{"generator":"Jekyll","link":[{"@attributes":{"href":"https:\/\/dorbodwolf.github.io\/\/feed.xml","rel":"self","type":"application\/atom+xml"}},{"@attributes":{"href":"https:\/\/dorbodwolf.github.io\/\/","rel":"alternate","type":"text\/html"}}],"updated":"2021-08-12T07:54:37+00:00","id":"https:\/\/dorbodwolf.github.io\/\/feed.xml","title":"dorbodwolf(Deyu Tian)","subtitle":"I love coding!","entry":[{"title":"\ud83d\udd25Jekyll\u672c\u5730\u6d4b\u8bd5\u547d\u4ee4","link":{"@attributes":{"href":"https:\/\/dorbodwolf.github.io\/\/jekyll\/update\/2029\/01\/07\/%E4%B8%B4%E6%97%B6-Jekyll%E6%9C%AC%E5%9C%B0%E6%B5%8B%E8%AF%95%E6%96%B9%E6%B3%95.html","rel":"alternate","type":"text\/html","title":"\ud83d\udd25Jekyll\u672c\u5730\u6d4b\u8bd5\u547d\u4ee4"}},"published":"2029-01-07T03:03:05+00:00","updated":"2029-01-07T03:03:05+00:00","id":"https:\/\/dorbodwolf.github.io\/\/jekyll\/update\/2029\/01\/07\/%E3%80%90%E4%B8%B4%E6%97%B6%E3%80%91Jekyll%E6%9C%AC%E5%9C%B0%E6%B5%8B%E8%AF%95%E6%96%B9%E6%B3%95","content":"

bundle exec jekyll serve<\/code><\/p>","author":{"name":{}},"category":[{"@attributes":{"term":"jekyll"}},{"@attributes":{"term":"update"}}],"summary":"bundle exec jekyll serve"},{"title":"\u3010ai\u5168\u6808\u3011\u5143\u5377\u79ef","link":{"@attributes":{"href":"https:\/\/dorbodwolf.github.io\/\/2021\/08\/06\/AI%E5%85%A8%E6%A0%88-%E5%85%83%E5%8D%B7%E7%A7%AF.html","rel":"alternate","type":"text\/html","title":"\u3010ai\u5168\u6808\u3011\u5143\u5377\u79ef"}},"published":"2021-08-06T00:00:00+00:00","updated":"2021-08-06T00:00:00+00:00","id":"https:\/\/dorbodwolf.github.io\/\/2021\/08\/06\/%E3%80%90AI%E5%85%A8%E6%A0%88%E3%80%91%E5%85%83%E5%8D%B7%E7%A7%AF","content":"

\u8fd9\u91cc\u8bf4\u7684\u662f1\u00d71\u5377\u79ef<\/p>\n\n

inception\u589e\u5f3a\u5377\u79ef\u6a21\u5757\u529f\u80fd<\/h2>\n\n

\"image.png\"<\/p>\n\n

\u5347\u7ef4\uff0c\u964d\u7ef4\uff0c\u51cf\u5c11\u53c2\u6570\uff0c\u901a\u9053\u95f4\u878d\u5408<\/p>\n\n

resnet\u6b8b\u5dee\u8fde\u63a5\u7f51\u7edc<\/h2>\n\n

\u51cf\u5c11\u53c2\u6570\u91cf<\/p>\n\n

\u5176\u4e2dbottleneck\u5f62\u5f0f\u7684resnet\u6a21\u5757\u57283\u00d73\u5377\u79ef\u524d\u540e\u90fd\u63a51\u00d71\u5377\u79ef\uff0c\u8fd9\u6837\u5148\u964d\u7ef4\u518d\u5347\u7ef4\uff0c\u76f8\u6bd4\u63a5\u4e24\u4e2a3\u00d73\u5377\u79ef\u6765\u8bf4\u5927\u5927\u964d\u4f4e\u53c2\u6570\u91cf\uff0c\u9002\u5408\u8f83\u6df1\u7684\u7f51\u7edc\u5982resnet101<\/p>\n\n

hrnet\u76f4\u63a5\u7528\u4e8e\u8bed\u4e49\u5206\u5272<\/h2>\n\n

\u6700\u540e\u9762\u7684\u7279\u5f81\u805a\u5408 \u76f8\u5f53\u4e8e\u5168\u8fde\u63a5\uff1f<\/p>","author":{"name":{}},"summary":"\u8fd9\u91cc\u8bf4\u7684\u662f1\u00d71\u5377\u79ef"},{"title":"\u3010ai\u5168\u6808\u3011\u63d2\u503c\u7b97\u6cd5","link":{"@attributes":{"href":"https:\/\/dorbodwolf.github.io\/\/2021\/08\/06\/AI%E5%85%A8%E6%A0%88-%E6%8F%92%E5%80%BC%E7%AE%97%E6%B3%95.html","rel":"alternate","type":"text\/html","title":"\u3010ai\u5168\u6808\u3011\u63d2\u503c\u7b97\u6cd5"}},"published":"2021-08-06T00:00:00+00:00","updated":"2021-08-06T00:00:00+00:00","id":"https:\/\/dorbodwolf.github.io\/\/2021\/08\/06\/%E3%80%90AI%E5%85%A8%E6%A0%88%E3%80%91%E6%8F%92%E5%80%BC%E7%AE%97%E6%B3%95","content":"

\u5728\u91cd\u89c6\u7a7a\u95f4\u4f4d\u7f6e\u7684\u60c5\u51b5\u4e0b\u4f7f\u7528<\/p>\n\n

\u9065\u611f\u56fe\u50cf\u5904\u7406<\/h2>\n\n

\u9065\u611f\u56fe\u50cf\u51e0\u4f55\u7ea0\u6b63\u7b49\u90fd\u7528\u5230\uff0c\u6316\u5751\u3002<\/p>\n\n

AOI Align<\/h2>\n\n

\u53cc\u7ebf\u6027\u63d2\u503c<\/p>\n\n

\u5728Mask R-CNN\u4e2d\uff0croi\u548c\u7279\u5f81\u56fe\u5bf9\u9f50<\/p>\n\n

\u795e\u7ecf\u7f51\u7edc\u7684\u4e0a\u91c7\u6837<\/h2>\n\n

\u548c\u6269\u5f20\u5377\u79ef\u540c\u5c5e\u4e8e\u5168\u5377\u79ef\u7f51\u7edc\u4e0a\u91c7\u6837\u7684\u4e00\u79cd\u65b9\u6cd5<\/p>\n\n

\u6a21\u578b\u8bbe\u8ba1\u4e0a\u968f\u65f6\u7528<\/h2>\n\n

\u539f\u59cbtf\u7248\u672cInception Resnet\u6a21\u5757\u548cInception\u6a21\u5757\u7684pytorch\u590d\u73b0\u7528\u5230\u63d2\u503c\u8ba1\u7b97\u3002<\/p>\n\n

\u5982\u679cstride=1,\u8981\u60f3\u4fdd\u6301tf\u7684padding=\u2019same\u2019\u6548\u679c\uff0c\u5373\u5377\u79ef\u8f93\u51fa\u5c3a\u5bf8\u4fdd\u6301\uff0cpadding=(kernel_size-1)\/2\nhttps:\/\/stackoverflow.com\/questions\/62166719\/padding-same-conversion-to-pytorch-padding<\/p>\n\n

pytorch\u7684maxpool\u4e0d\u652f\u6301\u592a\u5927\u7684pad\uff0c\u4e3a\u4e86\u5728\u6c60\u5316\u540e\u4fdd\u6301\u5c3a\u5bf8\uff0c\u4e8b\u5148\u8fdb\u884c\u7a7a\u95f4\u63d2\u503c\n\u5982\u679cmaxpool\u7684stride=kernel=2,pad=1,\u63d2\u503c\u540e\u5c3a\u5bf8 = (\u63d2\u503c\u524d\u5c3a\u5bf8-1)*2<\/p>","author":{"name":{}},"summary":"\u5728\u91cd\u89c6\u7a7a\u95f4\u4f4d\u7f6e\u7684\u60c5\u51b5\u4e0b\u4f7f\u7528"},{"title":"\u3010ai\u5168\u6808\u3011darknet\u6846\u67b6\u4ee3\u7801\u9605\u8bfb","link":{"@attributes":{"href":"https:\/\/dorbodwolf.github.io\/\/2021\/08\/03\/AI%E5%85%A8%E6%A0%88-darknet%E6%A1%86%E6%9E%B6%E4%BB%A3%E7%A0%81%E9%98%85%E8%AF%BB.html","rel":"alternate","type":"text\/html","title":"\u3010ai\u5168\u6808\u3011darknet\u6846\u67b6\u4ee3\u7801\u9605\u8bfb"}},"published":"2021-08-03T00:00:00+00:00","updated":"2021-08-03T00:00:00+00:00","id":"https:\/\/dorbodwolf.github.io\/\/2021\/08\/03\/%E3%80%90AI%E5%85%A8%E6%A0%88%E3%80%91darknet%E6%A1%86%E6%9E%B6%E4%BB%A3%E7%A0%81%E9%98%85%E8%AF%BB","content":"

