{"id":2577,"date":"2019-05-21T18:18:34","date_gmt":"2019-05-21T10:18:34","guid":{"rendered":"https:\/\/yanjingang.com\/blog\/?p=2577"},"modified":"2019-08-07T17:56:07","modified_gmt":"2019-08-07T09:56:07","slug":"darknet-%e4%b8%80%e4%b8%aa%e5%9f%ba%e4%ba%8ec%e7%9a%84%e6%b7%b1%e5%ba%a6%e5%ad%a6%e4%b9%a0%e6%a1%86%e6%9e%b6","status":"publish","type":"post","link":"https:\/\/yanjingang.com\/blog\/?p=2577","title":{"rendered":"\u5c0f\u732a\u5b66Darknet\u2014\u57fa\u4e8eC\u7684\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6"},"content":{"rendered":"

\u6700\u8fd1\u5728\u7814\u7a76\u89c6\u9891\u4e8c\u7ef4\u7801\u5b9a\u4f4d\u6e05\u7406\uff0c\u6d89\u53ca\u5230Object Detection\u5bf9\u8c61\u68c0\u6d4b\u6280\u672f\uff0c\u53d1\u73b0Darknet\u7528\u8d77\u6765\u6548\u679c\u4e0d\u9519\uff0c\u672c\u6587\u7b80\u5355\u4ecb\u7ecd\u4e0b\u3002\u5148\u8d34\u51e0\u4e2a\u793a\u4f8b\uff1a<\/p>\n

\u513f\u5b50\u6bd4\u8d5b\u89c6\u9891\u4e2d\u7684\u666e\u901a\u76ee\u6807\u8bc6\u522b\uff1a<\/p>\n

\"\"<\/a><\/p>\n

\u81ea\u52a8\u68c0\u6d4b\u6e05\u7406\u89c6\u9891\u4e2d\u7684\u4e8c\u7ef4\u7801\uff1a<\/p>\n

\"\"<\/a>\"\"<\/a><\/p>\n

 <\/p>\n

\u6982\u8ff0<\/strong><\/h1>\n

Darknet\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\u662f\u7531Joseph Redmon\u63d0\u51fa\u7684\u4e00\u4e2a\u7528C\u548cCUDA\u7f16\u5199\u7684\u5f00\u6e90\u795e\u7ecf\u7f51\u7edc\u6846\u67b6\u3002\u5b83\u5b89\u88c5\u901f\u5ea6\u5feb\uff0c\u6613\u4e8e\u5b89\u88c5\uff0c\u5e76\u652f\u6301CPU\u548cGPU\u8ba1\u7b97\u3002<\/p>\n

GitHub\uff1ahttps:\/\/github.com\/pjreddie\/darknet<\/a><\/p>\n

\u5b98\u7f51\uff1ahttps:\/\/pjreddie.com\/darknet\/<\/a><\/p>\n

\u5982\u4f55\u5b89\u88c5\uff1ahttps:\/\/pjreddie.com\/darknet\/install\/<\/a><\/p>\n

\u5b9e\u65f6\u5bf9\u8c61\u68c0\u6d4b&\u5982\u4f55\u8bad\u7ec3\u81ea\u5df1\u7684\u7279\u6b8a\u6a21\u578b\uff1ahttps:\/\/pjreddie.com\/darknet\/yolo\/<\/a><\/p>\n

