{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"CLS 07 · 检测几何、匹配与评价","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html","markdown":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.md","pages":[{"id":"reading-opening","title":"CLS 07 · 检测几何、匹配与评价","unit":"CLS 07 · 检测几何、匹配与评价","unitId":"cls08-检测几何匹配与评价","chapter":"course","number":1},{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","unit":"CLS 07 · 检测几何、匹配与评价","unitId":"cls08-检测几何匹配与评价","chapter":"course","number":2},{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","unit":"CLS 07 · 检测几何、匹配与评价","unitId":"cls08-检测几何匹配与评价","chapter":"course","number":3}],"aliases":[{"oldId":"b-71e8c366ad156f7a17fa16f9","targetId":"b-71e8c366ad156f7a17fa16f9","operation":"historical-explanation","sourceScope":"block-8f46ef44ed4e4529beedf33817f14d3f"},{"oldId":"b-b75015e77d041ef2fa20f13e","targetId":"b-b75015e77d041ef2fa20f13e","operation":"historical-explanation","sourceScope":"block-8f46ef44ed4e4529beedf33817f14d3f"},{"oldId":"b-73cd275977d3f236486bd96e","targetId":"b-73cd275977d3f236486bd96e","operation":"historical-explanation","sourceScope":"block-8f46ef44ed4e4529beedf33817f14d3f"},{"oldId":"b-5f33c3f2f2819707dd9de5f3","targetId":"b-5f33c3f2f2819707dd9de5f3","operation":"historical-explanation","sourceScope":"block-8f46ef44ed4e4529beedf33817f14d3f"},{"oldId":"b-affe83e634aec87ec35c794f","targetId":"b-affe83e634aec87ec35c794f","page":"reading-opening","operation":"move","beforeSHA256":"5d6e7b33e278ad7f7a59a8a116727d408295a4963580ecb0cc72f0b51184d970","afterSHA256":"5d6e7b33e278ad7f7a59a8a116727d408295a4963580ecb0cc72f0b51184d970","reason":"Same source body in authored semantic page"},{"oldId":"b-3e5831dd77a73a3aa533b5e2","targetId":"b-3e5831dd77a73a3aa533b5e2","page":"reading-opening","operation":"move","beforeSHA256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","afterSHA256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","reason":"Same source body in authored semantic page"},{"oldId":"b-4a192b0514374d5df100182a","targetId":"b-4a192b0514374d5df100182a","page":"reading-opening","operation":"move","beforeSHA256":"88216e858779670a15bd42f20a06f136275ceee6a8685216f5821af4b1faf48f","afterSHA256":"88216e858779670a15bd42f20a06f136275ceee6a8685216f5821af4b1faf48f","reason":"Same source body in authored semantic page"},{"oldId":"b-08b237d474ea6fe9c89ad57c","targetId":"b-08b237d474ea6fe9c89ad57c","page":"reading-opening","operation":"move","beforeSHA256":"f22152fd55dcdb0090b5653d37145694fd19388d31e10226549464fe8b483578","afterSHA256":"f22152fd55dcdb0090b5653d37145694fd19388d31e10226549464fe8b483578","reason":"Same source body in authored semantic page"},{"oldId":"b-18e139e5d97eaba63fe7c737","targetId":"b-18e139e5d97eaba63fe7c737","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"93899883c4c044ee5171cc6d4515e67b10961c3d3e67eb507245ef30c5e625ce","afterSHA256":"93899883c4c044ee5171cc6d4515e67b10961c3d3e67eb507245ef30c5e625ce","reason":"Same source body in authored semantic page"},{"oldId":"b-059f9ebe54b6a66a5a81d0dc","targetId":"b-059f9ebe54b6a66a5a81d0dc","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"8560cc92f5ba742f1c60ed74ecfdb4342af5e340aef761676c2308d0b54f08bc","afterSHA256":"8560cc92f5ba742f1c60ed74ecfdb4342af5e340aef761676c2308d0b54f08bc","reason":"Same source body in authored