\"darknet.png\"<\/p>","author":{"name":{}},"summary":{"@attributes":{"type":"html"}}},{"title":"\u3010ai\u5168\u6808\u3011\u8bed\u4e49\u5206\u5272\u7b97\u6cd5\u603b\u7ed3","link":{"@attributes":{"href":"https:\/\/dorbodwolf.github.io\/\/2021\/08\/03\/AI%E5%85%A8%E6%A0%88-%E8%AF%AD%E4%B9%89%E5%88%86%E5%89%B2%E7%AE%97%E6%B3%95%E6%80%BB%E7%BB%93.html","rel":"alternate","type":"text\/html","title":"\u3010ai\u5168\u6808\u3011\u8bed\u4e49\u5206\u5272\u7b97\u6cd5\u603b\u7ed3"}},"published":"2021-08-03T00:00:00+00:00","updated":"2021-08-03T00:00:00+00:00","id":"https:\/\/dorbodwolf.github.io\/\/2021\/08\/03\/%E3%80%90AI%E5%85%A8%E6%A0%88%E3%80%91%E8%AF%AD%E4%B9%89%E5%88%86%E5%89%B2%E7%AE%97%E6%B3%95%E6%80%BB%E7%BB%93","content":"

1 \u8bed\u4e49\u5206\u5272\u7efc\u8ff0<\/h1>\n\n

1.1 \u8bc4\u4ef7\u6307\u6807-miou\/fwiou<\/h2>\n\n

\u57fa\u4e8e\u6df7\u6dc6\u77e9\u9635\u8ba1\u7b97<\/p>\n\n

1.2 \u6a21\u578b\u539f\u7406<\/h2>\n\n

\u8bed\u4e49\u5206\u5272\u53d1\u5c55\u53f2\uff1a<\/p>\n\n

    \n
  1. \u6df1\u5ea6\u5b66\u4e60\u65b9\u6cd5\u78be\u538b\u4f20\u7edf\u65b9\u6cd5<\/li>\n
  2. \u5728\u5168\u5377\u673a\u7f51\u7edcFCN\u4e4b\u524d\uff0c\u6df1\u5ea6\u5b66\u4e60\u5904\u7406\u8bed\u4e49\u5206\u5272\u7684\u601d\u8def\u662fpatch classification\uff0c\u5373\u5bf9\u56fe\u50cf\u5207\u5757\u6765\u505a\u50cf\u7d20\u5206\u7c7b<\/li>\n
  3. FCN\u662f\u5177\u6709\u91cc\u7a0b\u7891\u610f\u4e49\uff0c\u5377\u79ef\u540e\u63a5\u53cd\u5377\u79ef\uff08\u4e0a\u91c7\u6837\uff09\u800c\u4e0d\u662f\u5168\u8fde\u63a5\u5c42\u6765\u5b9e\u73b0\u9010\u50cf\u7d20\u5206\u7c7b\u3002\u4e0a\u91c7\u6837\u635f\u5931\u7ec6\u8282\u4fe1\u606f\uff0c\u53ef\u4ee5\u901a\u8fc7\u8df3\u8dc3\u8fde\u63a5\u6765\u6709\u6548\u8fd8\u539f\u90e8\u5206\u7ec6\u8282\u4fe1\u606f\u3002<\/li>\n
  4. \u5f52\u5c5e\u4e8eencoder-decoder\u7684unet\u662f\u5728fcn\u7684\u601d\u60f3\u4e0a\u5efa\u7acb\u7684\u7f51\u7edc\uff1b\u7528\u4e8e\u533b\u5b66\u5f71\u50cf\u5206\u5272\uff1b\u7279\u5f81\u63d0\u53d6\u7528\u7684\u4e0d\u662fvgg\u7b49\u9884\u8bad\u7ec3\u7f51\u7edc\uff0c\u800c\u662f\u81ea\u7531\u6269\u5c55\u6df1\u5ea6\u7684cnn\uff1bunet\u8fd8\u6709\u4e00\u4e2a\u7279\u70b9\u662f\u6377\u5f84\u8fde\u63a5\u3002<\/li>\n
  5. \u7a7a\u6d1e\u5377\u79ef\u53d6\u4ee3pooling\u6765\u6269\u5927\u611f\u53d7\u91ce\u4e14\u80fd\u591f\u4fdd\u6301\u56fe\u50cf\u5c3a\u5bf8\u3002<\/li>\n<\/ol>\n\n

    PSPNet\uff1a<\/p>\n

      \n
    1. \u573a\u666f\u5206\u6790<\/li>\n
    2. \u57fa\u4e8efcn\u7684\u65b9\u6cd5\u7f3a\u4e4f\u6070\u5f53\u7684\u624b\u6bb5\u6765\u5229\u7528\u5168\u5c40\u573a\u666f\u7c7b\u522b\u7ebf\u7d22\u3002<\/li>\n
    3. \u7a7a\u95f4\u91d1\u5b57\u5854\u6c60\u5316\u5bf9\u4e8e\u6355\u6349\u5168\u5c40\u4fe1\u606f\u6548\u679c\u5f88\u597d\uff0c\u7a7a\u95f4\u91d1\u5b57\u5854\u7f51\u7edc\u8fdb\u4e00\u6b65\u52a0\u5f3a\u4e86\u8fd9\u4e00\u80fd\u529b\u3002<\/li>\n
    4. \u9664\u4e86\u7ed3\u5408\u5168\u5c40\u548c\u5c40\u90e8\u7684\u7ebf\u7d22\uff0c\u8be5\u8bba\u6587\u63d0\u51fa\u7ed3\u5408\u3010\u6df1\u5ea6\u76d1\u7763\u635f\u5931\u3011\u7684\u4f18\u5316\u7b56\u7565\u3002<\/li>\n<\/ol>\n\n

      deeplab v3+\uff1a<\/p>\n

        \n
      1. modified xception<\/li>\n
      2. xception\u9aa8\u5e72\u7f51\u7edc input flow\u8c03\u7528\u4e09\u6b21block\u51fd\u6570<\/li>\n
      3. \u6df1\u5ea6\u53ef\u5206\u79bb\u5377\u79ef\u7684\u5b9e\u73b0\u7528\u5230\u4e86pytorch\u7684pad\u51fd\u6570<\/li>\n
      4. \u7f51\u7edc\u8bbe\u8ba1\u7406\u5ff5\uff1a\u6a21\u5757\u5316\u8bbe\u8ba1<\/li>\n
      5. encoder-decoder \u6709\u52a9\u4e8e\u63d0\u53d6\u9510\u5229\u8fb9\u7f18\uff1f\uff08deeplabv3+\u6458\u8981\u4e2d\u63cf\u8ff0\uff0c\u7406\u89e3\u6709\u8bef\uff09<\/li>\n
      6. Unet\u548cfcn\u90fd\u5c5e\u4e8e\u7f16\u7801-\u89e3\u7801\u7ed3\u6784<\/li>\n
      7. pytorch\u5b9e\u73b0\uff1asequential \u5b9a\u4e49\u5757<\/li>\n
      8. pytorch forward \u51fd\u6570\u662f\u6a21\u578b\u5b9a\u4e49\u7684\u903b\u8f91\u6d41<\/li>\n<\/ol>\n\n

        \u6ce8\u610f\u529b\u8bed\u4e49\u5206\u5272\uff08\u4fe1\u606f\u878d\u5408\uff09\uff1a<\/p>\n

