{"id":974091,"date":"2024-12-27T06:02:20","date_gmt":"2024-12-26T22:02:20","guid":{"rendered":"https:\/\/docs.pingcode.com\/ask\/ask-ask\/974091.html"},"modified":"2024-12-27T06:02:22","modified_gmt":"2024-12-26T22:02:22","slug":"%e5%a6%82%e4%bd%95%e4%bd%bf%e7%94%a8python%e8%ae%a1%e7%ae%97oks","status":"publish","type":"post","link":"https:\/\/docs.pingcode.com\/ask\/974091.html","title":{"rendered":"\u5982\u4f55\u4f7f\u7528python\u8ba1\u7b97oks"},"content":{"rendered":"

\"\u5982\u4f55\u4f7f\u7528python\u8ba1\u7b97oks\"<\/p>\n

\u4f7f\u7528Python\u8ba1\u7b97OKS\uff08Object Keypoint Similarity\uff09\u9700\u8981\u7406\u89e3\u5173\u952e\u70b9\u5339\u914d\u3001\u6743\u91cd\u5206\u914d\u3001\u4ee5\u53ca\u5c3a\u5ea6\u89c4\u8303\u5316\u7b49\u6b65\u9aa4\u3002\u9996\u5148\uff0c\u8ba1\u7b97\u5173\u952e\u70b9\u7684\u6b27\u5f0f\u8ddd\u79bb\u3001\u7136\u540e\u5e94\u7528\u5c3a\u5ea6\u89c4\u8303\u5316\u3001\u6700\u540e\u52a0\u6743\u5e73\u5747\u5f97\u5230OKS\u503c\u3002\u63a5\u4e0b\u6765\u5c06\u8be6\u7ec6\u4ecb\u7ecd\u8fd9\u4e9b\u6b65\u9aa4\u4e2d\u7684\u4e00\u4e2a\uff1a\u8ba1\u7b97\u5173\u952e\u70b9\u7684\u6b27\u5f0f\u8ddd\u79bb\u3002\u8fd9\u4e00\u6b65\u901a\u8fc7\u6bd4\u8f83\u9884\u6d4b\u5173\u952e\u70b9\u4e0e\u771f\u5b9e\u5173\u952e\u70b9\u4e4b\u95f4\u7684\u8ddd\u79bb\u6765\u8bc4\u4f30\u5339\u914d\u7a0b\u5ea6\u3002<\/strong><\/p>\n<\/p>\n

\u4e00\u3001\u7406\u89e3OKS\u7684\u57fa\u672c\u6982\u5ff5<\/p>\n<\/p>\n

OKS\uff08Object Keypoint Similarity\uff09\u662f\u4e00\u79cd\u7528\u4e8e\u8bc4\u4f30\u68c0\u6d4b\u4efb\u52a1\u4e2d\u5173\u952e\u70b9\u9884\u6d4b\u7cbe\u5ea6\u7684\u6307\u6807\uff0c\u5e38\u7528\u4e8e\u59ff\u6001\u4f30\u8ba1\u7b49\u4efb\u52a1\u4e2d\u3002\u5b83\u901a\u8fc7\u6bd4\u8f83\u9884\u6d4b\u7684\u5173\u952e\u70b9\u4f4d\u7f6e\u548c\u771f\u5b9e\u7684\u5173\u952e\u70b9\u4f4d\u7f6e\u4e4b\u95f4\u7684\u8ddd\u79bb\uff0c\u7ed3\u5408\u5c3a\u5ea6\u56e0\u7d20\u548c\u5173\u952e\u70b9\u7684\u91cd\u8981\u6027\u6765\u8ba1\u7b97\u76f8\u4f3c\u6027\u5f97\u5206\u3002<\/p>\n<\/p>\n