semantic page"},{"oldId":"b-0f02087943e28c2d6cca375f","targetId":"b-0f02087943e28c2d6cca375f","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"d2c2e5495e53ef8a3f1ca587b170cd2deadae52ef91b4508655dad4f24e8253e","afterSHA256":"d2c2e5495e53ef8a3f1ca587b170cd2deadae52ef91b4508655dad4f24e8253e","reason":"Same source body in authored semantic page"},{"oldId":"b-4f7c40e926fa9bd65f967bc7","targetId":"b-4f7c40e926fa9bd65f967bc7","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"d075524e22eca70e822d1c51e8bd1fcec6ce17365859e6b7281d6ed011199d65","afterSHA256":"d075524e22eca70e822d1c51e8bd1fcec6ce17365859e6b7281d6ed011199d65","reason":"Same source body in authored semantic page"},{"oldId":"b-c78feaaba53e5517019cbbdb","targetId":"b-c78feaaba53e5517019cbbdb","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"dcd700347a57ddfff3e1b7965ed1d5138733ba426f6d770c0d667c4357403782","afterSHA256":"dcd700347a57ddfff3e1b7965ed1d5138733ba426f6d770c0d667c4357403782","reason":"Same source body in authored semantic page"},{"oldId":"b-80fa79a87a93bde84716d8a6","targetId":"b-80fa79a87a93bde84716d8a6","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"f242c7ee4328040c119d646550367cf3a3d8f45a7a7d5b769edcec25309f4975","afterSHA256":"f242c7ee4328040c119d646550367cf3a3d8f45a7a7d5b769edcec25309f4975","reason":"Same source body in authored semantic page"},{"oldId":"b-bb91601d0984113d77b094df","targetId":"b-bb91601d0984113d77b094df","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"5824845f2bd25d6b85eb2bedeb7fde8bbdc2adbd87bd830bf65b0450fb73702c","afterSHA256":"5824845f2bd25d6b85eb2bedeb7fde8bbdc2adbd87bd830bf65b0450fb73702c","reason":"Same source body in authored semantic page"},{"oldId":"b-4a5e4f005ff180e5f638b5c3","targetId":"b-4a5e4f005ff180e5f638b5c3","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"4069e14c5ad08dc619a5598db00977a4c0e0de429f00c9ecd0091eb7d58b8a1c","afterSHA256":"4069e14c5ad08dc619a5598db00977a4c0e0de429f00c9ecd0091eb7d58b8a1c","reason":"Same source body in authored semantic page"},{"oldId":"b-b0162f4941340f1772a93a21","targetId":"b-b0162f4941340f1772a93a21","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"0fe74dbe6635492abd693709fe30045ffdc781ab4cdbe8186bf03ddf8cbaf1fc","afterSHA256":"0fe74dbe6635492abd693709fe30045ffdc781ab4cdbe8186bf03ddf8cbaf1fc","reason":"Same source body in authored semantic page"},{"oldId":"b-7ac1f8a70ae52eac28ce1da6","targetId":"b-7ac1f8a70ae52eac28ce1da6","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"07781e6961e60ab655d9b67933a7916b6b48aaf8b55e0c0f2f00f89651cfd9a7","afterSHA256":"07781e6961e60ab655d9b67933a7916b6b48aaf8b55e0c0f2f00f89651cfd9a7","reason":"Same source body in authored semantic page"},{"oldId":"b-d96604e375b8a2fde3d8a3c0","targetId":"b-d96604e375b8a2fde3d8a3c0","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"e025ab4ea87ddd435bebe1613d16b6b5d848975ad5bc92e5256735cf2cb53561","afterSHA256":"e025ab4ea87ddd435bebe1613d16b6b5d848975ad5bc92e5256735cf2cb53561","reason":"Same source body in authored semantic page"},{"oldId":"b-dd5256ac58115ed6a0d5160e","targetId":"b-dd5256ac58115ed6a0d5160e","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"ad1f7ec36605770f96dde71af12439c5a535f3e77cb01cdaabb9a196ed5f5ff6","afterSHA256":"ad1f7ec36605770f96dde71af12439c5a535f3e77cb01cdaabb9a196ed5f5ff6","reason":"Same