          \n
        1. nonlocal \u4fdd\u8fb9\u6ee4\u6ce2\uff0c\u4fdd\u6301\u7c7b\u95f4\u65b9\u5dee\u7684\u540c\u65f6\uff0c\u51cf\u5c11\u7c7b\u5185\u65b9\u5dee\u7684\u6548\u679c\u3002<\/li>\n
        2. PSANET \u4e24\u8defattention\uff0c\u76f8\u5f53\u4e8etransformer\u4e2d\u7684\u4e24\u4e2ahead\uff0c\u4e24\u8def\u5206\u522b\u8d77\u5230collect\u548cdistribute\u7684\u4f5c\u7528\u3002<\/li>\n
        3. Senet \u901a\u9053\u4fe1\u606f\u878d\u5408<\/li>\n
        4. Cbam \u4f7f\u7528\u4e86\u538b\u7f29+\u6fc0\u6d3b\u7684\u601d\u60f3\u3002\u540c\u65f6\u4f7f\u7528\u901a\u9053+\u7a7a\u95f4\u4fe1\u606f\u878d\u5408\u3002<\/li>\n
        5. danet\u662fcbam\u548cnon-local\u7684\u7ed3\u5408<\/li>\n
        6. gcnet\u4f18\u5316danet\uff0c\u4f18\u5316no-local\u7684\u65f6\u95f4\u590d\u6742\u5ea6<\/li>\n
        7. Emanet\u662f\u671f\u671b\u6700\u5927\u503c\u7684\u6ce8\u610f\u529b\u673a\u5236<\/li>\n<\/ol>\n\n

          HRNET:<\/p>\n\n

          \u4e00\u4e2a\u7279\u5f81\u63d0\u53d6backbone\uff0c\u5177\u6709\u4fdd\u6301\u5206\u8fa8\u7387\u3001\u9891\u7e41\u4fe1\u606f\u878d\u5408\u7684\u7279\u5f81<\/p>\n\n

          \u7528\u4f5c\u8bed\u4e49\u5206\u5272\u9700\u8981\u628a\u7279\u5f81\u96c6\u5408\u8d77\u6765\u8fde\u63a5\u5206\u7c7b\u5668\u3002<\/p>\n\n

          OCRNET\uff1a<\/p>\n\n

          \u6ce8\u610f\u4e0d\u8981\u548c\u5149\u5b66\u5b57\u7b26\u8bc6\u522b\u7684OCR\u6df7\u6dc6\uff01<\/p>\n\n

          \u76f8\u6bd4\u7279\u5f81\u91d1\u5b57\u5854\u7f51\u7edc\u548c\u7a7a\u6d1e\u5377\u79ef\uff0c\u4e00\u79cd\u65b0\u7684\u6269\u5927\u611f\u53d7\u91ce\u8bbe\u8ba1\uff0c\u539f\u7406\u662f\u5229\u7528baseline\u5206\u5272\u6765\u51c6\u786e\u52fe\u7ed8\u5bf9\u8c61\u3002<\/p>\n\n