\u4f18\u70b9<\/strong><\/p>\n

Darknet\u662f\u4e00\u4e2a\u6bd4\u8f83\u5c0f\u4f17\u7684\u6df1\u5ea6\u5b66\u4e60\u6846\u67b6\uff0c\u6ca1\u6709\u793e\u533a\uff0c\u4e3b\u8981\u9760\u4f5c\u8005\u56e2\u961f\u7ef4\u62a4\uff0c\u6240\u4ee5\u63a8\u5e7f\u8f83\u5f31\uff0c\u7528\u7684\u4eba\u4e0d\u591a\u3002\u800c\u4e14\u7531\u4e8e\u7ef4\u62a4\u4eba\u5458\u6709\u9650\uff0c\u529f\u80fd\u4e5f\u4e0d\u5982tensorflow\u7b49\u6846\u67b6\u90a3\u4e48\u5f3a\u5927\uff0c\u4f46\u662f\u8be5\u6846\u67b6\u8fd8\u662f\u6709\u4e00\u4e9b\u72ec\u6709\u7684\u4f18\u70b9\uff1a
\n1.\u6613\u4e8e\u5b89\u88c5<\/span>
\n\u5728makefile\u91cc\u9762\u9009\u62e9\u81ea\u5df1\u9700\u8981\u7684\u9644\u52a0\u9879\uff08cuda\uff0ccudnn\uff0copencv\u7b49\uff09\u76f4\u63a5make\u5373\u53ef\uff0c\u51e0\u5206\u949f\u5b8c\u6210\u5b89\u88c5\uff1b
\n2.\u6ca1\u6709\u4efb\u4f55\u4f9d\u8d56\u9879<\/span>
\n\u6574\u4e2a\u6846\u67b6\u90fd\u7528C\u8bed\u8a00\u8fdb\u884c\u7f16\u5199\uff0c\u53ef\u4ee5\u4e0d\u4f9d\u8d56\u4efb\u4f55\u5e93\uff0c\u8fdeopencv\u4f5c\u8005\u90fd\u7f16\u5199\u4e86\u53ef\u4ee5\u5bf9\u5176\u8fdb\u884c\u66ff\u4ee3\u7684\u51fd\u6570\uff1b
\n3.\u8fd0\u884c\u901f\u5ea6\u5feb<\/span>
\nP4 gpu\u5361\u4e0a\u5b9e\u6d4b\u7ea635 fps\uff08\u9ad8\u6e05\u89c6\u9891\u5207\u6862\u56fe\u7247\uff09
\n4.\u7ed3\u6784\u660e\u6670\uff0c\u6e90\u4ee3\u7801\u67e5\u770b\u3001\u4fee\u6539\u65b9\u4fbf\uff0c\u6613\u4e8e\u79fb\u690d<\/span>
\n\u5176\u6846\u67b6\u7684\u57fa\u7840\u6587\u4ef6\u90fd\u5728src\u6587\u4ef6\u5939\uff0c\u800c\u5b9a\u4e49\u7684\u4e00\u4e9b\u68c0\u6d4b\u3001\u5206\u7c7b\u51fd\u6570\u5219\u5728example\u6587\u4ef6\u5939\uff0c\u53ef\u6839\u636e\u9700\u8981\u76f4\u63a5\u5bf9\u6e90\u4ee3\u7801\u8fdb\u884c\u67e5\u770b\u548c\u4fee\u6539\uff1b
\n\u8be5\u6846\u67b6\u90e8\u7f72\u5230\u673a\u5668\u672c\u5730\u5341\u5206\u7b80\u5355\uff0c\u4e14\u53ef\u4ee5\u6839\u636e\u673a\u5668\u60c5\u51b5\uff0c\u4f7f\u7528cpu\u548cgpu\uff0c\u7279\u522b\u662f\u68c0\u6d4b\u8bc6\u522b\u4efb\u52a1\u7684\u672c\u5730\u7aef\u90e8\u7f72\uff0cdarknet\u4f1a\u663e\u5f97\u5f02\u5e38\u65b9\u4fbf\u3002
\n5.\u53cb\u597dpython\u63a5\u53e3<\/span>
\n\u867d\u7136darknet\u4f7f\u7528c\u8bed\u8a00\u8fdb\u884c\u7f16\u5199\uff0c\u4f46\u662f\u4e5f\u63d0\u4f9b\u4e86python\u7684\u63a5\u53e3\uff0c\u901a\u8fc7python\u51fd\u6570\uff0c\u80fd\u591f\u4f7f\u7528python\u76f4\u63a5\u5bf9\u8bad\u7ec3\u597d\u7684.weight\u683c\u5f0f\u7684\u6a21\u578b\u8fdb\u884c\u8c03\u7528\uff1b<\/p>\n

 <\/p>\n

YOLO\u7b97\u6cd5<\/strong><\/p>\n

YOLO(You Only Look Once)\u662fJoseph Redmon\u9488\u5bf9\u8fd9\u4e00\u6846\u67b6\u63d0\u51fa\u7684\u6838\u5fc3\u76ee\u6807\u68c0\u6d4b\u7b97\u6cd5\u3002<\/p>\n

\"\"<\/a><\/p>\n

\u4f5c\u8005\u5728YOLO\u7b97\u6cd5\u4e2d\u628a\u7269\u4f53\u68c0\u6d4b\u95ee\u9898\u5904\u7406\u6210\u56de\u5f52\u95ee\u9898\uff0c\u7528\u4e00\u4e2a\u5377\u79ef\u795e\u7ecf\u7f51\u7edc\u7ed3\u6784\u5c31\u53ef\u4ee5\u4ece\u8f93\u5165\u56fe\u50cf\u76f4\u63a5\u9884\u6d4bbounding box\u548c\u7c7b\u522b\u6982\u7387\u3002<\/p>\n