source body in authored semantic page"},{"oldId":"b-ce80e8e778aebeb5ed497f5a","targetId":"b-ce80e8e778aebeb5ed497f5a","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"b1a8121bb9b156ad2aebe396b09b8e6178ff5376ae3b0d39fd29ae93f53af5aa","afterSHA256":"b1a8121bb9b156ad2aebe396b09b8e6178ff5376ae3b0d39fd29ae93f53af5aa","reason":"Same source body in authored semantic page"},{"oldId":"b-8e7f196c91b64b90b1cf135e","targetId":"b-8e7f196c91b64b90b1cf135e","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"90b1e23eb5816cc872991feea47f4be353693efb37b7aabcb3cb413d74365fca","afterSHA256":"90b1e23eb5816cc872991feea47f4be353693efb37b7aabcb3cb413d74365fca","reason":"Same source body in authored semantic page"},{"oldId":"b-9ba22d63bb9aa7f560ed915f","targetId":"b-9ba22d63bb9aa7f560ed915f","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"9098789457bdb4955ec4f933dc2f9e3ba5366f7356882e102ee2309a78302040","afterSHA256":"9098789457bdb4955ec4f933dc2f9e3ba5366f7356882e102ee2309a78302040","reason":"Same source body in authored semantic page"},{"oldId":"b-e37758de87cc17697aab140d","targetId":"b-e37758de87cc17697aab140d","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"8e6b0c8407cefcaeb3b9a235195b40b038695cddb1c3461c819b19ec767f0650","afterSHA256":"8e6b0c8407cefcaeb3b9a235195b40b038695cddb1c3461c819b19ec767f0650","reason":"Same source body in authored semantic page"},{"oldId":"b-beebf75e778f4bef3a5f419a","targetId":"b-beebf75e778f4bef3a5f419a","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"39e9d99c41a61b16e0edc703daef9d3075d4e5c7b18009e2090e7a4dfc508443","afterSHA256":"39e9d99c41a61b16e0edc703daef9d3075d4e5c7b18009e2090e7a4dfc508443","reason":"Same source body in authored semantic page"},{"oldId":"b-9d7d0dd4d72116403e38b7d2","targetId":"b-9d7d0dd4d72116403e38b7d2","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"f2e92753771abbafbcb6b78618e928db0d86d334070ed63cd8ae3c084e0a19f5","afterSHA256":"f2e92753771abbafbcb6b78618e928db0d86d334070ed63cd8ae3c084e0a19f5","reason":"Same source body in authored semantic page"},{"oldId":"b-d76fae133d2523ee907cd725","targetId":"b-d76fae133d2523ee907cd725","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"40858ec0f600a17096cc7466a4f922b9540d5c1a169226cf8af91498a8a711e3","afterSHA256":"40858ec0f600a17096cc7466a4f922b9540d5c1a169226cf8af91498a8a711e3","reason":"Same source body in authored semantic page"},{"oldId":"b-84cf6883740bdc3d1748e68d","targetId":"b-84cf6883740bdc3d1748e68d","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"072c7fcb91f010db36447b5f7636ad481ddad26681f26397e6b7e7661508f502","afterSHA256":"072c7fcb91f010db36447b5f7636ad481ddad26681f26397e6b7e7661508f502","reason":"Same source body in authored semantic page"},{"oldId":"b-8656bd9b4bd2d200803bcbb7","targetId":"b-8656bd9b4bd2d200803bcbb7","page":"reading-block-9264aa76ab44410a94ba0c31731e50ef","operation":"move","beforeSHA256":"842f4305a9f2d8fe7fb3192c4b4b2ff23e7ab466c311abaad3889acebfb6dd58","afterSHA256":"842f4305a9f2d8fe7fb3192c4b4b2ff23e7ab466c311abaad3889acebfb6dd58","reason":"Same source body in authored semantic page"},{"oldId":"b-6de54f04bd7c726ca86f63db","targetId":"b-6de54f04bd7c726ca86f63db","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"72d3c7834c3fafe18571e3ac3fc55ec05967604ca8e1c604fc52313b1c6b033b","afterSHA256":"72d3c7834c3fafe18571e3ac3fc55ec05967604ca8e1c604fc52313b1c6b033b","reason":"Same