fa8\u7387 RGB \u4f20\u611f\u5668\u548c\u8f85\u52a9\u4f4e\u5206\u8fa8\u7387 MS \u4f20\u611f\u5668\u7684\u6570\u636e\u3002\u8be5\u65b9\u6cd5\u5c06\u5b8c\u6574\u989c\u8272\u6821\u6b63\u6d41\u7a0b\u6574\u5408\u5230\u5355\u4e00\u6a21\u578b\u4e2d，\u751f\u6210\u4e00\u81f4\u4e14\u989c\u8272\u51c6\u786e\u7684\u8f93\u51fa\u3002\u4f5c\u8005\u901a\u8fc7\u91cd\u6784\u4e24\u79cd\u4e0d\u540c\u7684\u5148\u8fdb\u56fe\u50cf\u5230\u56fe\u50cf\u67b6\u6784，\u5c55\u793a\u4e86\u8be5\u6846\u67b6\u7684\u7075\u6d3b\u6027\u548c\u901a\u7528\u6027\u3002\u4e3a\u652f\u6301\u8bad\u7ec3\u548c\u8bc4\u4f30，\u672c\u6587\u805a\u5408\u5e76\u91cd\u65b0\u5229\u7528\u516c\u5f00\u5149\u8c31\u6570\u636e\u96c6，\u6784\u5efa\u4e86\u4e13\u7528\u6570\u636e\u96c6，\u5e76\u5728\u591a\u79cd RGB \u76f8\u673a\u5149\u8c31\u7075\u654f\u5ea6\u4e0b\u8fdb\u884c\u6e32\u67d3\u3002\u5927\u91cf\u5b9e\u9a8c\u8868\u660e，\u8be5\u65b9\u6cd5\u63d0\u5347\u4e86\u989c\u8272\u51c6\u786e\u6027\u548c\u7a33\u5b9a\u6027，\u76f8\u6bd4\u4ec5 RGB \u548c MS \u9a71\u52a8\u7684\u57fa\u7ebf，\u8bef\u5dee\u6700\u9ad8\u964d\u4f4e 50%\u3002\u4ee3\u7801\u3001\u6a21\u578b\u548c\u6570\u636e\u96c6\u5747\u5df2\u516c\u5f00\u3002<\/p>\n

\u6587\u7ae0\u94fe\u63a5<\/h2>\n

\u8bba\u6587\u5730\u5740：\u8bba\u6587\u5730\u5740<\/p>\n

\u4ee3\u7801\u5730\u5740：\u4ee3\u7801\u5730\u5740<\/p>\n

\u57fa\u672c\u539f\u7406<\/h2>\n

1. \u89e3\u51b3\u7684\u5173\u952e\u95ee\u9898<\/p>\n

IRA \u6240\u5728\u7684\u8fd9\u7bc7\u8bba\u6587\u4e3b\u8981\u89e3\u51b3\u79fb\u52a8\u7aef\u76f8\u673a\u989c\u8272\u6821\u6b63\u4e2d RGB \u4fe1\u606f\u4e0d\u8db3\u3001\u4f20\u7edf\u7ba1\u7ebf\u8bef\u5dee\u9010\u7ea7\u4f20\u64ad\u4ee5\u53ca\u591a\u5149\u8c31\u4fe1\u606f\u5229\u7528\u4e0d\u5145\u5206\u7684\u95ee\u9898\u3002\u4f20\u7edf\u76f8\u673a\u989c\u8272\u6821\u6b63\u901a\u5e38\u88ab\u62c6\u6210\u81ea\u52a8\u767d\u5e73\u8861（AWB）\u548c\u989c\u8272\u7a7a\u95f4\u53d8\u6362（CST）\u4e24\u4e2a\u9636\u6bb5：\u524d\u8005\u4f30\u8ba1\u5e76\u62b5\u6d88\u573a\u666f\u5149\u7167\u989c\u8272，\u540e\u8005\u5c06\u767d\u5e73\u8861\u540e\u7684\u76f8\u673a RGB \u6620\u5c04\u5230 CIE XYZ \u6216 ProPhoto RGB \u7b49\u8bbe\u5907\u65e0\u5173\u989c\u8272\u7a7a\u95f4\u3002\u8fd9\u79cd\u6a21\u5757\u5316\u6d41\u7a0b\u867d\u7136\u9ad8\u6548，\u4f46\u6bcf\u4e2a\u9636\u6bb5\u76f8\u4e92\u72ec\u7acb，\u524d\u4e00\u6b65\u7684\u7167\u660e\u4f30\u8ba1\u8bef\u5dee\u4f1a\u5f71\u54cd\u540e\u7eed\u989c\u8272\u53d8\u6362，\u590d\u6742\u5149\u7167\u4e0b\u5bb9\u6613\u51fa\u73b0\u504f\u8272\u548c\u4e0d\u7a33\u5b9a\u3002<\/p>\n

\u66f4\u6839\u672c\u7684\u95ee\u9898\u5728\u4e8e，\u666e\u901a RGB \u4f20\u611f\u5668\u53ea\u6709\u4e09\u4e2a\u5bbd\u5149\u8c31\u901a\u9053，\u96be\u4ee5\u5145\u5206\u533a\u5206\u7269\u4f53\u53cd\u5c04\u7387\u548c\u573a\u666f\u5149\u7167\u3002\u5f53\u4e0d\u540c\u5149\u7167\u3001\u4e0d\u540c\u6750\u8d28\u6216\u4e0d\u540c\u76f8\u673a\u5149\u8c31\u54cd\u5e94\u7ec4\u5408\u5728\u4e00\u8d77\u65f6，RGB \u6570\u503c\u53ef\u80fd\u9ad8\u5ea6\u76f8\u4f3c，\u4f46\u771f\u5b9e\u989c\u8272\u5374\u4e0d\u540c，\u8fd9\u4f1a\u9020\u6210\u989c\u8272\u6052\u5e38\u6027\u548c\u989c\u8272\u7a7a\u95f4\u6620\u5c04\u7684\u6b67\u4e49\u3002\u5feb\u7167\u5f0f\u4f4e\u5206\u8fa8\u7387\u591a\u5149\u8c31\u4f20\u611f\u5668\u53ef\u4ee5\u63d0\u4f9b\u989d\u5916\u5149\u8c31\u7ebf\u7d22，\u4f46\u73b0\u6709\u65b9\u6cd5\u5e38\u5e38\u53ea\u628a MS \u4fe1\u606f\u7528\u4e8e\u7167\u660e\u4f30\u8ba1，\u4e4b\u540e\u4ecd\u7136\u56de\u5230\u4f20\u7edf RGB \u7ba1\u7ebf，\u5bfc\u81f4\u591a\u5149\u8c31\u4fe1\u606f\u6ca1\u6709\u8d2f\u7a7f\u6574\u4e2a\u989c\u8272\u6821\u6b63\u8fc7\u7a0b\u3002IRA \u6a21\u5757\u6240\u5728\u7684 LPIENet \u6539\u9020\u65b9\u6848，\u5c31\u662f\u4e3a\u4e86\u5728\u8f7b\u91cf\u56fe\u50cf\u5230\u56fe\u50cf\u7f51\u7edc\u4e2d\u66f4\u6709\u6548\u5730\u5904\u7406 RGB \u4e0e MS \u7279\u5f81，\u8ba9\u7f51\u7edc\u7aef\u5230\u7aef\u5730\u8054\u5408\u5b8c\u6210\u7167\u660e\u4f30\u8ba1\u3001\u7167\u660e\u62b5\u6d88\u548c\u989c\u8272\u7a7a\u95f4\u53d8\u6362\u3002<\/p>\n