\"\"<\/a><\/p>\n

YOLO\u7b97\u6cd5\u7684\u4f18\u70b9<\/strong>\uff1a<\/p>\n

1\u3001YOLO\u7684\u901f\u5ea6\u975e\u5e38\u5feb\u3002\u5728Titan X GPU\u4e0a\u7684\u901f\u5ea6\u662f45 fps\uff08frames per second\uff09\uff0c\u52a0\u901f\u7248\u7684YOLO\u5dee\u4e0d\u591a\u662f150fps\u3002<\/p>\n

2\u3001YOLO\u662f\u57fa\u4e8e\u56fe\u50cf\u7684\u5168\u5c40\u4fe1\u606f\u8fdb\u884c\u9884\u6d4b\u7684\u3002\u8fd9\u4e00\u70b9\u548c\u57fa\u4e8esliding window\u4ee5\u53caregion proposal\u7b49\u68c0\u6d4b\u7b97\u6cd5\u4e0d\u4e00\u6837\u3002\u4e0eFast R-CNN\u76f8\u6bd4\uff0cYOLO\u5728\u8bef\u68c0\u6d4b\uff08\u5c06\u80cc\u666f\u68c0\u6d4b\u4e3a\u7269\u4f53\uff09\u65b9\u9762\u7684\u9519\u8bef\u7387\u80fd\u964d\u4f4e\u4e00\u534a\u591a\u3002<\/p>\n

3\u3001\u53ef\u4ee5\u5b66\u5230\u7269\u4f53\u7684generalizable-representations\u3002\u53ef\u4ee5\u7406\u89e3\u4e3a\u6cdb\u5316\u80fd\u529b\u5f3a\u3002<\/p>\n

4\u3001\u51c6\u786e\u7387\u9ad8\u3002\u6709\u5b9e\u9a8c\u8bc1\u660e\u3002<\/p>\n

\u4e8b\u5b9e\u4e0a\uff0c\u76ee\u6807\u68c0\u6d4b\u7684\u672c\u8d28\u5c31\u662f\u56de\u5f52\uff0c\u56e0\u6b64\u4e00\u4e2a\u5b9e\u73b0\u56de\u5f52\u529f\u80fd\u7684CNN\u5e76\u4e0d\u9700\u8981\u590d\u6742\u7684\u8bbe\u8ba1\u8fc7\u7a0b\u3002YOLO\u6ca1\u6709\u9009\u62e9\u6ed1\u7a97\u6216\u63d0\u53d6proposal\u7684\u65b9\u5f0f\u8bad\u7ec3\u7f51\u7edc\uff0c\u800c\u662f\u76f4\u63a5\u9009\u7528\u6574\u56fe\u8bad\u7ec3\u6a21\u578b\u3002\u8fd9\u6837\u505a\u7684\u597d\u5904\u5728\u4e8e\u53ef\u4ee5\u66f4\u597d\u7684\u533a\u5206\u76ee\u6807\u548c\u80cc\u666f\u533a\u57df\uff0c\u76f8\u6bd4\u4e4b\u4e0b\uff0c\u91c7\u7528proposal\u8bad\u7ec3\u65b9\u5f0f\u7684Fast-R-CNN\u5e38\u5e38\u628a\u80cc\u666f\u533a\u57df\u8bef\u68c0\u4e3a\u7279\u5b9a\u76ee\u6807\u3002\u5f53\u7136,YOLO\u5728\u63d0\u5347\u68c0\u6d4b\u901f\u5ea6\u7684\u540c\u65f6\u727a\u7272\u4e86\u4e00\u4e9b\u7cbe\u5ea6\u3002<\/p>\n

\n
\"\"<\/a><\/div>\n<\/figure>\n

YOLO\u7684\u8bbe\u8ba1\u7406\u5ff5\u9075\u5faa\u7aef\u5230\u7aef\u8bad\u7ec3\u548c\u5b9e\u65f6\u68c0\u6d4b\u3002YOLO\u5c06\u8f93\u5165\u56fe\u50cf\u5212\u5206\u4e3aS*S\u4e2a\u7f51\u683c\uff0c\u5982\u679c\u4e00\u4e2a\u7269\u4f53\u7684\u4e2d\u5fc3\u843d\u5728\u67d0\u7f51\u683c(cell)\u5185\uff0c\u5219\u76f8\u5e94\u7f51\u683c\u8d1f\u8d23\u68c0\u6d4b\u8be5\u7269\u4f53\u3002<\/p>\n