source body in authored semantic page"},{"oldId":"b-741c18a315124e1fd3152db4","targetId":"b-741c18a315124e1fd3152db4","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"b117cf446d2ad20730ef6f594c83df176d7660e6471a02cd3f1b71e3e15edf9e","afterSHA256":"b117cf446d2ad20730ef6f594c83df176d7660e6471a02cd3f1b71e3e15edf9e","reason":"Same source body in authored semantic page"},{"oldId":"b-93bdc2a6600e835e5bc702df","targetId":"b-93bdc2a6600e835e5bc702df","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"315a10d85c66d1a4cc67a41fd3d34c08e2b14631a03e3d0174dc9b1de08132f4","afterSHA256":"315a10d85c66d1a4cc67a41fd3d34c08e2b14631a03e3d0174dc9b1de08132f4","reason":"Same source body in authored semantic page"},{"oldId":"b-3c41d9b855efb5e7618f0f81","targetId":"b-3c41d9b855efb5e7618f0f81","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"ac5d224648f4ebde30f06c9a67e260c09658b54cddb2f0ff694279ac674969bc","afterSHA256":"ac5d224648f4ebde30f06c9a67e260c09658b54cddb2f0ff694279ac674969bc","reason":"Same source body in authored semantic page"},{"oldId":"b-04ce59162ab0695e1069f8fc","targetId":"b-04ce59162ab0695e1069f8fc","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"fdf90e10bf1db95f074762e097e80c61cd82e960b2ac2808385f306853065c3c","afterSHA256":"fdf90e10bf1db95f074762e097e80c61cd82e960b2ac2808385f306853065c3c","reason":"Same source body in authored semantic page"},{"oldId":"b-1cf8c4a92267dce4cda866f6","targetId":"b-1cf8c4a92267dce4cda866f6","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"9f8de4572c912ab854fcccf5932a372018412e10c44dbb237cf7260d10cfc308","afterSHA256":"9f8de4572c912ab854fcccf5932a372018412e10c44dbb237cf7260d10cfc308","reason":"Same source body in authored semantic page"},{"oldId":"b-7a322e6157d86fb9cf764251","targetId":"b-7a322e6157d86fb9cf764251","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"79ae2f09f6174dcb8948392160784f39aefcad99854bef12aef60b9ed330c357","afterSHA256":"79ae2f09f6174dcb8948392160784f39aefcad99854bef12aef60b9ed330c357","reason":"Same source body in authored semantic page"},{"oldId":"b-c089aa1a6789d3dc45bd8114","targetId":"b-c089aa1a6789d3dc45bd8114","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"939c4761628c5da65993b40a6d844bfb7378ef29dbeeb37b428d08c76b31f013","afterSHA256":"939c4761628c5da65993b40a6d844bfb7378ef29dbeeb37b428d08c76b31f013","reason":"Same source body in authored semantic page"},{"oldId":"b-43334f1c1312bb8cffc0da33","targetId":"b-43334f1c1312bb8cffc0da33","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"rewrite","beforeSHA256":"5c0fc9438aff52b91880680acb5caaf5089e0667dc765caf1933d42591dad233","afterSHA256":"f24808c9a952fb28fdcdc02be7ca78258ce96c4f5df42045c6c1f671959e3cb0","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-7c53bf4393cf6b1a1809d60f","targetId":"b-7c53bf4393cf6b1a1809d60f","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"8a25829866cb20b99f921364d86c33657dae9e1547549d9560d15ceb1a6f9537","afterSHA256":"8a25829866cb20b99f921364d86c33657dae9e1547549d9560d15ceb1a6f9537","reason":"Same source body in authored semantic page"},{"oldId":"b-b594d04da5348f75d19fffc8","targetId":"b-b594d04da5348f75d19fffc8","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"9d2d37fbdcea1339fbcd1aeaef278a3912086f9ddca5746afa81e9059b4bd22b","afterSHA256":"09e76a6115a60db2f769dcaf9677a8e9e048c037ec149a8c650a05d73a12af9f","reason":"Same