          2 \u4f18\u5316\u7b97\u6cd5<\/h1>\n\n

          2.1 \u68af\u5ea6\u4e0b\u964d<\/h2>\n\n

          important kind of search in current AI<\/p>\n\n

          based on concepts from multivariable calculus<\/p>\n\n

          search in continuous state space<\/p>\n\n

          gradientDescent(L){  \/\/\u635f\u5931\u51fd\u6570\u8f93\u5165\n    S = [s1, s2, ..., sn];  \/\/ \u521d\u59cb\u53d8\u91cf\n    repeat{  \n        G = gradient(L,S); \/\/\u6c42L\u5728S\u5904\u7684\u68af\u5ea6G\n        S = S - e*G; \/\/\u66f4\u65b0S\uff0c\u901f\u7387\u662fe\n    } until(termination condition)\n    return S;\n}\n<\/code><\/pre><\/div><\/div>\n\n
          3 \u635f\u5931\u51fd\u6570<\/h1>\n\n
          3.1 \u4ea4\u53c9\u71b5\u635f\u5931\u51fd\u6570<\/h2>\n\n
          3.2 IOU\u635f\u5931\u51fd\u6570<\/h2>","author":{"name":{}},"summary":"1 \u8bed\u4e49\u5206\u5272\u7efc\u8ff0"},{"title":"\u3010ai\u5168\u6808\u3011\u76ee\u6807\u68c0\u6d4b\u7b97\u6cd5\u603b\u7ed3","link":{"@attributes":{"href":"https:\/\/dorbodwolf.github.io\/\/2021\/08\/01\/AI%E5%85%A8%E6%A0%88-%E7%9B%AE%E6%A0%87%E6%A3%80%E6%B5%8B%E7%AE%97%E6%B3%95%E6%80%BB%E7%BB%93.html","rel":"alternate","type":"text\/html","title":"\u3010ai\u5168\u6808\u3011\u76ee\u6807\u68c0\u6d4b\u7b97\u6cd5\u603b\u7ed3"}},"published":"2021-08-01T00:00:00+00:00","updated":"2021-08-01T00:00:00+00:00","id":"https:\/\/dorbodwolf.github.io\/\/2021\/08\/01\/%E3%80%90AI%E5%85%A8%E6%A0%88%E3%80%91%E7%9B%AE%E6%A0%87%E6%A3%80%E6%B5%8B%E7%AE%97%E6%B3%95%E6%80%BB%E7%BB%93","content":"
          \u76ee\u6807\u68c0\u6d4b\u7efc\u8ff0<\/h1>\n\n
          \u76ee\u6807\u68c0\u6d4b\u5c31\u662f\u6c42\u89e3\u56fe\u50cf\u4e2d\u7684\u76ee\u6807\u6846\u5982\u732b:(x, y, w, h)<\/code>\u3002<\/p>\n\n
          \u5982\u679c\u6309\u7167\u5206\u7c7b\u7684\u601d\u8def\u53bb\u505a\u76ee\u6807\u68c0\u6d4b\uff0c\u5373\u5efa\u7acb\u4e00\u4e2a\u5377\u79ef\u795e\u7ecf\u7f51\u7edc+\u5168\u8054\u63a5\u5c42\u6765\u4ece\u56fe\u50cf\u4e2d\u53d6\u4e0d\u540c\u7684\u611f\u5174\u8da3\u533a\u57dfRoI\u6765\u5206\u7c7b\uff0c\u4f46\u662f\u8fd9\u79cd\u505a\u6cd5\u5e76\u4e0d\u9ad8\u6548\uff0c\u6240\u4ee5\u624d\u6709\u4e86R-CNN\uff0cYOLO\u7b49\u7b97\u6cd5\u6765\u9ad8\u6548\u7684\u89e3\u51b3\u8fd9\u7c7b\u95ee\u9898\u3002<\/p>\n\n
          R-CNN<\/h1>\n\n
          \u4e3a\u4e86\u907f\u514d\u5728\u56fe\u50cf\u4e2d\u9009\u62e9\u5927\u91cf\u7684Regions\uff0cRoss Girshick et al.<\/a>\u63d0\u51fa\u4e86\u4e00\u79cd\u65b9\u6cd5\uff0c\u8fd9\u79cd\u65b9\u6cd5\u4f7f\u7528selective search\u53ea\u9700\u8981\u4ece\u56fe\u50cf\u4e2d\u63d0\u53d62000\u4e2a\u533a\u57df\uff0c\u79f0\u4f5cregion proposals\u3002selective search\u7b97\u6cd5\u7684\u539f\u7406\u5982\u4e0b\uff1a<\/p>\n
          1. \u751f\u6210\u521d\u59cb\u7684sub-segementation\uff0c\u751f\u6210\u4e86\u8bb8\u591a\u5019\u9009\u7684region\n2. \u4f7f\u7528\u8d2a\u5fc3\u7b97\u6cd5\u9012\u5f52\u7684\u5408\u5e76\u5c0f\u533a\u57df\u4e3a\u5927\u533a\u57df\n3. \u4f7f\u7528\u751f\u6210\u7684\u533a\u57df\u6765\u751f\u4ea7\u6700\u7ec8\u7684\u5019\u9009region proposals\n<\/code><\/pre><\/div><\/div>\n
          \"R-CNN\"<\/p>\n\n
          \u63d0\u53d6\u76842000\u4e2aregion proposals\u88ab\u9001\u8fdbCNN\u7f51\u7edc\u5f97\u52304096\u7ef4\u7684\u7279\u5f81\u5411\u91cf\u8f93\u51fa\uff0c\u6700\u7ec8\u9001\u5230SVM\u5206\u7c7b\u5668\u786e\u8ba4\u5176\u4e2d\u662f\u5426\u6709\u67d0\u4e00\u7c7b\u76ee\u6807\u3002\u5bf9\u4e8e\u76ee\u6807\u68c0\u6d4b\uff0c\u5149\u6709\u5206\u7c7b\u5668\u8fd8\u4e0d\u591f\uff0c\u7f51\u7edc\u6700\u540e\u8fd8\u8981\u9001\u5165\u4e00\u4e2aBbox reg\u6a21\u5757\uff0c\u6765\u9884\u6d4bbounding box\u7684\u56db\u4e2a\u504f\u79fb\u91cf\uff0c\u4ee5\u7cbe\u51c6\u7684\u5f97\u5230\u76ee\u6807\u7684\u8303\u56f4\u3002<\/p>\n\n
          \"R-CNN\"<\/p>\n\n
          selective search\u7b97\u6cd5\u5c3d\u7ba1\u4e00\u5b9a\u7a0b\u5ea6\u4e0a\u964d\u4f4e\u4e86\u8ba1\u7b97\u91cf\uff0c\u4f46\u662f\u5728\u63d0\u53d6region proposals\u7684\u8fc7\u7a0b\u4e2d\u6ca1\u6709\u8bad\u7ec3\u53c2\u4e0e\uff0c\u6240\u4ee5\u5f97\u5230\u7684\u5019\u9009rp\u4e0d\u4e00\u5b9a\u597d\u3002<\/p>\n\n
          Fast R-CNN<\/h1>\n\n
          \u76f8\u6bd4R-CNN\u8981\u628a2000\u4e2arp\u9001\u8fdbCNN\u5206\u522b\u8fdb\u884c\u5377\u79ef\u8017\u8d39\u65f6\u95f4\uff0cFast R-CNN\u53ea\u9700\u8981\u5c06\u56fe\u7247\u9001\u8fdbCNN\u505a\u4e00\u6b21\u5377\u79ef\uff0c\u7136\u540e\u4ece\u5f97\u5230\u7684\u7279\u5f81\u6620\u5c04\u4e2d\u8bc6\u522brp\uff0c\u7136\u540e\u63a5roi pooling\u5c06\u76ee\u6807\u6846\u5bf9\u5e94\u7279\u5f81\u56feresize\u5230\u76f8\u540c\u7684\u5c3a\u5ea6\u4ee5\u6ee1\u8db3\u5168\u8fde\u63a5\u5c42\u56fa\u5b9a\u5c3a\u5bf8\u7684\u8f93\u5165\u5236\u7ea6\uff0c\u63a5\u4e24\u5c42\u5168\u8fde\u63a5\u5c42\u540e\u5206\u522b\u8fdb\u884c\u76ee\u6807\u7684softmax\u5206\u7c7b\u548c\u57fa\u4e8e\u56de\u5f52\u7684\u76ee\u6807\u6846\u504f\u79fb\u9884\u6d4b\u3002<\/p>\n\n
          Fast R-CNN\u65e0\u8bba\u662f\u8bad\u7ec3\u8fd8\u662f\u6d4b\u8bd5\u90fd\u8981\u6bd4R-CNN\u5feb\u7684\u591a\uff0c\u7136\u800c\uff08\u53f3\u56fe\u53ef\u89c1\uff09region proposals\u4f9d\u7136\u662f\u5236\u7ea6Fast R-CNN\u7684test\u73af\u8282\u901f\u5ea6\u7684\u74f6\u9888\u3002\n\"Comparison<\/p>\n\n
          RoI Pooling\u7684\u8ba1\u7b97\u539f\u7406\u53ef\u4ee5\u53c2\u8003\u4ee5\u4e0b\u52a8\u56feRoI Pooling<\/a>\uff0c\u5206\u6790\u4e4b\u524d\u5148\u6765\u770b\u770bRoI Pooling Layer\u7684caffe prototxt\u7684\u5b9a\u4e49\uff1a<\/p>\n