2. \u6574\u4f53\u67b6\u6784<\/p>\n

\u8bba\u6587\u63d0\u51fa\u7684\u662f\u4e00\u4e2a\u53cc\u8f93\u5165\u7aef\u5230\u7aef\u989c\u8272\u6821\u6b63\u6846\u67b6：\u8f93\u5165\u5305\u62ec\u9ad8\u5206\u8fa8\u7387 RGB \u56fe\u50cf\u548c\u8f85\u52a9\u4f4e\u5206\u8fa8\u7387\u591a\u5149\u8c31\u56fe\u50cf，\u8f93\u51fa\u4e3a\u989c\u8272\u6821\u6b63\u540e\u7684 CIE XYZ \u56fe\u50cf\u3002\u4f5c\u8005\u6ca1\u6709\u4ece\u96f6\u8bbe\u8ba1\u5168\u65b0\u7f51\u7edc，\u800c\u662f\u91cd\u6784\u4e86\u4e24\u7c7b\u5df2\u6709\u8f7b\u91cf\u56fe\u50cf\u5230\u56fe\u50cf\u67b6\u6784，\u5176\u4e2d LPIENet \u5206\u652f\u7684\u6838\u5fc3\u7ec4\u4ef6\u5c31\u662f IRA block\u3002LPIENet \u539f\u672c\u9762\u5411\u79fb\u52a8\u7aef\u5b9e\u65f6\u56fe\u50cf\u589e\u5f3a，\u91c7\u7528\u7c7b\u4f3c U-Net \u7684\u7f16\u7801\u5668-\u89e3\u7801\u5668\u7ed3\u6784，\u9002\u5408\u9700\u8981\u7a7a\u95f4\u4e00\u81f4\u8f93\u51fa\u7684\u989c\u8272\u6821\u6b63\u4efb\u52a1\u3002<\/p>\n

\u5728\u6539\u9020\u540e\u7684 LPIENet \u4e2d，RGB \u56fe\u50cf\u9996\u5148\u7ecf\u8fc7 RGB encoder \u63d0\u53d6\u591a\u5c3a\u5ea6\u7279\u5f81，\u7f51\u7edc\u5305\u542b\u4e09\u4e2a\u7f16\u7801\u5757\u3001\u4e24\u4e2a\u89e3\u7801\u5757\u3001\u8df3\u8dc3\u8fde\u63a5\u4ee5\u53ca\u6700\u7ec8 refinement block\u3002\u4e3a\u4e86\u5f15\u5165\u591a\u5149\u8c31\u6a21\u6001，\u8bba\u6587\u989d\u5916\u52a0\u5165\u4e00\u4e2a spectral encoder，\u8be5\u5206\u652f\u955c\u50cf RGB encoder \u7684\u7ed3\u6784，\u7531\u4e09\u4e2a IRA block \u7ec4\u6210，\u4f46\u4e0d\u6267\u884c\u4e0b\u91c7\u6837，\u4ee5\u9002\u914d\u4f4e\u5206\u8fa8\u7387 MS \u8f93\u5165\u5e76\u63d0\u53d6\u5149\u8c31\u8f85\u52a9\u7279\u5f81\u3002\u6bcf\u4e00\u5c42\u591a\u5149\u8c31\u7f16\u7801\u5668\u63d0\u53d6\u7684\u7279\u5f81，\u4f1a\u901a\u8fc7 skip connection \u4e0e RGB \u7f16\u7801\u5668\u5bf9\u5e94\u5c42\u7684\u7279\u5f81\u878d\u5408，\u5177\u4f53\u91c7\u7528\u9010\u5143\u7d20\u76f8\u52a0，\u5e76\u5728\u6bcf\u4e2a\u89e3\u7801\u9636\u6bb5\u4e4b\u524d\u6ce8\u5165\u3002\u8fd9\u6837，RGB \u5206\u652f\u8d1f\u8d23\u4fdd\u7559\u9ad8\u5206\u8fa8\u7387\u7a7a\u95f4\u7ed3\u6784，MS \u5206\u652f\u8d1f\u8d23\u63d0\u4f9b\u66f4\u4e30\u5bcc\u7684\u5149\u8c31\u5148\u9a8c，IRA \u5219\u4f5c\u4e3a\u4e24\u4e2a\u5206\u652f\u4e2d\u7684\u8f7b\u91cf\u7279\u5f81\u63d0\u53d6\u4e0e\u6ce8\u610f\u529b\u589e\u5f3a\u5355\u5143，\u4e3a\u540e\u7eed\u989c\u8272\u6821\u6b63\u8f93\u51fa\u63d0\u4f9b\u7a33\u5b9a\u8868\u5f81\u3002<\/p>\n

3. \u6280\u672f\u539f\u7406<\/p>\n

IRA \u7684\u5168\u79f0\u662f Inverted Residual Attention，\u5373\u5012\u6b8b\u5dee\u6ce8\u610f\u529b\u6a21\u5757\u3002\u5b83\u7ed3\u5408\u4e86 MobileNet \u98ce\u683c\u7684\u8f7b\u91cf\u5377\u79ef\u7ed3\u6784\u548c CBAM \u7c7b\u4f3c\u7684\u901a\u9053\/\u7a7a\u95f4\u6ce8\u610f\u529b\u601d\u60f3\u3002\u5012\u6b8b\u5dee\u7ed3\u6784\u7684\u4ef7\u503c\u5728\u4e8e\u4ee5\u8f83\u4f4e\u53c2\u6570\u91cf\u548c\u8ba1\u7b97\u91cf\u5b8c\u6210\u6709\u6548\u7279\u5f81\u53d8\u6362：\u901a\u5e38\u5148\u901a\u8fc7 1\u00d71 \u5377\u79ef\u8fdb\u884c\u901a\u9053\u53d8\u6362，\u518d\u4f7f\u7528\u6df1\u5ea6\u53ef\u5206\u79bb\u5377\u79ef\u63d0\u53d6\u5c40\u90e8\u7a7a\u95f4\u4fe1\u606f，\u6700\u540e\u518d\u901a\u8fc7 1\u00d71 \u5377\u79ef\u6295\u5f71\u56de\u76ee\u6807\u901a\u9053\u7ef4\u5ea6，\u5e76\u901a\u8fc7\u6b8b\u5dee\u8fde\u63a5\u4fdd\u6301\u4fe1\u606f\u6d41\u52a8\u3002\u8fd9\u79cd\u8bbe\u8ba1\u975e\u5e38\u9002\u5408\u79fb\u52a8\u7aef\u989c\u8272\u6821\u6b63，\u56e0\u4e3a\u989c\u8272\u6821\u6b63\u9700\u8981\u5bf9\u6574\u5e45\u56fe\u50cf\u4ea7\u751f\u7a7a\u95f4\u8fde\u7eed\u3001\u7ec6\u8282\u4e00\u81f4\u7684\u8f93\u51fa，\u540c\u65f6\u53c8\u4e0d\u80fd\u5f15\u5165\u8fc7\u91cd\u8ba1\u7b97\u3002<\/p>\n