\n
\"\"<\/a><\/div>\n<\/figure>\n

\u5728\u8bad\u7ec3\u548c\u6d4b\u8bd5\u65f6\uff0c\u6bcf\u4e2a\u7f51\u7edc\u9884\u6d4bB\u4e2abounding boxes\uff0c\u6bcf\u4e2abounding box\u5bf9\u5e945\u4e2a\u9884\u6d4b\u53c2\u6570\uff0c\u5373bounding box\u7684\u4e2d\u5fc3\u70b9\u5750\u6807(x,y)\uff0c\u5bbd\u9ad8\uff08w,h\uff09\uff0c\u548c\u7f6e\u4fe1\u5ea6\u8bc4\u5206\u3002<\/p>\n

\u8fd9\u91cc\u7684\u7f6e\u4fe1\u5ea6\u8bc4\u5206(Pr(Object)*IOU(predtruth))\u7efc\u5408\u53cd\u6620\u57fa\u4e8e\u5f53\u524d\u6a21\u578bbounding box\u5185\u5b58\u5728\u76ee\u6807\u7684\u53ef\u80fd\u6027Pr(Object)\u548cbounding box\u9884\u6d4b\u76ee\u6807\u4f4d\u7f6e\u7684\u51c6\u786e\u6027IOU(predtruth)\u3002\u5982\u679cbouding box\u5185\u4e0d\u5b58\u5728\u7269\u4f53\uff0c\u5219Pr(Object)=0\u3002\u5982\u679c\u5b58\u5728\u7269\u4f53\uff0c\u5219\u6839\u636e\u9884\u6d4b\u7684bounding box\u548c\u771f\u5b9e\u7684bounding box\u8ba1\u7b97IOU\uff0c\u540c\u65f6\u4f1a\u9884\u6d4b\u5b58\u5728\u7269\u4f53\u7684\u60c5\u51b5\u4e0b\u8be5\u7269\u4f53\u5c5e\u4e8e\u67d0\u4e00\u7c7b\u7684\u540e\u9a8c\u6982\u7387Pr(Class_iObject)\u3002<\/p>\n

\u5047\u5b9a\u4e00\u5171\u6709C\u7c7b\u7269\u4f53\uff0c\u90a3\u4e48\u6bcf\u4e00\u4e2a\u7f51\u683c\u53ea\u9884\u6d4b\u4e00\u6b21C\u7c7b\u7269\u4f53\u7684\u6761\u4ef6\u7c7b\u6982\u7387Pr(Class_iObject), i=1,2,…,C;\u6bcf\u4e00\u4e2a\u7f51\u683c\u9884\u6d4bB\u4e2abounding box\u7684\u4f4d\u7f6e\u3002\u5373\u8fd9B\u4e2abounding box\u5171\u4eab\u4e00\u5957\u6761\u4ef6\u7c7b\u6982\u7387Pr(Class_iObject), i=1,2,…,C\u3002\u57fa\u4e8e\u8ba1\u7b97\u5f97\u5230\u7684Pr(Class_iObject)\uff0c\u5728\u6d4b\u8bd5\u65f6\u53ef\u4ee5\u8ba1\u7b97\u67d0\u4e2abounding box\u7c7b\u76f8\u5173\u7f6e\u4fe1\u5ea6\uff1aPr(Class_iObject)*Pr(Object)*IOU(predtruth)=Pr(Class_i)*IOU(predtruth)\u3002<\/p>\n

\u5982\u679c\u5c06\u8f93\u5165\u56fe\u50cf\u5212\u5206\u4e3a7*7\u7f51\u683c\uff08S=7\uff09\uff0c\u6bcf\u4e2a\u7f51\u683c\u9884\u6d4b2\u4e2abounding box (B=2)\uff0c\u670920\u7c7b\u5f85\u68c0\u6d4b\u7684\u76ee\u6807\uff08C=20\uff09\uff0c\u5219\u76f8\u5f53\u4e8e\u6700\u7ec8\u9884\u6d4b\u4e00\u4e2a\u957f\u5ea6\u4e3aS*S*(B*5+C)=7*7*30\u7684\u5411\u91cf\uff0c\u4ece\u800c\u5b8c\u6210\u68c0\u6d4b+\u8bc6\u522b\u4efb\u52a1\uff0c\u6574\u4e2a\u6d41\u7a0b\u53ef\u4ee5\u901a\u8fc7\u4e0b\u56fe\u7406\u89e3\u3002<\/p>\n