          layer {\n  name: \"roi_pool5\"\n  type: \"ROIPooling\"\n  bottom: \"conv5_3\"\n  bottom: \"rois\"\n  top: \"pool5\"\n  roi_pooling_param {\n    pooled_w: 7\n    pooled_h: 7\n    spatial_scale: 0.0625 # 1\/16\n  }\n}\n<\/code><\/pre><\/div><\/div>\n
          \u5176\u4e2d\u6709\u65b0\u53c2\u6570pooled_w\u548cpooled_h\uff0c\u53e6\u5916\u4e00\u4e2a\u53c2\u6570spatial_scale\u8ba4\u771f\u9605\u8bfb\u7684\u8bfb\u8005\u80af\u5b9a\u5df2\u7ecf\u77e5\u9053\u77e5\u9053\u7528\u9014\u3002RoI Pooling layer forward\u8fc7\u7a0b\uff1a<\/p>\n
          1. \u7531\u4e8eproposal\u662f\u5bf9\u5e94MxN\u5c3a\u5ea6\u7684\uff0c\u6240\u4ee5\u9996\u5148\u4f7f\u7528spatial_scale\u53c2\u6570\u5c06\u5176\u6620\u5c04\u56de(M\/16)x(N\/16)\u5927\u5c0f\u7684feature map\u5c3a\u5ea6\uff1b\n2. \u518d\u5c06\u6bcf\u4e2aproposal\u5bf9\u5e94\u7684feature map\u533a\u57df\u6c34\u5e73\u5206\u4e3a[pooled_w x pooled_h]\u7684\u7f51\u683c\uff1b\n3. \u5bf9\u7f51\u683c\u7684\u6bcf\u4e00\u4efd\u90fd\u8fdb\u884cmax pooling\u5904\u7406\u3002\n<\/code><\/pre><\/div><\/div>\n
          \u8fd9\u6837\u5904\u7406\u540e\uff0c\u5373\u4f7f\u5927\u5c0f\u4e0d\u540c\u7684proposal\u8f93\u51fa\u7ed3\u679c\u90fd\u662fpooled_w x pooled_h<\/strong>\u56fa\u5b9a\u5927\u5c0f\uff0c\u5b9e\u73b0\u4e86\u56fa\u5b9a\u957f\u5ea6\u8f93\u51fa\u3002<\/p>\n\n
          Faster R-CNN<\/h1>\n\n
          R-CNN & Fast R-CNN\u901a\u8fc7selective search\u7b97\u6cd5\u6765\u5bfb\u627eregion proposals\u65f6\u95f4\u6548\u7387\u592a\u4f4e\uff0cShaoqing Ren et al.<\/a>\u6448\u5f03\u4e86selective search\u7b97\u6cd5\uff0c\u8ba9\u7f51\u7edc\u53bb\u5b66\u4e60region proposals\uff0c\u8fd9\u4e2a\u72ec\u7acb\u7684\u7f51\u7edc\u53eb\u533a\u57df\u5019\u9009\u7f51\u7edcregion proposals network\uff08rpn\uff09\u3002<\/p>\n\n
          \u76f8\u6bd4Fast R-CNN\u5bf9cnn\u63d0\u53d6\u7684feature map\u65bd\u52a0selective search\u7b97\u6cd5\u751f\u6210rp\uff0cFaster R-CNN\u63a5rpn\u7f51\u7edc\u751f\u6210rp\uff0c\u4e4b\u540e\u7684\u64cd\u4f5c\u548cFast R-CNN\u65e0\u592a\u5927\u5dee\u522b\u3002<\/p>\n\n
          RPN\u7f51\u7edc\u7528\u4e8e\u751f\u6210region proposals\uff0c\u539f\u7406\u5982\u4e0b\uff1a<\/p>\n
          1. Conv layers\u8ba1\u7b97\u83b7\u5f97\u7684feature maps\uff0c\u4e3a\u6bcf\u4e00\u4e2a\u70b9\u90fd\u914d\u5907\u4e0d\u540c\u5c3a\u5ea6\u76849\u79cdanchors\u4f5c\u4e3a\u521d\u59cb\u7684\u68c0\u6d4b\u6846;\n2. \u901a\u8fc7softmax\u5224\u65adanchors\u5c5e\u4e8epositive\u6216\u8005negative\uff0c\u6309\u7167\u6b63\u8d1f\u6bd4\u4f8b1:3\u4fdd\u7559anchors;\n3. \u518d\u5229\u7528bounding box regression\u4fee\u6b63anchors\u83b7\u5f97\u7cbe\u786e\u7684proposals;\n<\/code><\/pre><\/div><\/div>\n\n
          Faster R-CNN\u6a21\u578b\u4e00\u5e45\u56fe0.2s\u7684\u68c0\u6d4b\u901f\u5ea6\u5df2\u7ecf\u53ef\u4ee5\u4f5c\u4e3a\u4e00\u4e2a\u5b9e\u65f6\u68c0\u6d4b\u7b97\u6cd5\u4f7f\u7528\u4e86\uff1a\n\"Comparison<\/p>\n\n
          Mask R-CNN<\/h1>\n\n
          \u4f55\u51ef\u660e\u5927\u795e\u6240\u8a00\uff0c\u5b9e\u4f8b\u5206\u5272\u6709\u4e24\u79cd\u601d\u8def\uff0c\u4e00\u662fsegmentation-first\uff0c\u4e8c\u662finstance-first\uff0cMask R-CNN\u662f\u540e\u8005<\/p>\n\n
          RoI Pooling\u7684\u7f3a\u9677\u548cRoIAlign\u7684\u5f15\u5165\uff1a\n\u5206\u5757\u6c60\u5316\u5bfc\u81f4RoI\u548cFeaure Map\u9519\u4f4d(misalignments)\uff0c\u8fd9\u5bf9\u5206\u7c7b\u6765\u8bf4\u5e26\u6765\u8bef\u5dee\u53ef\u4ee5\u5ffd\u7565\u4e0d\u8bb0\uff0c\u4f46\u662f\u5bf9\u4e8e\u9010\u50cf\u7d20\u7684\u5206\u5272\u63a9\u819c\u6765\u8bf4\u5f71\u54cd\u5c31\u5927\u4e86\u3002<\/p>\n\n
          Mask-RCNN\uff1a\n\u76ee\u6807\u68c0\u6d4b\u7528\u4e86faster r-cnn\uff0c\u7279\u5f81\u63d0\u53d6\u7528\u4e86\u7279\u5f81\u91d1\u5b57\u5854\u7f51\u7edcfpn<\/p>\n\n
          \u5b9e\u9a8c\u63d0\u5347tricks\uff1a<\/p>\n
            \n  
          1. \u7aef\u5230\u7aef\u8bad\u7ec3\u76f8\u6bd4\u5148\u8bad\u7ec3RPN\u518d\u8bad\u7ec3Mask R-CNN\u4f1a\u63d0\u5347\u7cbe\u5ea6<\/li>\n  
          2. \u57fa\u4e8eImageNet\u9884\u8bad\u7ec3\u53ef\u4ee5\u63d0\u5347\u7cbe\u5ea6<\/li>\n  
          3. \u8bad\u7ec3\u65f6\u589e\u5f3a\u53ef\u4ee5\u63d0\u5347\u7cbe\u5ea6<\/li>\n  
          4. \u628aResNeXt\u4ece101\u5c42\u63d0\u5347\u5230152\u5c42\u53ef\u4ee5\u63d0\u5347\u7cbe\u5ea6<\/li>\n  
          5. \u6d4b\u8bd5\u65f6\u589e\u5f3a\u53ef\u4ee5\u63d0\u5347\u7cbe\u5ea6<\/li>\n<\/ol>\n\n
            YOLOv1<\/h1>\n\n
            \u539f\u7406\u4ecb\u7ecd<\/strong><\/p>\n\n
            \u628a\u76ee\u6807\u68c0\u6d4b\u5f53\u4f5c\u4e00\u4e2a\u56de\u5f52\u95ee\u9898\u6c42\u89e3\uff0cinference\u53ea\u9700\u4e00\u6b21\u56de\u5f52\uff0c\u7528\u901f\u5ea6\u6362\u7cbe\u5ea6\u3002<\/p>\n\n
            yolov1\u628a\u8f93\u5165\u56fe\u50cf\u5212\u5206\u4e3aS*S\u7684\u683c\u6805\uff0c\u6bcf\u4e2a\u683c\u6805\u8d1f\u8d23\u68c0\u6d4b\u4e2d\u5fc3\u843d\u5728\u8be5\u683c\u6805\u7684\u76ee\u6807\uff0c\u5305\u62ec\u76ee\u6807\u7684\u5750\u6807\u503c(x, y, w, h)\u548cconfidence scores\u3002\u53e6\u5916\uff0c\u6bcf\u4e2a\u683c\u6805\u9884\u6d4b\u4e00\u4e2a\u7c7b\u522b\u6982\u7387\u503c\uff0c\u8f93\u51fa\u4e00\u5f20\u7c7b\u522b\u6982\u7387\u56fe\u3002<\/p>\n
            confidence = P(obj) * IoU(truth,pred)\n<\/code><\/pre><\/div><\/div>\n\n
            \u4e0a\u5f0f\u4e2d\uff0c\u5f53\u683c\u6805\u4e2d\u5fc3\u70b9\u6ca1\u6709\u76ee\u6807\u65f6P(obj)\u4e3a0\uff0c\u6b64\u65f6confidence\u4e3a0\uff1b\u5426\u5219\u4e3aIOU\u7684\u503c\u3002<\/p>\n\n
            \"yolov1\"<\/p>\n\n
            \u7f51\u7edc\u7ed3\u6784<\/strong><\/p>\n\n
            \"YOLO.png\"<\/p>\n\n
            \u8bad\u7ec3\u9636\u6bb5<\/strong><\/p>\n\n