\u5728 IRA \u4e2d，\u6ce8\u610f\u529b\u673a\u5236\u8fdb\u4e00\u6b65\u589e\u5f3a\u4e86\u5012\u6b8b\u5dee\u5377\u79ef\u7684\u9009\u62e9\u80fd\u529b\u3002\u901a\u9053\u6ce8\u610f\u529b\u7528\u4e8e\u5224\u65ad\u54ea\u4e9b\u7279\u5f81\u901a\u9053\u5bf9\u5f53\u524d\u989c\u8272\u6821\u6b63\u66f4\u91cd\u8981，\u4f8b\u5982\u4e0e\u5149\u7167\u989c\u8272\u3001\u6750\u8d28\u53cd\u5c04\u3001\u5c40\u90e8\u8272\u504f\u76f8\u5173\u7684\u54cd\u5e94；\u7a7a\u95f4\u6ce8\u610f\u529b\u5219\u7528\u4e8e\u5b9a\u4f4d\u56fe\u50cf\u4e2d\u66f4\u9700\u8981\u6821\u6b63\u6216\u66f4\u5177\u53c2\u8003\u4ef7\u503c\u7684\u533a\u57df，\u4f8b\u5982\u9ad8\u4eae\u533a\u57df\u3001\u9634\u5f71\u533a\u57df\u3001\u8272\u5f69\u8fb9\u754c\u548c\u6750\u8d28\u53d8\u5316\u4f4d\u7f6e\u3002\u8bba\u6587\u56fe\u793a\u4e2d，IRA block \u5305\u542b IRB \u5377\u79ef\u5206\u652f，\u5e76\u5e76\u884c\u7ed3\u5408 channel attention \u4e0e spatial attention，\u6700\u7ec8\u901a\u8fc7\u9010\u5143\u7d20\u4e58\u6cd5\u548c\u6b8b\u5dee\/\u878d\u5408\u64cd\u4f5c\u5f3a\u5316\u6709\u6548\u7279\u5f81\u3002\u76f8\u6bd4\u7eaf\u5377\u79ef\u6a21\u5757，IRA \u80fd\u66f4\u81ea\u9002\u5e94\u5730\u9009\u62e9\u989c\u8272\u76f8\u5173\u4fe1\u606f；\u76f8\u6bd4\u91cd\u578b Transformer \u6ce8\u610f\u529b，\u5b83\u53c8\u4fdd\u6301\u4e86\u79fb\u52a8\u7aef\u53cb\u597d\u7684\u8f7b\u91cf\u7279\u6027\u3002<\/p>\n

\u5728\u591a\u5149\u8c31\u989c\u8272\u6821\u6b63\u573a\u666f\u4e2d，IRA \u7684\u4f5c\u7528\u4e0d\u4ec5\u662f\u589e\u5f3a RGB \u56fe\u50cf\u7279\u5f81，\u8fd8\u627f\u62c5\u4e86\u5149\u8c31\u7279\u5f81\u7f16\u7801\u7684\u57fa\u7840\u5355\u5143\u3002\u5bf9\u4e8e spectral encoder \u6765\u8bf4，\u4e09\u4e2a IRA block \u4ece\u4f4e\u5206\u8fa8\u7387 MS \u56fe\u50cf\u4e2d\u63d0\u53d6\u4e0e\u7167\u660e\u3001\u53cd\u5c04\u7387\u548c\u76f8\u673a\u5149\u8c31\u54cd\u5e94\u76f8\u5173\u7684\u8f85\u52a9\u7279\u5f81；\u8fd9\u4e9b\u7279\u5f81\u901a\u8fc7\u8df3\u8dc3\u8fde\u63a5\u6ce8\u5165 RGB \u89e3\u7801\u6d41\u7a0b，\u4f7f\u7f51\u7edc\u4e0d\u518d\u53ea\u4f9d\u8d56 RGB \u4e09\u901a\u9053\u731c\u6d4b\u771f\u5b9e\u989c\u8272，\u800c\u662f\u5728\u7aef\u5230\u7aef\u6620\u5c04\u4e2d\u6301\u7eed\u5229\u7528 MS \u7ebf\u7d22\u3002\u5b9e\u9a8c\u7ed3\u679c\u4e5f\u652f\u6301\u8fd9\u4e00\u8bbe\u8ba1：\u8bba\u6587\u5728\u6d88\u878d\u4e2d\u79fb\u9664 spectral encoder \u540e，\u6027\u80fd\u660e\u663e\u4e0b\u964d；\u52a0\u5165 MS \u4fe1\u606f\u540e，LPIENet\u3001LPIENet-small \u548c cmKAN-light \u7684\u5e73\u5747 \u0394E00 \u4e0e Reproduction Error \u5747\u663e\u8457\u6539\u5584，\u6574\u4f53\u8bef\u5dee\u6700\u9ad8\u53ef\u6bd4 RGB-only \u8bbe\u7f6e\u964d\u4f4e\u7ea6 50%\u3002\u56e0\u6b64，IRA \u53ef\u4ee5\u88ab\u7406\u89e3\u4e3a\u8be5\u8f7b\u91cf\u591a\u5149\u8c31\u989c\u8272\u6821\u6b63\u6846\u67b6\u4e2d\u7684\u5173\u952e\u79ef\u6728：\u5b83\u7528\u4f4e\u6210\u672c\u5377\u79ef\u4fdd\u8bc1\u79fb\u52a8\u7aef\u53ef\u90e8\u7f72\u6027，\u7528\u901a\u9053\u4e0e\u7a7a\u95f4\u6ce8\u610f\u529b\u589e\u5f3a\u989c\u8272\u76f8\u5173\u8868\u5f81，\u518d\u901a\u8fc7 RGB-MS \u53cc\u5206\u652f\u878d\u5408\u63d0\u5347\u590d\u6742\u5149\u7167\u4e0b\u7684\u989c\u8272\u51c6\u786e\u6027\u548c\u7a33\u5b9a\u6027\u3002<\/p>\n