    \n
  1. \n

    \u5173\u952e\u70b9\u5339\u914d<\/strong><\/p>\n<\/p>\n

    \u5728\u59ff\u6001\u4f30\u8ba1\u4efb\u52a1\u4e2d\uff0c\u6bcf\u4e2a\u76ee\u6807\u901a\u5e38\u6709\u591a\u4e2a\u5173\u952e\u70b9\uff0c\u6bd4\u5982\u4eba\u4f53\u7684\u5173\u8282\u70b9\u3002OKS\u4e3b\u8981\u7528\u4e8e\u8bc4\u4f30\u8fd9\u4e9b\u5173\u952e\u70b9\u7684\u9884\u6d4b\u6548\u679c\u3002OKS\u7684\u6838\u5fc3\u601d\u60f3\u662f\u6839\u636e\u9884\u6d4b\u548c\u771f\u5b9e\u5173\u952e\u70b9\u4e4b\u95f4\u7684\u8ddd\u79bb\u6765\u8ba1\u7b97\u76f8\u4f3c\u6027\u5f97\u5206\uff0c\u56e0\u6b64\u5339\u914d\u5173\u952e\u70b9\u662f\u8ba1\u7b97OKS\u7684\u57fa\u7840\u3002<\/p>\n<\/p>\n<\/li>\n

  2. \n

    \u5c3a\u5ea6\u89c4\u8303\u5316<\/strong><\/p>\n<\/p>\n

    \u7531\u4e8e\u4e0d\u540c\u76ee\u6807\u7684\u5927\u5c0f\u4e0d\u540c\uff0c\u76f4\u63a5\u6bd4\u8f83\u5173\u952e\u70b9\u7684\u8ddd\u79bb\u53ef\u80fd\u4f1a\u5bfc\u81f4\u8bef\u5dee\uff0c\u56e0\u6b64OKS\u5728\u8ba1\u7b97\u65f6\u4f1a\u4f7f\u7528\u76ee\u6807\u7684\u5c3a\u5ea6\u8fdb\u884c\u89c4\u8303\u5316\u3002\u8fd9\u6837\u53ef\u4ee5\u6709\u6548\u5730\u6d88\u9664\u76ee\u6807\u5927\u5c0f\u5bf9\u76f8\u4f3c\u6027\u8bc4\u4f30\u7684\u5f71\u54cd\u3002<\/p>\n<\/p>\n<\/li>\n

  3. \n

    \u6743\u91cd\u5206\u914d<\/strong><\/p>\n<\/p>\n

    \u5728OKS\u8ba1\u7b97\u4e2d\uff0c\u4e0d\u540c\u5173\u952e\u70b9\u7684\u91cd\u8981\u6027\u53ef\u80fd\u4e0d\u540c\uff0c\u56e0\u6b64\u4f1a\u5bf9\u6bcf\u4e2a\u5173\u952e\u70b9\u5206\u914d\u6743\u91cd\u3002\u6743\u91cd\u7684\u8bbe\u7f6e\u901a\u5e38\u6839\u636e\u4efb\u52a1\u9700\u6c42\u548c\u7ecf\u9a8c\u8fdb\u884c\u8c03\u6574\uff0c\u4ee5\u4fbf\u66f4\u51c6\u786e\u5730\u8bc4\u4f30\u5173\u952e\u70b9\u7684\u5339\u914d\u6548\u679c\u3002<\/p>\n<\/p>\n<\/li>\n<\/ol>\n

    \u4e8c\u3001\u5b9e\u73b0OKS\u8ba1\u7b97\u7684\u6b65\u9aa4<\/p>\n<\/p>\n

      \n
    1. \n

      \u8ba1\u7b97\u5173\u952e\u70b9\u7684\u6b27\u5f0f\u8ddd\u79bb<\/strong><\/p>\n<\/p>\n