            \u975e\u7ebf\u6027\u6fc0\u6d3b\u51fd\u6570\uff1aLeaky RELU<\/p>\n\n
            \u635f\u5931\u51fd\u6570\u8bbe\u8ba1\uff1a\u635f\u5931\u51fd\u6570\u8bbe\u8ba1\u7684\u76ee\u6807\u662f\u5c06\u5750\u6807 (x, y, w, h)\uff0cconfidence\uff0cclassification\u8fd9\u4e09\u4e2a\u65b9\u9762\u8fbe\u5230\u5f88\u597d\u7684\u5e73\u8861\uff0c\u6240\u4ee5\u4e0d\u80fd\u7b80\u5355\u7684\u7528\u5e73\u65b9\u6839\u8bef\u5dee\u6765\u5bf9\u6240\u6709\u53d8\u91cf\u8fdb\u884c\u56de\u5f52\u3002<\/p>\n\n
            \"YOLOv1\u635f\u5931\u51fd\u6570\u8bbe\u8ba1\"<\/p>\n\n
            \u4e0a\u56fe\u53ef\u89c1\u635f\u5931\u51fd\u6570\u5305\u542b\u4e94\u9879\u7684\u548c\u3002<\/p>\n\n
            p1 \u2014\u2014\u2014\u2014\u7b2c\u4e00\u9879+\u7b2c\u4e8c\u9879\u662f\u5750\u6807\u7684\u504f\u79fb\uff0c\u8d4b\u4e88\u5f88\u9ad8\u7684loss\u6743\u91cd\uff0c\u5b9e\u9a8c\u4e2d\u53d65\u3002\u7b2c\u4e00\u9879\u662fx\u548cy\u5750\u6807\u7684\u504f\u79fb\uff0c\u7b2c\u4e8c\u9879\u662fw\u548ch\u5750\u6807\u7684\u504f\u79fb\u3002<\/p>\n\n
            p2 \u2014\u2014\u2014\u2014\u9488\u5bf9\u7b2c\u4e8c\u9879\uff0c\u4f5c\u8005\u7528w\u548ch\u7684\u5e73\u65b9\u6839\u504f\u5dee\u6765\u8ba1\u7b97loss\u662f\u4e3a\u4e86\u51f8\u663e\u5c0f\u76ee\u6807\u53d1\u751f\u504f\u5dee\u7684\u635f\u5931\u3002<\/p>\n\n
            \u89e3\u91ca\uff1a\u5728yolo\u7684\u6837\u672c\u6807\u7b7e\u4e2d\uff0cw\u548ch\u662fbbox\u76f8\u5bf9\u56fe\u50cf\u7684\u5bbd\u9ad8\u5927\u5c0f\uff0c\u76ee\u6807\u8d8a\u5927\uff0cw\u548ch\u5c31\u8d8a\u5927\uff0c\u76ee\u6807\u8d8a\u5c0f\uff0cw\u548ch\u5c31\u8d8a\u5c0f\u3002\u5c0f\u7684\u76ee\u6807\u53d1\u751f\u0394w\u548c\u0394h\u7684\u504f\u5dee\u53cd\u6620\u5728\u5e73\u65b9\u6839\u4e0a\u8981\u5927\u4e8e\u5927\u76ee\u6807\u7684\u503c\uff0c\u4ee5\u6b64\u6765\u653e\u5927\u5c0f\u76ee\u6807\u7684\u635f\u5931\uff0c\u6765\u63d0\u5347\u5c0f\u76ee\u6807\u7684\u68c0\u6d4b\u6548\u679c\uff0c\u5177\u4f53\u539f\u7406\u89c1\u4e0b\u56fe\uff1a<\/p>\n\n
            \"yolov1\u5c0f\u76ee\u6807loss\u653e\u5927\u6570\u5b66\u539f\u7406\"<\/p>\n\n
            p3 \u2014\u2014\u2014\u2014\u7b2c\u4e09\u9879+\u7b2c\u56db\u9879\u662fbbox\u7684confidence\u635f\u5931\uff0c\u4f46\u662f\u5206\u4e3a\u542b\u6709object\u548c\u4e0d\u542bobject\uff0c\u4e0d\u542bobject\u7684confidence\u635f\u5931\u7ed9\u4e860.5\u7684\u6743\u91cd\u3002\u6211\u662f\u8fd9\u6837\u7406\u89e3\u7684\uff0c\u4e0d\u542bobject\u7684bbox\u8981\u6bd4\u542b\u6709object\u7684bbox\u591a\u7684\u591a\uff0c\u4e3a\u4e86\u9632\u6b62\u5176\u631f\u6301\u68af\u5ea6\u4e0b\u964d\u8fc7\u7a0b\u5bfc\u81f4\u6a21\u578b\u96be\u6536\u655b\u6216\u53d1\u6563\u800c\u7ed9\u4e88\u5c0f\u7684\u6743\u91cd\u3002<\/p>\n\n
            p4 \u2014\u2014\u2014\u2014\u7b2c\u4e94\u9879\u662f\u7c7b\u522b\u56fe\u7684\u635f\u5931\uff0c\u6743\u91cd\u4e3a1\u3002<\/p>\n\n
            \u63a8\u7406\u9636\u6bb5<\/strong>\n\u6bcf\u4e2a\u683c\u6805\u5305\u542bSS<\/em>B\u4e2abbox\u7684\u9884\u6d4b\uff0c\u5982\u679cS=7 B=2\uff0c\u5219\u6bcf\u4e2a\u683c\u6805\u670998\u4e2abbox\u9884\u6d4b\u7ed3\u679c.<\/p>\n\n
            \u5c06bbox\u7684IoU\u548c\u683c\u6805\u7684\u7c7b\u522b\u6982\u7387\u76f8\u4e58\u5f97\u5230class-special confidence score\uff0c\u8fd9\u4e2ascore\u8861\u91cf\u9884\u6d4b\u7684\u7cbe\u5ea6.<\/p>\n\n
            1. \u6bcf\u4e2a\u683c\u6805\u7684\u6bcf\u4e2abbox\u90fd\u8981\u8ba1\u7b97class-special confidence score;\n2. \u6309\u7167score\u8fdb\u884c\u9608\u503c\u8fc7\u6ee4\u6765\u5269\u4e0b\u9ad8\u5206\u6570\u7684bbox;\n3. \u518d\u8fdb\u884cNMS\u91cd\u53e0\u8fc7\u6ee4\u6765\u8f93\u51fa\u6700\u7ec8\u7684\u9884\u6d4b\u7ed3\u679c\u3002\n<\/code><\/pre><\/div><\/div>\n\n
            YOLOv2<\/h1>\n\n
            \u63d0\u5347\uff1aanchor boxes<\/strong><\/p>\n\n
            \u76f8\u6bd4yolov1\u5212\u5206SS\u5927\u5c0f\u7684\u683c\u6805\uff0c\u6bcf\u4e2a\u683c\u6805\u9884\u6d4bB\u4e2a\u76ee\u6807\uff0cyolov2\u5f15\u5165anchor boxes\uff0c\u5728\u7279\u5f81\u56fe\u7684S<\/em>S\uff0813 * 13\uff09\u7684\u683c\u6805\u4e0a\u9884\u6d4b5\u4e2aanchor boxes\uff0c\u5bf9\u4e8e\u6bcf\u4e2aanchor box\u9884\u6d4bbbox\u7684\u7684\u5750\u6807\u3001\u7f6e\u4fe1\u5ea6\u548c\u5206\u7c7b\u6982\u7387\uff0c\u53ef\u4ee5\u9884\u6d4b13 * 13 * 5 = 845\u4e2abbox\uff0c\u5927\u5927\u63d0\u5347\u6a21\u578b\u53ec\u56de\u7387\uff08\u6f0f\u68c0\u5c11\u4e86\uff09\u3002<\/p>\n\n
            \u540c\u6837\u662fancher-based\u7684\u68c0\u6d4b\u5668\uff0c\u4e0d\u540c\u4e8efaster rcnn\u5728\u6bcf\u4e2acell\u624b\u52a8\u8bbe\u7f6e9\u4e2aanchor boxes\uff0cyolov2\u7684\u4f5c\u8005\u901a\u8fc7\u5728COCO\u548cVOC\u6570\u636e\u96c6\u4e0a\u505akmeans\u805a\u7c7b\u6765\u786e\u5b9aanchor boxes\u7684\u4e2a\u6570\u3001\u5206\u5e03\u548c\u5f62\u6001\u3002\u5982\u4e0b\u56fe\uff1a<\/p>\n\n
            \"kmeans\u805a\u7c7b\u786e\u5b9aanchor<\/p>\n\n
            \u601d\u8003\uff1a\u5bf9\u4e8e\u6211\u4eec\u7684\u9065\u611f\u6570\u636e\u96c6\u4e5f\u6309\u8fd9\u6837\u7684\u65b9\u5f0f\u8bbe\u8ba1anchor\u662f\u4e0d\u662f\u53ef\u4ee5\u63d0\u5347\u68c0\u6d4b\u6548\u679c\uff1f<\/em><\/p>\n\n
            \u8def\u4eba\u7532\uff1a\u805a\u7c7b\u5b8c\u66f4\u5dee<\/p>\n\n
            \u8def\u4eba\u4e59\uff1a\u805a\u7c7b\u4e86\u6536\u655b\u6bd4\u8f83\u5feb\uff0c\u6027\u80fd\u63d0\u5347\u4e00\u4e22\u4e22<\/p>\n\n
            \u63d0\u5347\uff1a\u65b0\u7684backbone<\/strong><\/p>\n\n
            Darknet-19<\/p>\n\n
            \u63d0\u5347\uff1a\u591a\u5c3a\u5ea6\u8bad\u7ec3<\/strong><\/p>\n\n
            \u4e0d\u56fa\u5b9a\u8bad\u7ec3\u56fe\u50cf\u5c3a\u5bf8<\/p>\n\n
            \u63d0\u5347\uff1a\u635f\u5931\u51fd\u6570<\/strong><\/p>\n\n