\u6838\u5fc3\u4ee3\u7801<\/h2>\n

YOLO11\u5f15\u5165\u4ee3\u7801<\/h2>\n

\u5728\u6839\u76ee\u5f55\u4e0b\u7684ultralytics\/nn\/\u76ee\u5f55，\u65b0\u5efa\u4e00\u4e2a C3k2\u76ee\u5f55，\u7136\u540e\u65b0\u5efa\u4e00\u4e2a\u4ee5 C3k2_IRA\u4e3a\u6587\u4ef6\u540d\u7684py\u6587\u4ef6， \u628a\u4ee3\u7801\u62f7\u8d1d\u8fdb\u53bb\u3002<\/p>\n

import<\/span> torch
\nimport<\/span> torch.<\/span>nn as<\/span> nn
\nimport<\/span> torch.<\/span>nn.<\/span>functional as<\/span> F
\nfrom<\/span> ultralytics.<\/span>nn.<\/span>modules.<\/span>conv import<\/span> RepConv,<\/span> Conv
\nfrom<\/span> ultralytics.<\/span>nn.<\/span>modules.<\/span>block import<\/span> C3k,<\/span> C3k2<\/p>\n

class<\/span> AttentionBlock<\/span>(<\/span>nn.<\/span>Module)<\/span>:<\/span>
\n def<\/span> __init__<\/span>(<\/span>self,<\/span> dim:<\/span> int<\/span>)<\/span>:<\/span>
\n super<\/span>(<\/span>AttentionBlock,<\/span> self)<\/span>.<\/span>__init__(<\/span>)<\/span>
\n self.<\/span>_spatial_attention_conv =<\/span> nn.<\/span>Conv2d(<\/span>2<\/span>,<\/span> dim,<\/span> kernel_size=<\/span>3<\/span>,<\/span> padding=<\/span>1<\/span>)<\/span><\/p>\n

# Channel attention MLP<\/span>
\n self.<\/span>_channel_attention_conv0 =<\/span> nn.<\/span>Conv2d(<\/span>1<\/span>,<\/span> dim,<\/span> kernel_size=<\/span>1<\/span>,<\/span> padding=<\/span>0<\/span>)<\/span>
\n self.<\/span>_channel_attention_conv1 =<\/span> nn.<\/span>Conv2d(<\/span>dim,<\/span> dim,<\/span> kernel_size=<\/span>1<\/span>,<\/span> padding=<\/span>0<\/span>)<\/span><\/p>\n

self.<\/span>_out_conv =<\/span> nn.<\/span>Conv2d(<\/span>2<\/span> *<\/span> dim,<\/span> dim,<\/span> kernel_size=<\/span>1<\/span>,<\/span> padding=<\/span>0<\/span>)<\/span><\/p>\n

def<\/span> forward<\/span>(<\/span>self,<\/span> x:<\/span> torch.<\/span>Tensor)<\/span>:<\/span>
\n if<\/span> len<\/span>(<\/span>x.<\/span>shape)<\/span> !=<\/span> 4<\/span>:<\/span>
\n raise<\/span> ValueError(<\/span>f"Expected [B, C, H, W] input, got <\/span>{<\/span>x.<\/span>shape}<\/span><\/span>."<\/span><\/span>)<\/span><\/p>\n

# Spatial attention<\/span>
\n mean =<\/span> torch.<\/span>mean(<\/span>x,<\/span> dim=<\/span>1<\/span>,<\/span> keepdim=<\/span>True<\/span>)<\/span> # Mean\/Max on C axis<\/span>
\n max<\/span>,<\/span> _ =<\/span> torch.<\/span>max<\/span>(<\/span>x,<\/span> dim=<\/span>1<\/span>,<\/span> keepdim=<\/span>True<\/span>)<\/span>
\n spatial_attention =<\/span> torch.<\/span>cat(<\/span>[<\/span>mean,<\/span> max<\/span>]<\/span>,<\/span> dim=<\/span>1<\/span>)<\/span> # [B, 2, H, W]<\/span>
\n spatial_attention =<\/span> self.<\/span>_spatial_attention_conv(<\/span>spatial_attention)<\/span>
\n spatial_attention =<\/span> torch.<\/span>sigmoid(<\/span>spatial_attention)<\/span> *<\/span> x<\/p>\n

# Channel attention. TODO: Correct that it only uses average pool contrary to CBAM?<\/span>
\n # NOTE\/TODO: This differs from CBAM as it uses Channel pooling, not spatial pooling!<\/span>
\n # In a way, this is 2x spatial attention<\/span>
\n channel_attention =<\/span> torch.<\/span>relu(<\/span>self.<\/span>_channel_attention_conv0(<\/span>mean)<\/span>)<\/span>
\n channel_attention =<\/span> self.<\/span>_channel_attention_conv1(<\/span>channel_attention)<\/span>
\n channel_attention =<\/span> torch.<\/span>sigmoid(<\/span>channel_attention)<\/span> *<\/span> x<\/p>\n

attention =<\/span> torch.<\/span>cat(<\/span>[<\/span>spatial_attention,<\/span> channel_attention]<\/span>,<\/span> dim=<\/span>1<\/span>)<\/span> # [B, 2*dim, H, W]<\/span>
\n attention =<\/span> self.<\/span>_out_conv(<\/span>attention)<\/span>
\n return<\/span> x +<\/span> attention<\/p>\n

class<\/span> BaseBlock<\/span>(<\/span>nn.<\/span>Module)<\/span>:<\/span>
\n def<\/span> __init__<\/span>(<\/span>self,<\/span> channels:<\/span> int<\/span>)<\/span>:<\/span>
\n super<\/span>(<\/span>BaseBlock,<\/span> self)<\/span>.<\/span>__init__(<\/span>)<\/span><\/p>\n

self.<\/span>_conv0 =<\/span> nn.<\/span>Conv2d(<\/span>channels,<\/span> channels,<\/span> kernel_size=<\/span>1<\/span>)<\/span>
\n self.<\/span>_dw_conv =<\/span> nn.<\/span>Conv2d(<\/span>channels,<\/span> channels,<\/span> kernel_size=<\/span>3<\/span>,<\/span> padding=<\/span>1<\/span>,<\/span> groups=<\/span>channels)<\/span>
\n self.<\/span>_conv1 =<\/span> nn.<\/span>Conv2d(<\/span>channels,<\/span> channels,<\/span> kernel_size=<\/span>1<\/span>)<\/span><\/p>\n

self.<\/span>_conv2 =<\/span> nn.<\/span>Conv2d(<\/span>channels,<\/span> channels,<\/span> kernel_size=<\/span>1<\/span>)<\/span>
\n self.<\/span>_conv3 =<\/span> nn.<\/span>Conv2d(<\/span>channels,<\/span> channels,<\/span> kernel_size=<\/span>1<\/span>)<\/span><\/p>\n