      \u5173\u952e\u70b9\u7684\u6b27\u5f0f\u8ddd\u79bb\u662fOKS\u8ba1\u7b97\u7684\u57fa\u7840\u3002\u5bf9\u4e8e\u6bcf\u4e00\u5bf9\u9884\u6d4b\u548c\u771f\u5b9e\u7684\u5173\u952e\u70b9\uff0c\u8ba1\u7b97\u5176\u5728\u4e8c\u7ef4\u5e73\u9762\u4e0a\u7684\u6b27\u5f0f\u8ddd\u79bb\u3002\u8fd9\u4e00\u6b65\u53ef\u4ee5\u901a\u8fc7\u4ee5\u4e0b\u516c\u5f0f\u5b9e\u73b0\uff1a<\/p>\n<\/p>\n

      [<\/p>\n

      d = \\sqrt{(x_{pred} – x_{true})^2 + (y_{pred} – y_{true})^2}<\/p>\n

      ]<\/p>\n<\/p>\n

      \u5176\u4e2d\uff0c(x_{pred}, y_{pred}) \u548c (x_{true}, y_{true}) \u5206\u522b\u662f\u9884\u6d4b\u548c\u771f\u5b9e\u5173\u952e\u70b9\u7684\u5750\u6807\u3002<\/p>\n<\/p>\n<\/li>\n

    2. \n

      \u5c3a\u5ea6\u89c4\u8303\u5316<\/strong><\/p>\n<\/p>\n

      \u8ba1\u7b97\u51fa\u7684\u6b27\u5f0f\u8ddd\u79bb\u9700\u8981\u8fdb\u884c\u5c3a\u5ea6\u89c4\u8303\u5316\uff0c\u4ee5\u6d88\u9664\u76ee\u6807\u5927\u5c0f\u7684\u5f71\u54cd\u3002\u89c4\u8303\u5316\u901a\u5e38\u4f1a\u4f7f\u7528\u76ee\u6807\u7684\u9762\u79ef\u3001\u5468\u957f\u6216\u5176\u4ed6\u7279\u5f81\u4f5c\u4e3a\u53c2\u8003\u503c\u3002<\/p>\n<\/p>\n<\/li>\n

    3. \n

      \u5e94\u7528\u6743\u91cd<\/strong><\/p>\n<\/p>\n

      \u5bf9\u6bcf\u4e2a\u5173\u952e\u70b9\u7684\u8ddd\u79bb\u8fdb\u884c\u52a0\u6743\u5e73\u5747\uff0c\u6743\u91cd\u7684\u8bbe\u7f6e\u53ef\u4ee5\u6839\u636e\u5173\u952e\u70b9\u7684\u91cd\u8981\u6027\u8fdb\u884c\u8c03\u6574\u3002\u6743\u91cd\u8d8a\u5927\u7684\u5173\u952e\u70b9\u5bf9\u6700\u7ec8OKS\u503c\u7684\u5f71\u54cd\u4e5f\u8d8a\u5927\u3002<\/p>\n<\/p>\n<\/li>\n<\/ol>\n

      \u4e09\u3001Python\u5b9e\u73b0OKS\u8ba1\u7b97<\/p>\n<\/p>\n

      \u4e0b\u9762\u662f\u4e00\u4e2a\u4f7f\u7528Python\u5b9e\u73b0OKS\u8ba1\u7b97\u7684\u57fa\u672c\u793a\u4f8b\uff1a<\/p>\n<\/p>\n