\n
\"\"<\/a><\/div>\n<\/figure>\n

YOLO\u7f51\u7edc\u8bbe\u8ba1\u9075\u5faa\u4e86GoogleNet\u7684\u601d\u60f3\uff0c\u4f46\u4e0e\u4e4b\u6709\u6240\u533a\u522b\u3002YOLO\u4f7f\u7528\u4e8624\u4e2a\u7ea7\u8054\u7684\u5377\u79ef\uff08conv\uff09\u5c42\u548c2\u4e2a\u5168\u8fde\u63a5\uff08fc\uff09\u5c42\uff0c\u5176\u4e2dconv\u5c42\u5305\u62ec3*3\u548c1*1\u4e24\u79cdKernel\uff0c\u6700\u540e\u4e00\u4e2afc\u5c42\u5373YOLO\u7f51\u7edc\u7684\u8f93\u51fa\uff0c\u957f\u5ea6\u4e3aS*S*(B*5+C)=7*7*30.\u6b64\u5916\uff0c\u4f5c\u8005\u8fd8\u8bbe\u8ba1\u4e86\u4e00\u4e2a\u7b80\u5316\u7248\u7684YOLO-small\u7f51\u7edc\uff0c\u5305\u62ec9\u4e2a\u7ea7\u8054\u7684conv\u5c42\u548c2\u4e2afc\u5c42\uff0c\u7531\u4e8econv\u5c42\u7684\u6570\u91cf\u5c11\u4e86\u5f88\u591a\uff0c\u56e0\u6b64YOLO-small\u901f\u5ea6\u6bd4YOLO\u5feb\u5f88\u591a\u3002\u5982\u4e0b\u56fe\u6240\u793a\u7ed9\u51fa\u4e86YOLO\u7f51\u7edc\u7684\u67b6\u6784\u3002<\/p>\n

\n
\"\"<\/a><\/div>\n<\/figure>\n

YOLOv3\u4e0e\u5176\u4ed6\u7b97\u6cd5\u7684\u6027\u80fd\u5bf9\u6bd4\uff1a<\/p>\n

\"\"<\/a><\/p>\n

 <\/p>\n

YOLO\u7b97\u6cd5\u7684\u7f3a\u70b9<\/strong>\uff1a<\/p>\n

1\u3001\u4f4d\u7f6e\u7cbe\u786e\u6027\u5dee\uff0c\u5bf9\u4e8e\u5c0f\u76ee\u6807\u7269\u4f53\u4ee5\u53ca\u7269\u4f53\u6bd4\u8f83\u5bc6\u96c6\u7684\u573a\u666f\u68c0\u6d4b\u6548\u679c\u4e0d\u597d<\/span>\uff0c\u6bd4\u5982\u4e00\u7fa4\u5c0f\u9e1f\u3002<\/p>\n

2\u3001YOLO\u867d\u7136\u53ef\u4ee5\u964d\u4f4e\u5c06\u80cc\u666f\u68c0\u6d4b\u4e3a\u7269\u4f53\u7684\u6982\u7387\uff0c\u4f46\u540c\u65f6\u5bfc\u81f4\u53ec\u56de\u7387\u8f83\u4f4e\u3002<\/p>\n

\n
\"\"<\/a><\/div>\n<\/figure>\n

 <\/p>\n

YOLO\u7684\u8bba\u6587\uff1ahttps:\/\/www.cv-foundation.org\/openaccess\/content_cvpr_2016\/papers\/Redmon_You_Only_Look_CVPR_2016_paper.pdf<\/p>\n

 <\/p>\n

\u5982\u4f55\u4f7f\u7528<\/h1>\n

\u4ee3\u7801\u7ed3\u6784<\/strong>
\n\u4e0b\u56fe\u662fdarknet\u6e90\u4ee3\u7801\u6587\u4ef6\u5939\u7684\u5206\u5e03\u60c5\u51b5\uff1a<\/p>\n