def<\/span> forward<\/span>(<\/span>self,<\/span> x:<\/span> torch.<\/span>Tensor)<\/span>:<\/span>
\n features =<\/span> self.<\/span>_conv0(<\/span>x)<\/span>
\n features =<\/span> F.<\/span>elu(<\/span>self.<\/span>_dw_conv(<\/span>features)<\/span>)<\/span> # TODO: ELU or ReLU?<\/span>
\n features =<\/span> self.<\/span>_conv1(<\/span>features)<\/span>
\n x =<\/span> x +<\/span> features<\/p>\n

features =<\/span> F.<\/span>elu(<\/span>self.<\/span>_conv2(<\/span>x)<\/span>)<\/span>
\n features =<\/span> self.<\/span>_conv3(<\/span>features)<\/span>
\n return<\/span> x +<\/span> features<\/p>\n

class<\/span> IRA<\/span>(<\/span>nn.<\/span>Module)<\/span>:<\/span>
\n def<\/span> __init__<\/span>(<\/span>self,<\/span> inc,<\/span> ouc)<\/span>:<\/span>
\n super<\/span>(<\/span>)<\/span>.<\/span>__init__(<\/span>)<\/span><\/p>\n

self.<\/span>proj =<\/span> nn.<\/span>Conv2d(<\/span>inc,<\/span> ouc,<\/span> kernel_size=<\/span>3<\/span>,<\/span> padding=<\/span>1<\/span>)<\/span><\/p>\n

self.<\/span>irb =<\/span> nn.<\/span>Sequential(<\/span>
\n nn.<\/span>Conv2d(<\/span>ouc,<\/span> ouc,<\/span> kernel_size=<\/span>1<\/span>)<\/span>,<\/span>
\n nn.<\/span>Conv2d(<\/span>ouc,<\/span> ouc,<\/span> kernel_size=<\/span>3<\/span>,<\/span> padding=<\/span>1<\/span>,<\/span> groups=<\/span>ouc)<\/span>,<\/span>
\n nn.<\/span>ELU(<\/span>inplace=<\/span>True<\/span>)<\/span>,<\/span>
\n nn.<\/span>Conv2d(<\/span>ouc,<\/span> ouc,<\/span> kernel_size=<\/span>1<\/span>)<\/span>,<\/span>
\n )<\/span><\/p>\n

self.<\/span>ffn =<\/span> nn.<\/span>Sequential(<\/span>
\n nn.<\/span>Conv2d(<\/span>ouc,<\/span> ouc,<\/span> kernel_size=<\/span>1<\/span>)<\/span>,<\/span>
\n nn.<\/span>ELU(<\/span>inplace=<\/span>True<\/span>)<\/span>,<\/span>
\n nn.<\/span>Conv2d(<\/span>ouc,<\/span> ouc,<\/span> kernel_size=<\/span>1<\/span>)<\/span>,<\/span>
\n )<\/span><\/p>\n

self.<\/span>attn =<\/span> AttentionBlock(<\/span>ouc)<\/span><\/p>\n

def<\/span> forward<\/span>(<\/span>self,<\/span> x)<\/span>:<\/span>
\n x =<\/span> self.<\/span>proj(<\/span>x)<\/span><\/p>\n

identity =<\/span> x
\n x =<\/span> identity +<\/span> self.<\/span>irb(<\/span>x)<\/span><\/p>\n

identity =<\/span> x
\n x =<\/span> identity +<\/span> self.<\/span>ffn(<\/span>x)<\/span><\/p>\n

x =<\/span> self.<\/span>attn(<\/span>x)<\/span>
\n return<\/span> x<\/p>\n

class<\/span> C3k_IRA<\/span>(<\/span>C3k)<\/span>:<\/span>
\n def<\/span> __init__<\/span>(<\/span>self,<\/span> c1,<\/span> c2,<\/span> n=<\/span>1<\/span>,<\/span> shortcut=<\/span>False<\/span>,<\/span> g=<\/span>1<\/span>,<\/span> e=<\/span>0.5<\/span>,<\/span> k=<\/span>3<\/span>)<\/span>:<\/span>
\n super<\/span>(<\/span>)<\/span>.<\/span>__init__(<\/span>c1,<\/span> c2,<\/span> n,<\/span> shortcut,<\/span> g,<\/span> e,<\/span> k)<\/span>
\n c_ =<\/span> int<\/span>(<\/span>c2 *<\/span> e)<\/span> # hidden channels<\/span>
\n self.<\/span>m =<\/span> nn.<\/span>Sequential(<\/span>*<\/span>(<\/span>IRA(<\/span>c_,<\/span> c_)<\/span> for<\/span> _ in<\/span> range<\/span>(<\/span>n)<\/span>)<\/span>)<\/span><\/p>\n

class<\/span> C3k2_IRA<\/span>(<\/span>C3k2)<\/span>:<\/span>
\n def<\/span> __init__<\/span>(<\/span>self,<\/span> c1,<\/span> c2,<\/span> n=<\/span>1<\/span>,<\/span> c3k=<\/span>False<\/span>,<\/span> e=<\/span>0.5<\/span>,<\/span> g=<\/span>1<\/span>,<\/span> shortcut=<\/span>True<\/span>)<\/span>:<\/span>
\n super<\/span>(<\/span>)<\/span>.<\/span>__init__(<\/span>c1,<\/span> c2,<\/span> n,<\/span> c3k,<\/span> e,<\/span> g,<\/span> shortcut)<\/span>
\n self.<\/span>m =<\/span> nn.<\/span>ModuleList(<\/span>C3k_IRA(<\/span>self.<\/span>c,<\/span> self.<\/span>c,<\/span> 2<\/span>,<\/span> shortcut,<\/span> g)<\/span> if<\/span> c3k else<\/span> IRA(<\/span>self.<\/span>c,<\/span> self.<\/span>c)<\/span> for<\/span> _ in<\/span> range<\/span>(<\/span>n)<\/span>)<\/span><\/p>\n

\u6ce8\u518c<\/h2>\n

\u5728ultralytics\/nn\/tasks.py\u4e2d\u8fdb\u884c\u5982\u4e0b\u64cd\u4f5c：<\/p>\n

\u6b65\u9aa41:<\/h3>\n

from<\/span> ultralytics.<\/span>nn.<\/span>C3k2.<\/span>C3k2_IRA import<\/span> C3k2_IRA<\/p>\n

\u6b65\u9aa42<\/h3>\n

\u4fee\u6539def parse_model(d, ch, verbose=True):<\/p>\n

C3k2_IRA<\/p>\n

\"image-20260812220414836\"<\/p>\n

\u914d\u7f6eyolo11-C3k2_IRA.yaml<\/h2>\n

# Ultralytics YOLO 🚀, AGPL-3.0 license<\/span>
\n# YOLO11 object detection model with P3-P5 outputs. For Usage examples see https:\/\/docs.ultralytics.com\/tasks\/detect<\/span><\/p>\n