      import numpy as np<\/p>\n
      def calculate_oks(pred_keypoints, true_keypoints, scale, sigmas):<\/p>\n
          """<\/p>\n
          \u8ba1\u7b97OKS\u503c<\/p>\n
          :param pred_keypoints: \u9884\u6d4b\u5173\u952e\u70b9\u5750\u6807\u6570\u7ec4 (n, 2)<\/p>\n
          :param true_keypoints: \u771f\u5b9e\u5173\u952e\u70b9\u5750\u6807\u6570\u7ec4 (n, 2)<\/p>\n
          :param scale: \u7528\u4e8e\u89c4\u8303\u5316\u7684\u5c3a\u5ea6\u503c<\/p>\n
          :param sigmas: \u5173\u952e\u70b9\u7684\u6807\u51c6\u5dee\u6570\u7ec4<\/p>\n
          :return: OKS\u503c<\/p>\n
          """<\/p>\n
          assert pred_keypoints.shape == true_keypoints.shape<\/p>\n
          # \u8ba1\u7b97\u6b27\u5f0f\u8ddd\u79bb<\/p>\n
          distances = np.linalg.norm(pred_keypoints - true_keypoints, axis=1)<\/p>\n
          # \u8ba1\u7b97OKS<\/p>\n
          oks = np.exp(-distances<strong>2 \/ (2 * (scale * sigmas)<\/strong>2))<\/p>\n
          return np.mean(oks)<\/p>\n
      \u793a\u4f8b\u4f7f\u7528<\/strong><\/h2>\n
      pred_keypoints = np.array([[100, 200], [150, 250], [200, 300]])<\/p>\n
      true_keypoints = np.array([[110, 210], [145, 255], [190, 290]])<\/p>\n
      scale = 0.5<\/p>\n
      sigmas = np.array([0.1, 0.1, 0.1])<\/p>\n
      oks_value = calculate_oks(pred_keypoints, true_keypoints, scale, sigmas)<\/p>\n
      print("OKS Value:", oks_value)<\/p>\n
      <\/code><\/pre>\n<\/p>\n
      \u56db\u3001\u5e94\u7528\u4e0e\u4f18\u5316\u5efa\u8bae<\/p>\n<\/p>\n
        \n
      1. \n
        \u8c03\u6574\u5c3a\u5ea6\u548c\u6743\u91cd<\/strong><\/p>\n<\/p>\n
        \u6839\u636e\u5177\u4f53\u7684\u4efb\u52a1\u9700\u6c42\uff0c\u53ef\u4ee5\u5bf9\u5c3a\u5ea6\u548c\u6743\u91cd\u8fdb\u884c\u8c03\u6574\u3002\u4f8b\u5982\uff0c\u5bf9\u4e8e\u4e00\u4e9b\u5173\u952e\u70b9\u975e\u5e38\u91cd\u8981\u7684\u4efb\u52a1\uff0c\u53ef\u4ee5\u589e\u52a0\u8fd9\u4e9b\u5173\u952e\u70b9\u7684\u6743\u91cd\uff0c\u4ee5\u63d0\u9ad8\u8bc4\u4f30\u7684\u51c6\u786e\u6027\u3002<\/p>\n<\/p>\n<\/li>\n
      2. \n
        \u6570\u636e\u9884\u5904\u7406<\/strong><\/p>\n<\/p>\n
        \u5728\u8ba1\u7b97OKS\u4e4b\u524d\uff0c\u786e\u4fdd\u6570\u636e\u5df2\u7ecf\u8fc7\u9002\u5f53\u7684\u9884\u5904\u7406\uff0c\u5305\u62ec\u5750\u6807\u5f52\u4e00\u5316\u3001\u6570\u636e\u6e05\u6d17\u7b49\uff0c\u4ee5\u63d0\u9ad8\u8ba1\u7b97\u7ed3\u679c\u7684\u7a33\u5b9a\u6027\u548c\u51c6\u786e\u6027\u3002<\/p>\n<\/p>\n<\/li>\n
      3. \n
        \u4f7f\u7528\u66f4\u590d\u6742\u7684\u6a21\u578b<\/strong><\/p>\n<\/p>\n