\"\"<\/a><\/p>\n

1.cfg<\/span>\u6587\u4ef6\u5939\u5185\u662f\u4e00\u4e9b\u6a21\u578b\u7684\u67b6\u6784\uff0c\u6bcf\u4e2acfg\u6587\u4ef6\u7c7b\u4f3c\u4e0ecaffe\u7684prototxt\u6587\u4ef6\uff0c\u901a\u8fc7\u8be5\u6587\u4ef6\u5b9a\u4e49\u7684\u6574\u4e2a\u6a21\u578b\u7684\u67b6\u6784<\/span>
\n2.data\u6587\u4ef6\u5939\u5185\u653e\u7f6e\u4e86\u4e00\u4e9blabel\u6587\u4ef6\uff0c\u5982coco9k\u7684\u7c7b\u522b\u540d\u7b49\uff0c\u548c\u4e00\u4e9b\u6837\u4f8b\u56fe\uff08\u8be5\u6587\u4ef6\u5939\u4e3b\u8981\u4e3a\u6f14\u793a\u7528\uff0c\u6216\u8005\u662f\u76f4\u63a5\u8bad\u7ec3coco\u7b49\u5bf9\u5e94\u6570\u636e\u96c6\u65f6\u6709\u7528\uff0c\u5982\u679c\u8981\u7528\u81ea\u5df1\u7684\u6570\u636e\u81ea\u884c\u8bad\u7ec3\uff0c\u8be5\u6587\u4ef6\u5939\u5185\u7684\u4e1c\u897f\u90fd\u4e0d\u662f\u6211\u4eec\u9700\u8981\u7684\uff09
\n3.src<\/span>\u6587\u4ef6\u5939\u5185\u5168\u662f\u6700\u5e95\u5c42\u7684\u6846\u67b6\u5b9a\u4e49\u6587\u4ef6\uff0c\u6240\u6709\u5c42\u7684\u5b9a\u4e49\u7b49\u6700\u57fa\u672c\u7684\u51fd\u6570\u5168\u90e8\u5728\u8be5\u6587\u4ef6\u5939\u5185\uff0c\u53ef\u4ee5\u7406\u89e3\u4e3a\u8be5\u6587\u4ef6\u5939\u5c31\u662f\u6846\u67b6\u7684\u6e90\u7801\uff1b
\n4.examples<\/span>\u6587\u4ef6\u5939\u662f\u66f4\u4e3a\u9ad8\u5c42\u7684\u4e00\u4e9b\u51fd\u6570\uff0c\u5982\u68c0\u6d4b\u51fd\u6570\uff0c\u8bc6\u522b\u51fd\u6570<\/span>\u7b49\uff0c\u8fd9\u4e9b\u51fd\u6570\u76f4\u63a5\u8c03\u7528\u4e86\u5e95\u5c42\u7684\u51fd\u6570\uff0c\u6211\u4eec\u7ecf\u5e38\u4f7f\u7528\u7684\u5c31\u662fexample\u4e2d\u7684\u51fd\u6570\uff1b
\n5.include\u6587\u4ef6\u5939\uff0c\u987e\u540d\u601d\u4e49\uff0c\u5b58\u653e\u5934\u6587\u4ef6\u7684\u5730\u65b9\uff1b
\n6.python\u6587\u4ef6\u5939\u91cc\u662f\u4f7f\u7528python\u5bf9\u6a21\u578b\u7684\u8c03\u7528\u65b9\u6cd5\uff0c\u57fa\u672c\u90fd\u5728darknet.py<\/span>\u4e2d\u3002\u5f53\u7136\uff0c\u8981\u5b9e\u73b0python\u7684\u8c03\u7528\uff0c\u8fd8\u9700\u8981\u7528\u5230darknet\u7684\u52a8\u6001\u5e93libdarknet.so<\/span>\uff0c\u8fd9\u4e2a\u52a8\u6001\u5e93\u7a0d\u540e\u518d\u4ecb\u7ecd\uff1b
\n7.scripts\u6587\u4ef6\u5939\u4e2d\u662f\u4e00\u4e9b\u811a\u672c\uff0c\u5982\u4e0b\u8f7dcoco\u6570\u636e\u96c6\uff0c\u5c06voc\u683c\u5f0f\u7684\u6570\u636e\u96c6\u8f6c\u6362\u4e3a\u8bad\u7ec3\u6240\u9700\u683c\u5f0f\u7684\u811a\u672c\u7b49
\n8.Makefile\u6587\u4ef6\uff0c\u6587\u4ef6\u5f00\u5934\u6709\u4e00\u4e9b\u9009\u9879\uff0c\u628a\u4f60\u9700\u8981\u4f7f\u7528\u7684\u9009\u9879\u8bbe\u4e3a1\u5373\u53ef<\/p>\n