# Parameters<\/span>
\nnc<\/span>:<\/span> 80<\/span> # number of classes<\/span>
\nscales<\/span>:<\/span> # model compound scaling constants, i.e. 'model=yolo11n.yaml' will call yolo11.yaml with scale 'n'<\/span>
\n # [depth, width, max_channels]<\/span>
\n n<\/span>:<\/span> [<\/span>0.50<\/span>,<\/span> 0.25<\/span>,<\/span> 1024<\/span>]<\/span> # summary: 319 layers, 2624080 parameters, 2624064 gradients, 6.6 GFLOPs<\/span>
\n s<\/span>:<\/span> [<\/span>0.50<\/span>,<\/span> 0.50<\/span>,<\/span> 1024<\/span>]<\/span> # summary: 319 layers, 9458752 parameters, 9458736 gradients, 21.7 GFLOPs<\/span>
\n m<\/span>:<\/span> [<\/span>0.50<\/span>,<\/span> 1.00<\/span>,<\/span> 512<\/span>]<\/span> # summary: 409 layers, 20114688 parameters, 20114672 gradients, 68.5 GFLOPs<\/span>
\n l<\/span>:<\/span> [<\/span>1.00<\/span>,<\/span> 1.00<\/span>,<\/span> 512<\/span>]<\/span> # summary: 631 layers, 25372160 parameters, 25372144 gradients, 87.6 GFLOPs<\/span>
\n x<\/span>:<\/span> [<\/span>1.00<\/span>,<\/span> 1.50<\/span>,<\/span> 512<\/span>]<\/span> # summary: 631 layers, 56966176 parameters, 56966160 gradients, 196.0 GFLOPs<\/span><\/p>\n

# YOLO11n backbone<\/span>
\nbackbone<\/span>:<\/span>
\n # [from, repeats, module, args]<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 1<\/span>,<\/span> Conv,<\/span> [<\/span>64<\/span>,<\/span> 3<\/span>,<\/span> 2<\/span>]<\/span>]<\/span> # 0-P1\/2<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 1<\/span>,<\/span> Conv,<\/span> [<\/span>128<\/span>,<\/span> 3<\/span>,<\/span> 2<\/span>]<\/span>]<\/span> # 1-P2\/4<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 2<\/span>,<\/span> C3k2_IRA,<\/span> [<\/span>256<\/span>,<\/span> False<\/span>,<\/span> 0.25<\/span>]<\/span>]<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 1<\/span>,<\/span> Conv,<\/span> [<\/span>256<\/span>,<\/span> 3<\/span>,<\/span> 2<\/span>]<\/span>]<\/span> # 3-P3\/8<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 2<\/span>,<\/span> C3k2_IRA,<\/span> [<\/span>512<\/span>,<\/span> False<\/span>,<\/span> 0.25<\/span>]<\/span>]<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 1<\/span>,<\/span> Conv,<\/span> [<\/span>512<\/span>,<\/span> 3<\/span>,<\/span> 2<\/span>]<\/span>]<\/span> # 5-P4\/16<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 2<\/span>,<\/span> C3k2_IRA,<\/span> [<\/span>512<\/span>,<\/span> True<\/span>]<\/span>]<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 1<\/span>,<\/span> Conv,<\/span> [<\/span>1024<\/span>,<\/span> 3<\/span>,<\/span> 2<\/span>]<\/span>]<\/span> # 7-P5\/32<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 2<\/span>,<\/span> C3k2_IRA,<\/span> [<\/span>1024<\/span>,<\/span> True<\/span>]<\/span>]<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 1<\/span>,<\/span> SPPF,<\/span> [<\/span>1024<\/span>,<\/span> 5<\/span>]<\/span>]<\/span> # 9<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 2<\/span>,<\/span> C2PSA,<\/span> [<\/span>1024<\/span>]<\/span>]<\/span> # 10<\/span><\/p>\n

# YOLO11n head<\/span>
\nhead<\/span>:<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 1<\/span>,<\/span> nn.Upsample,<\/span> [<\/span>None,<\/span> 2<\/span>,<\/span> "nearest"<\/span>]<\/span>]<\/span>
\n –<\/span> [<\/span>[<\/span>-1<\/span>,<\/span> 6<\/span>]<\/span>,<\/span> 1<\/span>,<\/span> Concat,<\/span> [<\/span>1<\/span>]<\/span>]<\/span> # cat backbone P4<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 2<\/span>,<\/span> C3k2_IRA,<\/span> [<\/span>512<\/span>,<\/span> False<\/span>]<\/span>]<\/span> # 13<\/span><\/p>\n

–<\/span> [<\/span>-1<\/span>,<\/span> 1<\/span>,<\/span> nn.Upsample,<\/span> [<\/span>None,<\/span> 2<\/span>,<\/span> "nearest"<\/span>]<\/span>]<\/span>
\n –<\/span> [<\/span>[<\/span>-1<\/span>,<\/span> 4<\/span>]<\/span>,<\/span> 1<\/span>,<\/span> Concat,<\/span> [<\/span>1<\/span>]<\/span>]<\/span> # cat backbone P3<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 2<\/span>,<\/span> C3k2_IRA,<\/span> [<\/span>256<\/span>,<\/span> False<\/span>]<\/span>]<\/span> # 16 (P3\/8-small)<\/span><\/p>\n

–<\/span> [<\/span>-1<\/span>,<\/span> 1<\/span>,<\/span> Conv,<\/span> [<\/span>256<\/span>,<\/span> 3<\/span>,<\/span> 2<\/span>]<\/span>]<\/span>
\n –<\/span> [<\/span>[<\/span>-1<\/span>,<\/span> 13<\/span>]<\/span>,<\/span> 1<\/span>,<\/span> Concat,<\/span> [<\/span>1<\/span>]<\/span>]<\/span> # cat head P4<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 2<\/span>,<\/span> C3k2_IRA,<\/span> [<\/span>512<\/span>,<\/span> False<\/span>]<\/span>]<\/span> # 19 (P4\/16-medium)<\/span><\/p>\n

–<\/span> [<\/span>-1<\/span>,<\/span> 1<\/span>,<\/span> Conv,<\/span> [<\/span>512<\/span>,<\/span> 3<\/span>,<\/span> 2<\/span>]<\/span>]<\/span>
\n –<\/span> [<\/span>[<\/span>-1<\/span>,<\/span> 10<\/span>]<\/span>,<\/span> 1<\/span>,<\/span> Concat,<\/span> [<\/span>1<\/span>]<\/span>]<\/span> # cat head P5<\/span>
\n –<\/span> [<\/span>-1<\/span>,<\/span> 2<\/span>,<\/span> C3k2_IRA,<\/span> [<\/span>1024<\/span>,<\/span> True<\/span>]<\/span>]<\/span> # 22 (P5\/32-large)<\/span><\/p>\n

–<\/span> [<\/span>[<\/span>16<\/span>,<\/span> 19<\/span>,<\/span> 22<\/span>]<\/span>,<\/span> 1<\/span>,<\/span> Detect,<\/span> [<\/span>nc]<\/span>]<\/span> # Detect(P3, P4, P5)<\/span><\/p>\n