            \u6cbf\u7528yolov1\u7684\u8bad\u7ec3\u65b9\u5f0f\uff0c\u5c06\u635f\u5931\u5206\u4e3a\u5750\u6807\u3001IoU\u548c\u7c7b\u522b\u6982\u7387\u4e09\u90e8\u5206\uff0c\u4f46\u662f\u505a\u4e86\u4e0d\u5c11\u6539\u8fdb\uff0c\u53d8\u5f97\u66f4\u590d\u6742\u4e86\u3002<\/p>\n\n
            YOLOv3<\/h1>\n\n
            \u770b\u4e0b\u56fe\u5c31\u591f\u4e86\uff0c\u76f8\u6bd4v2\u4e5f\u5c31\u589e\u52a0\u4e86\u6b8b\u5dee\u6a21\u578bDarknet-53\u548cFPN\u7ed3\u6784\u3002<\/p>\n\n
            \"network_of_yolov3\"<\/p>\n\n
            \u89e3\u91ca\u4e0a\u56fe\uff1a<\/p>\n\n
            \u5de6\u4fa7\u662f\u5305\u542b\u4e94\u4e2a\u6b8b\u5dee\u5757\u7684Darknet-53 backbone<\/p>\n\n
            \u53f3\u4fa7\u4e3aFPN\u7ed3\u6784\u53ca\u5728\u5176\u4e0a\u8fdb\u884c\u7684\u4e09\u4e2aanchor-base\u7684\u68c0\u6d4b\u8def. FPN\u5c31\u662f\u4e00\u79cd\u91d1\u5b57\u5854\u7ed3\u6784\uff0c\u5728\u4e0d\u540c\u5c3a\u5ea6\u4e0a\u8f93\u51fa\u7279\u5f81\u6620\u5c04<\/p>\n\n
            yolov3\u5728\u4e09\u4e2a\u5c3a\u5ea6\u4e0a\u8f93\u51fa\u7279\u5f81\u6620\u5c04\u8fdb\u884c\u9884\u6d4b\uff0c\u7b97\u662f\u591a\u5c3a\u5ea6\u9884\u6d4b\uff0c\u4e09\u4e2a\u8f93\u51fa\u7279\u5f81\u6620\u5c04\u5206\u522b\u662f(batch, 52, 52, 75), (batch, 26, 26, 75), (batch, 13, 13, 75)\u3002\u6bcf\u4e2a\u9884\u6d4b\u5c3a\u5ea6\u4e0a\u57fa\u4e8e\u7ef4\u5ea6\u805a\u7c7b\u7684\u65b9\u6cd5\u5728\u6bcf\u4e2acell\u4e0a\u751f\u62103\u4e2aanchor boxes\uff0c\u548cfaster rcnn\u4e00\u6837\u603b\u5171\u4e5f\u662f9\u4e2aanchor boxes\u3002<\/p>\n\n
            yolov3-tiny\u662f\u66f4\u52a0\u8f7b\u91cf\u7ea7\u7684yolov3\uff0c\u5c42\u6570\u5927\u5927\u7ea6\u51cf\uff0c\u4f46\u662f\u4f9d\u7136\u4fdd\u7559\u4e86\u4e24\u4e2a\u5c3a\u5ea6\u7684\u72ec\u7acb\u68c0\u6d4b\uff0c\u8f93\u51fa\u5206\u522b\u4e3a(, 13, 13, 255)\u548c(, 26, 26, 255),\u7f51\u7edc\u7ed3\u6784\u89c1\u4e0b\u56fe\uff1a<\/p>\n\n
            \"yolo_v3_tiny\"<\/p>\n\n
            YOLOv4<\/h1>\n\n
            \u7279\u5f81\u63d0\u53d6\u66f4\u590d\u6742\u4e86\uff0c\u6570\u636e\u6269\u5145\u65b9\u6cd5\u589e\u52a0\u4e86\u4e0d\u5c11\uff0c\u68c0\u6d4b\u5934\u6ca1\u4ec0\u4e48\u53d8\u5316\u3002<\/p>\n\n
            \u76ee\u6807\u68c0\u6d4b\u601d\u7ef4\u5bfc\u56fe<\/h1>\n\n
            \u76ee\u6807\u68c0\u6d4b\u7f51\u7edc\u4e00\u822c\u7ed3\u6784\uff1a<\/p>\n\n
            \"detector.png\"<\/p>\n\n
            \u76ee\u6807\u68c0\u6d4b\u63d0\u5347\u65b9\u6cd5\uff1a<\/p>\n\n
            \"detector_improve.png\"<\/p>","author":{"name":{}},"summary":"\u76ee\u6807\u68c0\u6d4b\u7efc\u8ff0"},{"title":"\u3010gis\u3011\u5de7\u7528gdal tiff\u521b\u5efa\u9009\u9879","link":{"@attributes":{"href":"https:\/\/dorbodwolf.github.io\/\/2021\/05\/28\/GIS-%E5%B7%A7%E7%94%A8GDAL-TIFF%E5%88%9B%E5%BB%BA%E9%80%89%E9%A1%B9.html","rel":"alternate","type":"text\/html","title":"\u3010gis\u3011\u5de7\u7528gdal tiff\u521b\u5efa\u9009\u9879"}},"published":"2021-05-28T00:00:00+00:00","updated":"2021-05-28T00:00:00+00:00","id":"https:\/\/dorbodwolf.github.io\/\/2021\/05\/28\/%E3%80%90GIS%E3%80%91%E5%B7%A7%E7%94%A8GDAL%20TIFF%E5%88%9B%E5%BB%BA%E9%80%89%E9%A1%B9","content":"
            1 \u7b80\u4ecb<\/h1>\n\n
            gdal\u652f\u6301\u4f17\u591a\u6805\u683c\u9a71\u52a8\u5668\uff0cGeoTIFF\u5c31\u662f\u5176\u4e2d\u6700\u5e38\u7528\u7684\uff0cTIFF\u683c\u5f0f\u6709\u5f88\u591a\u521b\u5efa\u9009\u9879\uff08Create Option\uff09\uff0c\u5728\u4ee3\u7801\u4e2d\u548c\u811a\u672c\u4e2d\u53ef\u4ee5\u6307\u5b9a\u8fd9\u4e9b\u521b\u5efa\u9009\u9879\u6765\u81ea\u5b9a\u4e49TIFF\u6587\u4ef6\u8f93\u51fa\u3002\u5728GDAL\u81ea\u5e26\u811a\u672c\u4e2d\u53ef\u4ee5\u901a\u8fc7-co<\/code>\u9009\u9879\u6765\u6dfb\u52a0\u521b\u5efa\u9009\u9879\u3002<\/p>\n\n
            2 \u811a\u672c\u5b9e\u4f8b\u2013\u9576\u5d4c\u65f6\u5e94\u7528<\/h1>\n\n
            \u5229\u7528Google earth engine\u7ebf\u4e0a\u505a\u5730\u7269\u5206\u7c7b\uff0c\u5206\u7c7b\u7ed3\u679c\u4e0b\u8f7d\u5230\u672c\u5730\u662f\u5206\u5e45\u5b58\u50a8\u7684\uff0c\u9700\u8981\u9576\u5d4c\u8d77\u6765\uff0c\u4fdd\u5b58\u7ed3\u679c\u60f3\u8981\u5c3d\u53ef\u80fd\u538b\u7f29\u6765\u63d0\u9ad8\u5b58\u50a8\u6548\u7387\u3002<\/p>\n\n
            \u7248\u672c1<\/h3>\n\n
            \u4e0d\u505a\u538b\u7f29\uff0c\u76f4\u63a5\u8f93\u51fa\uff0c\u4f1a\u5360\u7528\u7a7a\u95f4\u5f88\u5927\u3002<\/p>\n\n
            python3 -m gdal_merge -ot Byte -of GTiff -o F:\/\u66f9\u53bf666.tif<\/code><\/p>\n\n
            \u7248\u672c2<\/h3>\n\n
            \u8f93\u51fa\u7ed3\u679c\u505alzw\u538b\u7f29\uff0c\u53ef\u4ee5\u6709\u6548\u8282\u7701\u7a7a\u95f4\u3002<\/p>\n\n
            \u8bb8\u591a\u4eba\u6ca1\u6709\u610f\u8bc6\u5230\uff0c\u538b\u7f29\u65b9\u6848\u5728\u5f88\u5927\u7a0b\u5ea6\u4e0a\u53d6\u51b3\u4e8e\u8981\u5904\u7406\u7684\u6570\u636e\uff0c\u5728\u4e00\u4e2a\u6570\u636e\u4e0a\u8fd0\u884c\u826f\u597d\u7684\u65b9\u6848\u53ef\u80fd\u4f1a\u5728\u4e0b\u4e00\u4e2a\u5f88\u5dee\u3002<\/p>\n\n
            \u4f46\u662f\u7531\u4e8e\u6211\u4eec\u4e0d\u60f3\u7ed9\u4e16\u754c\u8d1f\u62c5\u592a\u591a\u538b\u7f29\u65b9\u6848\uff0c\u9700\u8981\u4e00\u79cd\u81ea\u9002\u5e94\u65b9\u6848\uff0c\u4e0d\u6c42\u6700\u597d\uff0c\u4f46\u6c42\u5728\u7edd\u5927\u591a\u6570\u56fe\u50cf\u4e0a\u8868\u73b0\u7684\u4e0d\u9519\uff0cLZW\u5177\u6709\u8fd9\u6837\u7684\u54c1\u8d28\u3002<\/p>\n\n
            python3 -m gdal_merge -ot Byte -of GTiff -co COMPRESS=LZW  -o F:\/\u66f9\u53bf666.tif<\/code><\/p>\n\n
            \u7248\u672c3<\/h3>\n\n