\u5b89\u88c5<\/strong>
\n<\/code><\/p>\n

#Mac\u5b89\u88c5opencv\r\nbrew install opencv opencv@2    #\u524d\u8fb9\u662f\u6700\u65b0\u7684v4.1\u7248\u672c\uff0c@2\u662f\u8001\u7248\u672c\u7684v2.4\r\nln -s \/usr\/local\/Cellar\/opencv@2\/2.4.13.7_3\/lib\/pkgconfig\/opencv.pc \/usr\/local\/lib\/pkgconfig\/opencv.pc  #\u5efa\u7acbopencv.pc\u8f6f\u94fe\r\n#Centos\u5b89\u88c5opencv\r\nyum install cmake gcc gcc-c++ gtk2-devel gimp-develgimp-devel-tools gimp-help-browser zlib-devel libtiff-devel libjpeg-devellibpng-devel gstreamer-devel libavc1394-devel libraw1394-devel libdc1394-develjasper-devel jasper-utils swig python libtool nasm  -y\r\n\r\nwget https:\/\/github.com\/Itseez\/opencv\/archive\/2.4.13.zip\r\nvim CMakeLists.txt  #\u8bbe\u7f6e\u5b89\u88c5\u4f4d\u7f6e\r\n   set(CMAKE_INSTALL_PREFIX \"\/home\/work\/.jumbo\/opt\/opencv\")\r\ncmake .\/\r\nmake\r\nmake install\r\n\r\n#\u68c0\u67e5opencv.pc\u6587\u4ef6\r\nll \/usr\/local\/lib\/pkgconfig\/opencv.pc \r\n#\u6dfb\u52a0path\r\nexport LD_LIBRARY_PATH=\/home\/work\/.jumbo\/opt\/opencv\/lib:${LD_LIBRARY_PATH}\r\nexport PKG_CONFIG_PATH=\/usr\/local\/lib\/pkgconfig    #\u6307\u5b9apkgconfig\u76ee\u5f55\u4f4d\u7f6e\r\n\r\n#\u7f16\u8f91\u914d\u7f6e\r\nvim Makefile\r\n  GPU=1    #\u4f7f\u7528GPU\r\n  CUDNN=1   #\u4f7f\u7528CUDNN\r\n  OPENCV=1  #\u4f7f\u7528OPENCV\r\n  OPENMP=0\r\n  DEBUG=0\r\n\r\n#\u5b89\u88c5\r\nmake\r\n\r\n#\u68c0\u67e5\u662f\u5426\u5b89\u88c5\u6210\u529f\r\n.\/darknet\r\n<\/code><\/pre>\n
 <\/p>\n
\u68c0\u6d4b<\/strong>
\n<\/code><\/p>\n
#\u4e0b\u8f7d\u8bad\u7ec3\u597d\u7684\u6743\u91cd\u6587\u4ef6\r\nwget https:\/\/pjreddie.com\/media\/files\/yolov3.weights         #\u5b8c\u6574\u7248 237M\r\nwget https:\/\/pjreddie.com\/media\/files\/yolov3-tiny.weights    #mini\u7248 34M\r\n\r\n#\u6d4b\u8bd5\u5bf9\u8c61\u68c0\u6d4b(-thresh\u53c2\u6570\uff1a\u5bf9\u8c61\u8bc6\u522b\u7f6e\u4fe1\u5ea6\u9ed8\u8ba4>0.25\uff1b\u68c0\u6d4b\u7ed3\u679c\u8f93\u51fa\u5230out.png)\r\n.\/darknet detect cfg\/yolov3.cfg yolov3.weights data\/dog.jpg -thresh 0.9 -out out\r\nor #\u4e24\u4e2a\u547d\u4ee4\u7b49\u540c\r\n.\/darknet detector test cfg\/coco.data cfg\/yolov3.cfg yolov3.weights data\/dog.jpg -thresh 0.25\r\n\u6216 \u4f7f\u7528\u7cbe\u7b80\u7248\u6a21\u578b\r\n.\/darknet detect cfg\/yolov3-tiny.cfg yolov3-tiny.weights data\/dog.jpg -thresh 0.55 -out out-tiny\r\n\r\n#python\u63a5\u53e3\r\npython .\/python\/darknet.py    #python3\u8fd0\u884c\u6240\u6709\u5b57\u7b26\u4e32\u53c2\u6570\u589e\u52a0str.encode(\"cfg\/yolov3.cfg\")\r\n\r\n<\/code><\/pre>\n
\u8fd0\u884c\u8f93\u51fa\uff1a\uff08\u6ce8\uff1a\u5728mac\u7b49\u975egpu\u673a\u5668\u4e0a\u6bd4\u8f83\u6162\uff09<\/p>\n
\"\"<\/a><\/p>\n
\u68c0\u6d4b\u6548\u679c\uff1aout.png<\/code><\/p>\n
\"\"<\/a><\/p>\n
\u513f\u5b50\u6bd4\u8d5b\u7684\u7167\u7247\u8bc6\u522b\u6548\u679c\uff1a<\/p>\n
\"\"<\/a><\/code>
\n<\/code><\/code><\/p>\n