\u5b9e\u9a8c<\/h2>\n

\u811a\u672c<\/h3>\n

import<\/span> warnings
\nwarnings.<\/span>filterwarnings(<\/span>'ignore'<\/span>)<\/span>
\nfrom<\/span> ultralytics import<\/span> YOLO<\/p>\n

if<\/span> __name__ ==<\/span> '__main__'<\/span>:<\/span>
\n# \u4fee\u6539\u4e3a\u81ea\u5df1\u7684\u914d\u7f6e\u6587\u4ef6\u5730\u5740<\/span>
\n model =<\/span> YOLO(<\/span>'.\/ultralytics\/cfg\/models\/11\/yolo11-C3k2_IRA.yaml'<\/span>)<\/span>
\n# \u4fee\u6539\u4e3a\u81ea\u5df1\u7684\u6570\u636e\u96c6\u5730\u5740<\/span>
\n model.<\/span>train(<\/span>data=<\/span>'.\/ultralytics\/cfg\/datasets\/coco8.yaml'<\/span>,<\/span>
\n cache=<\/span>False<\/span>,<\/span>
\n imgsz=<\/span>640<\/span>,<\/span>
\n epochs=<\/span>10<\/span>,<\/span>
\n single_cls=<\/span>False<\/span>,<\/span> # \u662f\u5426\u662f\u5355\u7c7b\u522b\u68c0\u6d4b<\/span>
\n batch=<\/span>8<\/span>,<\/span>
\n close_mosaic=<\/span>10<\/span>,<\/span>
\n workers=<\/span>0<\/span>,<\/span>
\n optimizer=<\/span>'SGD'<\/span>,<\/span>
\n amp=<\/span>True<\/span>,<\/span>
\n project=<\/span>'runs\/train'<\/span>,<\/span>
\n name=<\/span>'C3k2_IRA'<\/span>,<\/span>
\n )<\/span><\/p>\n

\u7ed3\u679c<\/h3>\n

\"image-20260812220348321\"<\/p>\n","protected":false},"excerpt":{"rendered":"