        \u5bf9\u4e8e\u5927\u89c4\u6a21\u7684\u59ff\u6001\u4f30\u8ba1\u4efb\u52a1\uff0c\u53ef\u80fd\u9700\u8981\u4f7f\u7528\u66f4\u52a0\u590d\u6742\u7684\u6a21\u578b\u548c\u7b97\u6cd5\u6765\u83b7\u5f97\u66f4\u9ad8\u7684\u7cbe\u5ea6\u3002\u5728\u8fd9\u79cd\u60c5\u51b5\u4e0b\uff0cOKS\u7684\u8ba1\u7b97\u53ef\u80fd\u9700\u8981\u7ed3\u5408\u5176\u4ed6\u8bc4\u4f30\u6307\u6807\u8fdb\u884c\u7efc\u5408\u5206\u6790\u3002<\/p>\n<\/p>\n<\/li>\n<\/ol>\n
        \u4e94\u3001\u603b\u7ed3<\/p>\n<\/p>\n
        OKS\u662f\u4e00\u79cd\u91cd\u8981\u7684\u8bc4\u4f30\u6307\u6807\uff0c\u5c24\u5176\u5728\u59ff\u6001\u4f30\u8ba1\u7b49\u8ba1\u7b97\u673a\u89c6\u89c9\u4efb\u52a1\u4e2d\u3002\u901a\u8fc7\u7406\u89e3\u5176\u8ba1\u7b97\u539f\u7406\u548c\u5b9e\u73b0\u6b65\u9aa4\uff0c\u53ef\u4ee5\u5e2e\u52a9\u7814\u7a76\u4eba\u5458\u66f4\u597d\u5730\u8bc4\u4f30\u548c\u6539\u8fdb\u5173\u952e\u70b9\u68c0\u6d4b\u6a21\u578b\u7684\u6027\u80fd\u3002\u5728\u5b9e\u9645\u5e94\u7528\u4e2d\uff0cOKS\u7684\u8ba1\u7b97\u8fd8\u53ef\u4ee5\u7ed3\u5408\u5176\u4ed6\u6307\u6807\uff0c\u5982PCK\uff08Percentage of Correct Keypoints\uff09\u7b49\uff0c\u4ee5\u63d0\u4f9b\u66f4\u5168\u9762\u7684\u6a21\u578b\u8bc4\u4f30\u7ed3\u679c\u3002<\/p>\n<\/p>\n
        \u76f8\u5173\u95ee\u7b54FAQs\uff1a<\/strong><\/h2>\n
         \u5982\u4f55\u7528Python\u8ba1\u7b97OKS\u7684\u57fa\u672c\u6b65\u9aa4\u662f\u4ec0\u4e48\uff1f<\/strong>
        \u8ba1\u7b97OKS\uff08Object Keypoint Similarity\uff09\u901a\u5e38\u6d89\u53ca\u5bf9\u76ee\u6807\u68c0\u6d4b\u6216\u5173\u952e\u70b9\u68c0\u6d4b\u7684\u7ed3\u679c\u8fdb\u884c\u8bc4\u4f30\u3002\u9996\u5148\uff0c\u9700\u8981\u5b89\u88c5\u76f8\u5173\u7684\u5e93\uff0c\u5982NumPy\u548cOpenCV\u3002\u63a5\u7740\uff0c\u5c06\u9884\u6d4b\u7ed3\u679c\u4e0e\u771f\u5b9e\u6807\u7b7e\u8fdb\u884c\u6bd4\u5bf9\uff0c\u8ba1\u7b97\u6bcf\u4e2a\u5173\u952e\u70b9\u7684\u8ddd\u79bb\uff0c\u63a5\u7740\u6839\u636e\u76f8\u4f3c\u5ea6\u8fdb\u884c\u52a0\u6743\uff0c\u6700\u7ec8\u5f97\u51faOKS\u503c\u3002\u5177\u4f53\u6b65\u9aa4\u5305\u62ec\uff1a\u8bfb\u53d6\u6570\u636e\u3001\u8ba1\u7b97\u5173\u952e\u70b9\u4e4b\u95f4\u7684\u8ddd\u79bb\u3001\u5e94\u7528\u76f8\u4f3c\u5ea6\u516c\u5f0f\u7b49\u3002<\/p>\n
        \u5728Python\u4e2d\u8ba1\u7b97OKS\u65f6\uff0c\u5982\u4f55\u5904\u7406\u5173\u952e\u70b9\u7f3a\u5931\u7684\u60c5\u51b5\uff1f<\/strong>