#\u6253\u5f00\u6444\u50cf\u5934\u5b9e\u65f6\u68c0\u6d4b\uff08\u4f9d\u8d56GPU=1 OPENCV=1 make\u7f16\u8bd1\uff1b-c\u53c2\u6570\uff1a\u6307\u5b9a\u4f7f\u7528\u54ea\u4e2a\u6444\u50cf\u5934\uff09\r\n.\/darknet detector demo cfg\/coco.data cfg\/yolov3.cfg yolov3.weights -c 0   #\u8fd9\u4e2a\u5b9e\u65f6\u68c0\u6d4b\u975eGPU\u57fa\u672c\u8dd1\u4e0d\u52a8\r\n\r\n#\u68c0\u6d4b\u6307\u5b9a\u89c6\u9891\u6587\u4ef6\uff08-prefix\u53c2\u6570\uff1a\u4ee5\u6b64\u53c2\u6570\u503c\u4e3a\u524d\u7f00\u4fdd\u5b58\u5904\u7406\u540e\u7684\u6bcf\u4e00\u5e27\u56fe\u7247\uff09\r\n.\/darknet detector demo cfg\/coco.data cfg\/yolov3.cfg yolov3.weights data\/football.mp4 -thresh 0.9 -prefix football -out_filename data\/football.out.mp4\r\n*\u6ce8\uff1a-out_filename\u4e0d\u597d\u4f7f\u4e0d\u77e5\u9053\u548b\u56de\u4e8b\uff0c\u7528ffmpeg\u5408\u6210\u4e86\u4e00\u4e0b\uff1affmpeg -i .\/football\/video-%d.jpg -c:v libx264 -vf \"fps=25,format=yuv420p\"  football.out.mp4\r\n \r\n<\/code><\/pre>\n
\u89c6\u9891\u5bf9\u8c61\u68c0\u6d4b\u6548\u679c\uff1a<\/p>\n
\nhttps:\/\/yanjingang.com\/blog\/wp-content\/uploads\/2019\/05\/football.out_-2.mp4<\/a><\/video><\/div>\n
\u56fe\u50cf\u7406\u89e3\uff1a<\/p>\n
\"\"<\/a><\/p>\n
<\/h3>\n
\u5982\u4f55\u8bad\u7ec3\u81ea\u5b9a\u4e49\u6a21\u578b<\/strong><\/h1>\n
\u4e0b\u7bc7\u5355\u72ec\u8bb2\uff0c\u53c2\u8003https:\/\/pjreddie.com\/darknet\/yolo\/\u4e2d\u7684 Training YOLO on VOC<\/p>\n
 <\/p>\n
 <\/p>\n
\u53c2\u8003\uff1a<\/span><\/p>\n
https:\/\/pjreddie.com\/<\/span><\/p>\n
https:\/\/www.cnblogs.com\/CZiFan\/p\/9516504.html<\/span><\/p>\n
https:\/\/blog.csdn.net\/u010122972\/article\/details\/83541978<\/span><\/p>\n
https:\/\/cloud.tencent.com\/developer\/news\/76803<\/span><\/p>\n
5\u5206\u949f\u5b66\u4f1aYolo – How to use YOLO with python<\/a><\/span><\/p>\n
\u5982\u4f55\u4fdd\u5b58yolo3\u8fd0\u884c\u68c0\u6d4b\u7684\u89c6\u9891\u7ed3\u679c<\/a><\/span><\/p>\n
DarkNet-YOLOv3 \u8bad\u7ec3\u81ea\u5df1\u7684\u6570\u636e\u96c6 Ubuntu16.04+cuda8.0<\/a><\/span><\/p>\n","protected":false},"excerpt":{"rendered":"
\u6700\u8fd1\u5728\u7814\u7a76\u89c6\u9891\u4e8c\u7ef4\u7801\u5b9a\u4f4d\u6e05\u7406\uff0c\u6d89\u53ca\u5230Object Detection\u5bf9\u8c61\u68c0\u6d4b\u6280\u672f\uff0c\u53d1\u73b0Darknet\u7528\u8d77\u6765\u6548\u679c […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":[],"categories":[788,539],"tags":[793,789,792,799,798,794,797,800,796,795,790,801,791],"_links":{"self":[{"href":"https:\/\/yanjingang.com\/blog\/index.php?rest_route=\/wp\/v2\/posts\/2577"}],"collection":[{"href":"https:\/\/yanjingang.com\/blog\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/yanjingang.com\/blog\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/yanjingang.com\/blog\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/yanjingang.com\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=2577"}],"version-history":[{"count":0,"href":"https:\/\/yanjingang.com\/blog\/index.php?rest_route=\/wp\/v2\/posts\/2577\/revisions"}],"wp:attachment":[{"href":"https:\/\/yanjingang.com\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=2577"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/yanjingang.com\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=2577"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/yanjingang.com\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=2577"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}