\u524d\u8a00
\n\u672c\u6587\u4ecb\u7ecd\u4e86\u7528\u4e8e\u8f7b\u91cf\u7aef\u5230\u7aef\u989c\u8272\u6821\u6b63\u7684\u5012\u6b8b\u5dee\u6ce8\u610f\u529b\u6a21\u5757 IRA，\u878d\u5408\u4e86 MobileNet \u98ce\u683c\u5012\u6b8b\u5dee\u7ed3\u6784\u4e0e\u901a\u9053\u3001\u7a7a\u95f4\u6ce8\u610f\u529b\u673a\u5236\u3002\u8be5\u6a21\u5757\u901a\u8fc7\u6df1\u5ea6\u53ef\u5206\u79bb\u5377\u79ef\u9ad8\u6548\u63d0\u53d6\u5c40\u90e8\u7279\u5f81，\u5e76\u5229\u7528\u6ce8\u610f\u529b\u81ea\u9002\u5e94\u5f3a\u5316\u4e0e\u989c\u8272\u3001\u7eb9\u7406\u3001\u8fb9\u7f18\u76f8\u5173\u7684\u5173\u952e\u4fe1\u606f，\u5728\u8f83\u4f4e\u8ba1\u7b97\u6210\u672c\u4e0b\u63d0\u5347\u7279\u5f81\u9009\u62e9\u80fd\u529b\u4e0e\u8868\u8fbe\u7a33\u5b9a\u6027\u3002\u6211\u4eec\u5c06 IRA \u6210\u529f\u96c6\u6210\u8fdb YOLO11 \u7684 C3k2 \u6a21\u5757，\u6784\u5efa C3k2_IRA，\u7528\u4e8e\u589e\u5f3a\u4e3b\u5e72\u4e0e\u9888\u90e8\u7f51\u7edc\u7684\u5c40\u90e8\u7ec6\u8282\u5efa\u6a21\u548c\u6709\u6548\u7279\u5f81<\/p>\n","protected":false},"author":2,"featured_media":94837,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[156,50,78,523],"topic":[],"class_list":["post-94840","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-yolo","tag-50","tag-78","tag-523"],"yoast_head":"\nYOLO11\u6539\u8fdb - C3k2\u878d\u5408 | IRA\u5012\u6b8b\u5dee\u6ce8\u610f\u529b\u6a21\u5757\uff0c\u8f7b\u91cf\u878d\u5408\u5377\u79ef\u8868\u5f81\u4e0e\u901a\u9053\u7a7a\u95f4\u6ce8\u610f\uff0c\u52a9\u529b\u79fb\u52a8\u7aef\u591a\u5149\u8c31\u989c\u8272\u6821\u6b63\u8bef\u5dee\u5927\u5e45\u964d\u4f4e | CVPR 2026 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/94840.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"YOLO11\u6539\u8fdb - C3k2\u878d\u5408 | IRA\u5012\u6b8b\u5dee\u6ce8\u610f\u529b\u6a21\u5757\uff0c\u8f7b\u91cf\u878d\u5408\u5377\u79ef\u8868\u5f81\u4e0e\u901a\u9053\u7a7a\u95f4\u6ce8\u610f\uff0c\u52a9\u529b\u79fb\u52a8\u7aef\u591a\u5149\u8c31\u989c\u8272\u6821\u6b63\u8bef\u5dee\u5927\u5e45\u964d\u4f4e | CVPR 2026 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u524d\u8a00 \u672c\u6587\u4ecb\u7ecd\u4e86\u7528\u4e8e\u8f7b\u91cf\u7aef\u5230\u7aef\u989c\u8272\u6821\u6b63\u7684\u5012\u6b8b\u5dee\u6ce8\u610f\u529b\u6a21\u5757 IRA，\u878d\u5408\u4e86 MobileNet \u98ce\u683c\u5012\u6b8b\u5dee\u7ed3\u6784\u4e0e\u901a\u9053\u3001\u7a7a\u95f4\u6ce8\u610f\u529b\u673a\u5236\u3002\u8be5\u6a21\u5757\u901a\u8fc7\u6df1\u5ea6\u53ef\u5206\u79bb\u5377\u79ef\u9ad8\u6548\u63d0\u53d6\u5c40\u90e8\u7279\u5f81，\u5e76\u5229\u7528\u6ce8\u610f\u529b\u81ea\u9002\u5e94\u5f3a\u5316\u4e0e\u989c\u8272\u3001\u7eb9\u7406\u3001\u8fb9\u7f18\u76f8\u5173\u7684\u5173\u952e\u4fe1\u606f，\u5728\u8f83\u4f4e\u8ba1\u7b97\u6210\u672c\u4e0b\u63d0\u5347\u7279\u5f81\u9009\u62e9\u80fd\u529b\u4e0e\u8868\u8fbe\u7a33\u5b9a\u6027\u3002\u6211\u4eec\u5c06 IRA \u6210\u529f\u96c6\u6210\u8fdb YOLO11 \u7684 C3k2 \u6a21\u5757，\u6784\u5efa C3k2_IRA，\u7528\u4e8e\u589e\u5f3a\u4e3b\u5e72\u4e0e\u9888\u90e8\u7f51\u7edc\u7684\u5c40\u90e8\u7ec6\u8282\u5efa\u6a21\u548c\u6709\u6548\u7279\u5f81\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/94840.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-08-15T04:32:56+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260815043255-6a7febf7113ec.png\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"6 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/94840.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/94840.html\",\"name\":\"YOLO11\u6539\u8fdb - C3k2\u878d\u5408 | IRA\u5012\u6b8b\u5dee\u6ce8\u610f\u529b\u6a21\u5757\uff0c\u8f7b\u91cf\u878d\u5408\u5377\u79ef\u8868\u5f81\u4e0e\u901a\u9053\u7a7a\u95f4\u6ce8\u610f\uff0c\u52a9\u529b\u79fb\u52a8\u7aef\u591a\u5149\u8c31\u989c\u8272\u6821\u6b63\u8bef\u5dee\u5927\u5e45\u964d\u4f4e | CVPR 2026 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-08-15T04:32:56+00:00\",\"dateModified\":\"2026-08-15T04:32:56+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/94840.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/94840.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/94840.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"YOLO11\u6539\u8fdb - C3k2\u878d\u5408 | IRA\u5012\u6b8b\u5dee\u6ce8\u610f\u529b\u6a21\u5757\uff0c\u8f7b\u91cf\u878d\u5408\u5377\u79ef\u8868\u5f81\u4e0e\u901a\u9053\u7a7a\u95f4\u6ce8\u610f\uff0c\u52a9\u529b\u79fb\u52a8\u7aef\u591a\u5149\u8c31\u989c\u8272\u6821\u6b63\u8bef\u5dee\u5927\u5e45\u964d\u4f4e | CVPR 2026\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"YOLO11\u6539\u8fdb - C3k2\u878d\u5408 | IRA\u5012\u6b8b\u5dee\u6ce8\u610f\u529b\u6a21\u5757\uff0c\u8f7b\u91cf\u878d\u5408\u5377\u79ef\u8868\u5f81\u4e0e\u901a\u9053\u7a7a\u95f4\u6ce8\u610f\uff0c\u52a9\u529b\u79fb\u52a8\u7aef\u591a\u5149\u8c31\u989c\u8272\u6821\u6b63\u8bef\u5dee\u5927\u5e45\u964d\u4f4e | CVPR 2026 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/94840.html","og_locale":"zh_CN","og_type":"article","og_title":"YOLO11\u6539\u8fdb - C3k2\u878d\u5408 | IRA\u5012\u6b8b\u5dee\u6ce8\u610f\u529b\u6a21\u5757\uff0c\u8f7b\u91cf\u878d\u5408\u5377\u79ef\u8868\u5f81\u4e0e\u901a\u9053\u7a7a\u95f4\u6ce8\u610f\uff0c\u52a9\u529b\u79fb\u52a8\u7aef\u591a\u5149\u8c31\u989c\u8272\u6821\u6b63\u8bef\u5dee\u5927\u5e45\u964d\u4f4e | CVPR 2026 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u524d\u8a00 \u672c\u6587\u4ecb\u7ecd\u4e86\u7528\u4e8e\u8f7b\u91cf\u7aef\u5230\u7aef\u989c\u8272\u6821\u6b63\u7684\u5012\u6b8b\u5dee\u6ce8\u610f\u529b\u6a21\u5757 IRA，\u878d\u5408\u4e86 MobileNet \u98ce\u683c\u5012\u6b8b\u5dee\u7ed3\u6784\u4e0e\u901a\u9053\u3001\u7a7a\u95f4\u6ce8\u610f\u529b\u673a\u5236\u3002\u8be5\u6a21\u5757\u901a\u8fc7\u6df1\u5ea6\u53ef\u5206\u79bb\u5377\u79ef\u9ad8\u6548\u63d0\u53d6\u5c40\u90e8\u7279\u5f81，\u5e76\u5229\u7528\u6ce8\u610f\u529b\u81ea\u9002\u5e94\u5f3a\u5316\u4e0e\u989c\u8272\u3001\u7eb9\u7406\u3001\u8fb9\u7f18\u76f8\u5173\u7684\u5173\u952e\u4fe1\u606f，\u5728\u8f83\u4f4e\u8ba1\u7b97\u6210\u672c\u4e0b\u63d0\u5347\u7279\u5f81\u9009\u62e9\u80fd\u529b\u4e0e\u8868\u8fbe\u7a33\u5b9a\u6027\u3002\u6211\u4eec\u5c06 IRA \u6210\u529f\u96c6\u6210\u8fdb YOLO11 \u7684 C3k2 \u6a21\u5757，\u6784\u5efa C3k2_IRA，\u7528\u4e8e\u589e\u5f3a\u4e3b\u5e72\u4e0e\u9888\u90e8\u7f51\u7edc\u7684\u5c40\u90e8\u7ec6\u8282\u5efa\u6a21\u548c\u6709\u6548\u7279\u5f81","og_url":"https:\/\/www.wsisp.com\/helps\/94840.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-08-15T04:32:56+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/08\/20260815043255-6a7febf7113ec.png"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"6 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/94840.html","url":"https:\/\/www.wsisp.com\/helps\/94840.html","name":"YOLO11\u6539\u8fdb - C3k2\u878d\u5408 | IRA\u5012\u6b8b\u5dee\u6ce8\u610f\u529b\u6a21\u5757\uff0c\u8f7b\u91cf\u878d\u5408\u5377\u79ef\u8868\u5f81\u4e0e\u901a\u9053\u7a7a\u95f4\u6ce8\u610f\uff0c\u52a9\u529b\u79fb\u52a8\u7aef\u591a\u5149\u8c31\u989c\u8272\u6821\u6b63\u8bef\u5dee\u5927\u5e45\u964d\u4f4e | CVPR 2026 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-08-15T04:32:56+00:00","dateModified":"2026-08-15T04:32:56+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/94840.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/94840.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/94840.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"YOLO11\u6539\u8fdb - C3k2\u878d\u5408 | IRA\u5012\u6b8b\u5dee\u6ce8\u610f\u529b\u6a21\u5757\uff0c\u8f7b\u91cf\u878d\u5408\u5377\u79ef\u8868\u5f81\u4e0e\u901a\u9053\u7a7a\u95f4\u6ce8\u610f\uff0c\u52a9\u529b\u79fb\u52a8\u7aef\u591a\u5149\u8c31\u989c\u8272\u6821\u6b63\u8bef\u5dee\u5927\u5e45\u964d\u4f4e | CVPR 2026"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/94840","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=94840"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/94840\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/94837"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=94840"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=94840"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=94840"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=94840"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}