        \u5728\u8ba1\u7b97OKS\u65f6\uff0c\u7f3a\u5931\u7684\u5173\u952e\u70b9\u4f1a\u5f71\u54cd\u6700\u7ec8\u7684\u76f8\u4f3c\u5ea6\u7ed3\u679c\u3002\u4e3a\u4e86\u5904\u7406\u8fd9\u4e00\u60c5\u51b5\uff0c\u53ef\u4ee5\u91c7\u7528\u4e00\u4e9b\u7b56\u7565\uff0c\u4f8b\u5982\u5728\u8ba1\u7b97\u65f6\u5bf9\u7f3a\u5931\u7684\u5173\u952e\u70b9\u8fdb\u884c\u6743\u91cd\u8c03\u6574\uff0c\u6216\u8005\u5728\u8ba1\u7b97OKS\u65f6\u53ea\u8003\u8651\u90a3\u4e9b\u5b58\u5728\u7684\u5173\u952e\u70b9\u3002\u901a\u8fc7\u8fd9\u79cd\u65b9\u6cd5\uff0c\u53ef\u4ee5\u786e\u4fdd\u8ba1\u7b97\u7ed3\u679c\u66f4\u4e3a\u51c6\u786e\uff0c\u907f\u514d\u56e0\u7f3a\u5931\u6570\u636e\u800c\u5bfc\u81f4\u7684\u504f\u5dee\u3002<\/p>\n
        OKS\u8ba1\u7b97\u7ed3\u679c\u7684\u8303\u56f4\u662f\u4ec0\u4e48\uff1f\u6211\u8be5\u5982\u4f55\u89e3\u8bfb\u8fd9\u4e9b\u7ed3\u679c\uff1f<\/strong>
        OKS\u7684\u503c\u901a\u5e38\u57280\u52301\u4e4b\u95f4\uff0c\u503c\u8d8a\u63a5\u8fd11\u8868\u793a\u68c0\u6d4b\u7ed3\u679c\u4e0e\u771f\u5b9e\u6807\u7b7e\u8d8a\u76f8\u4f3c\u3002\u5177\u4f53\u6765\u8bf4\uff0c0.5\u7684OKS\u503c\u901a\u5e38\u88ab\u8ba4\u4e3a\u662f\u4e00\u4e2a\u5408\u7406\u7684\u9608\u503c\uff0c\u8868\u793a\u6a21\u578b\u5728\u8be5\u5173\u952e\u70b9\u68c0\u6d4b\u4e0a\u8868\u73b0\u5408\u683c\u3002\u901a\u8fc7\u5206\u6790\u4e0d\u540c\u7684OKS\u503c\uff0c\u53ef\u4ee5\u8bc4\u4f30\u6a21\u578b\u5728\u4e0d\u540c\u6761\u4ef6\u4e0b\u7684\u8868\u73b0\uff0c\u5e2e\u52a9\u4f18\u5316\u6a21\u578b\u53c2\u6570\u6216\u8bad\u7ec3\u6570\u636e\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"\u4f7f\u7528Python\u8ba1\u7b97OKS\uff08Object Keypoint Similarity\uff09\u9700\u8981\u7406\u89e3\u5173\u952e\u70b9\u5339\u914d\u3001\u6743\u91cd\u5206\u914d […]","protected":false},"author":3,"featured_media":974098,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[37],"tags":[],"acf":[],"_links":{"self":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/974091"}],"collection":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/comments?post=974091"}],"version-history":[{"count":"1","href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/974091\/revisions"}],"predecessor-version":[{"id":974101,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/posts\/974091\/revisions\/974101"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/media\/974098"}],"wp:attachment":[{"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/media?parent=974091"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/categories?post=974091"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/docs.pingcode.com\/wp-json\/wp\/v2\/tags?post=974091"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}