            \u67d0\u4e9b\u56fe\u50cf\u9996\u5148\u7ecf\u8fc7\u4e00\u4e2a\u5904\u7406\uff0c\u5176\u4e2d\u6bcf\u4e2a\u50cf\u7d20\u503c\u90fd\u7528\u50cf\u7d20\u548c\u524d\u4e00\u4e2a\u50cf\u7d20\u4e4b\u95f4\u7684\u5dee\u503c\u4ee3\u66ff\uff0c\u5219\u53ef\u4ee5\u4f7f\u7528LZW\u7f16\u7801\u66f4\u597d\u5730\u538b\u7f29\u5b83\u4eec\uff0c\u5728\u4e8c\u7ef4\u4e0a\u6267\u884c\u8fd9\u79cd\u5dee\u503c\u8ba1\u7b97\u53ef\u4ee5\u5e2e\u52a9\u66f4\u591a\u56fe\u50cf\u3002<\/p>\n\n
            \u8fd9\u5c31\u662fLZW\u538b\u7f29\u4e2d\u7684PREDICTOR\u7b97\u6cd5\u3002<\/p>\n\n
            \u4f46\u662f\uff0c\u4f7f\u7528\u8fd9\u79cd\u989d\u5916\u7684\u9884\u5904\u7406\u540e\uff0c\u8bb8\u591a\u56fe\u50cf\u7684\u538b\u7f29\u6548\u679c\u5e76\u4e0d\u7406\u60f3\uff0c\u5e76\u4e14\u5bf9\u4e8e\u5927\u91cf\u56fe\u50cf\u800c\u8a00\uff0c\u538b\u7f29\u7387\u5b9e\u9645\u4e0a\u66f4\u5dee\u3002\u56e0\u6b64\uff0c\u9ed8\u8ba4PREDICTOR\u503c\u4e3a1\u5373\u4e0d\u505a\u8fd9\u4e2a\u9884\u5904\u7406\u3002<\/p>\n\n
            \u5b9e\u9a8c\u4e2d\u5c1d\u8bd5\u4e86PREDICTOR=2\u538b\u7f29\u6548\u679c\u63d0\u5347\u4e0d\u5927\u3002<\/p>\n\n
            python3 -m gdal_merge -ot Byte -of GTiff -co COMPRESS=LZW -co PREDICTOR=2 -o F:\/\u66f9\u53bf666.tif<\/code><\/p>\n\n
            \u7248\u672c4<\/h3>\n\n
            Tiled Multi-Resolution (or Tiled Pyramidal) TIFF\u662f\u7b80\u5355\u7684\u591a\u9875\u5206\u5e45 TIFF \u56fe\u50cf, \u6bcf\u4e00\u79cd\u5206\u8fa8\u7387\u5b58\u50a8\u4e00\u4e2a\u56fe\u5c42\u3002<\/p>\n\n
            \u8fd9\u662f\u4e00\u79cd\u6807\u51c6\u7684TIFF\u6269\u5c55\uff0c\u5f88\u591a\u8f6f\u4ef6\u90fd\u652f\u6301\u3002libtiff\u51fd\u6570\u5e93\u4e5f\u652f\u6301\u8fd9\u79cdtiff\u7684\u8bfb\u5199\uff0c\u56e0\u6b64\u53ef\u4ee5\u5728gdal\u4e2d\u4f7f\u7528\u3002<\/p>\n\n
            Tiled Pyramidal TIFF \u56fe\u50cf \u4e5f\u53ef\u4ee5\u4f7f\u7528 LZW\u8fd9\u6837\u7684\u65e0\u635f\u538b\u7f29\u6216\u8005JPEG\u8fd9\u6837\u7684\u65e0\u635f\u538b\u7f29\u3002<\/p>\n\n
            \u5b9e\u9a8c\u4e2d\u6dfb\u52a0\u4e86TILED<\/code>\u6807\u7b7e\u540e\u538b\u7f29\u6bd4\u663e\u8457\u63d0\u5347\u3002<\/p>\n\n
            python3 -m gdal_merge -ot Byte -of GTiff -co COMPRESS=LZW -co PREDICTOR=2 -co TILED=YES -o F:\/\u66f9\u53bf666.tif<\/code><\/p>\n\n
            3 \u4ee3\u7801\u4e2d\u4f7f\u7528\u2013\u538b\u7f29\u8f93\u51fa<\/h1>\n\n
            \u4e0b\u9762\u662f\u4e00\u4e2a\u5c01\u88c5\u7684\u538b\u7f29\u8f93\u51fa\u7684\u4f8b\u5b50\uff0c\u76f4\u63a5\u8c03\u7528gdal\u7684\u5305\u51fd\u6570Warp()\uff0c\u8fd8\u652f\u6301\u591a\u7ebf\u7a0b\u5904\u7406<\/p>\n\n
            import<\/span> gdal<\/span>\n\ndef<\/span> compressTifLZW<\/span>(<\/span>outraster<\/span>,<\/span> inraster<\/span>):<\/span>\n    options<\/span>=<\/span>gdal<\/span>.<\/span>WarpOptions<\/span>(<\/span>\n            creationOptions<\/span>=<\/span>[<\/span>\"COMPRESS=LZW\"<\/span>],<\/span> # creationOptions\u4e4b\u538b\u7f29\u9009\u9879\u8bbe\u7f6e\n<\/span>            multithread<\/span>=<\/span>True<\/span> # \u591a\u7ebf\u7a0b\n<\/span>            )<\/span>\n    gdal<\/span>.<\/span>Warp<\/span>(<\/span>outraster<\/span>,<\/span> inraster<\/span>,<\/span> options<\/span>=<\/span>options<\/span>)<\/span>\n\n<\/code><\/pre><\/div><\/div>\n\n
            \u5982\u679c\u60f3\u5728\u811a\u672c\u4e2d\u4f7f\u7528\u8fd9\u4e2a\u529f\u80fd\uff0c\u5728-co\u53c2\u6570\u6307\u5b9a\u538b\u7f29\u5373\u53ef<\/p>\n\n
            4 \u53c2\u8003\u8d44\u6599<\/h1>\n\n
            https:\/\/gis.stackexchange.com\/a\/258215\/88333<\/p>\n\n
            https:\/\/forum.xnview.com\/viewtopic.php?p=104569&sid=839f082ea7e26ca900a7b5db75ae9ad4#p104569<\/p>\n\n
            https:\/\/www.fileformat.info\/format\/tiff\/corion-lzw.htm<\/p>\n\n
            https:\/\/gdal.org\/programs\/gdal_merge.html<\/p>\n\n
            https:\/\/gdal.org\/programs\/gdalwarp.html<\/p>\n\n
            https:\/\/gdal.org\/drivers\/raster\/gtiff.html<\/p>","author":{"name":{}},"summary":"1 \u7b80\u4ecb"},{"title":"\u3010ai\u5168\u6808\u3011\u4e0d\u540c\u5e73\u53f0\u4e0bgcc\u7f16\u8bd1\u5b89\u88c5","link":{"@attributes":{"href":"https:\/\/dorbodwolf.github.io\/\/2021\/01\/18\/AI%E5%85%A8%E6%A0%88-%E4%B8%8D%E5%90%8C%E5%B9%B3%E5%8F%B0%E4%B8%8Bgcc%E7%BC%96%E8%AF%91%E5%AE%89%E8%A3%85.html","rel":"alternate","type":"text\/html","title":"\u3010ai\u5168\u6808\u3011\u4e0d\u540c\u5e73\u53f0\u4e0bgcc\u7f16\u8bd1\u5b89\u88c5"}},"published":"2021-01-18T00:00:00+00:00","updated":"2021-01-18T00:00:00+00:00","id":"https:\/\/dorbodwolf.github.io\/\/2021\/01\/18\/%20%E3%80%90AI%E5%85%A8%E6%A0%88%E3%80%91%E4%B8%8D%E5%90%8C%E5%B9%B3%E5%8F%B0%E4%B8%8Bgcc%E7%BC%96%E8%AF%91%E5%AE%89%E8%A3%85","content":"
            \u672c\u6587\u7684\u76ee\u6807\u662f\u5b89\u88c52010\u5e74\u53d1\u5e03\u7684GCC 4.4.5\u7248\u672c\uff0c\u8fd9\u662f\u56e0\u4e3a\u5f88\u591aOJ\u5e73\u53f0\u8fd8\u662f\u6cbf\u7528\u8fd9\u4e2a\u8001\u7248\u672c\uff0c\u65b9\u4fbf\u7ebf\u4e0b\u6d4b\u8bd5\u4ee3\u7801\u3002<\/p>\n\n
            GCC 4.8.1\u662f\u7b2c\u4e00\u4e2a\u5b8c\u6574\u5305\u542b2011 C++\u6807\u51c6\u7684\u7248\u672c\uff0c\u7f16\u8bd1\u65f6\u5019\u901a\u8fc7-std=c++11\u6765\u8bbe\u7f6eflag\u652f\u6301C++ 11\u6807\u51c6\u3002<\/p>\n\n
            linux<\/h1>\n\n
            linux\u4e0b\u7f16\u8bd1\u5b89\u88c5\u53ef\u4ee5\u7075\u6d3b\u914d\u7f6e\uff0c\u5177\u4f53\u53ef\u4ee5\u53c2\u8003GNU\u4e3b\u9875\u4e0a\u7684\u6559\u7a0b\u3002\u5982\u679c\u60f3\u8981\u5feb\u901f\u5b89\u88c5\u76f4\u63a5\u53c2\u8003\u8fd9\u7bc7\u5b98\u65b9wiki<\/a>\u3002<\/p>\n\n
            \u7f16\u8bd1\u6280\u5de7\uff1a<\/p>\n\n