source body in authored semantic page"},{"oldId":"b-487abc79fa988f34f27f021f","targetId":"b-487abc79fa988f34f27f021f","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"f0c22e62c5520d432ca61fc6bad6c1e01e0460e70f7f9f2bdc6150386fd7aa62","afterSHA256":"9ab6840d67cc4c3ac9ff62f6ec9a8c36064f02605214b40349dbf9191fcbd589","reason":"Same source body in authored semantic page"},{"oldId":"b-ccf3be989ba4a4ef634bae38","targetId":"b-ccf3be989ba4a4ef634bae38","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"d50b94eb722dee5ced832487e587531ad8fc0c7e1ea6a13ff5fe4d0906e93d7b","afterSHA256":"d50b94eb722dee5ced832487e587531ad8fc0c7e1ea6a13ff5fe4d0906e93d7b","reason":"Same source body in authored semantic page"},{"oldId":"b-a81974d09a52b6dc4ee142c2","targetId":"b-a81974d09a52b6dc4ee142c2","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"8b7d63701e0e921bee18b7f1c1e38d6db5d9a5db3cb2e40577eaa14a66c80021","afterSHA256":"8b7d63701e0e921bee18b7f1c1e38d6db5d9a5db3cb2e40577eaa14a66c80021","reason":"Same source body in authored semantic page"},{"oldId":"b-321aa9f046bea8cf21fdcac0","targetId":"b-321aa9f046bea8cf21fdcac0","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"47399423a3cd722f6cc81989106117ca9567f5f15f0b56d84fbad77bbc4db80c","afterSHA256":"47399423a3cd722f6cc81989106117ca9567f5f15f0b56d84fbad77bbc4db80c","reason":"Same source body in authored semantic page"},{"oldId":"b-e9345df2b7ee301512a859cb","targetId":"b-e9345df2b7ee301512a859cb","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"11cc60ecac5230cad0208d185147fdee5e90412fc3a83335039e6be3103df597","afterSHA256":"11cc60ecac5230cad0208d185147fdee5e90412fc3a83335039e6be3103df597","reason":"Same source body in authored semantic page"},{"oldId":"b-ba9184d24dc136bdd439dde8","targetId":"b-ba9184d24dc136bdd439dde8","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"bee32e87d504e0167d898e9c0f91a56e4c7b091c9b8feb49d487de4dab842691","afterSHA256":"bee32e87d504e0167d898e9c0f91a56e4c7b091c9b8feb49d487de4dab842691","reason":"Same source body in authored semantic page"},{"oldId":"b-6c06d14c4c4cbc99b2434a6f","targetId":"b-6c06d14c4c4cbc99b2434a6f","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"4746f9c7084f10c1639a38d91b9ad149f85aba184e86b9c42f324db145121c64","afterSHA256":"4746f9c7084f10c1639a38d91b9ad149f85aba184e86b9c42f324db145121c64","reason":"Same source body in authored semantic page"},{"oldId":"b-ea119481db408e15090f7706","targetId":"b-ea119481db408e15090f7706","page":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","operation":"move","beforeSHA256":"609abc3860560a383fa699b0b71970390c0f3d0f80d59952beb64f4d04b0862a","afterSHA256":"609abc3860560a383fa699b0b71970390c0f3d0f80d59952beb64f4d04b0862a","reason":"Same source body in authored semantic page"},{"oldId":"block-9264aa76ab44410a94ba0c31731e50ef--四个数怎样指向图像里的同一区域","targetId":"b-b0162f4941340f1772a93a21","operation":"retain-source-identity"},{"oldId":"block-9264aa76ab44410a94ba0c31731e50ef--锚框先给坐标一个起点匹配再赋予训练身份","targetId":"b-e37758de87cc17697aab140d","operation":"retain-source-identity"},{"oldId":"block-8f46ef44ed4e4529beedf33817f14d3f--nms-与评价匹配解决两个不同的重复问题","targetId":"b-6de54f04bd7c726ca86f63db","operation":"retain-source-identity"},{"oldId":"block-8f46ef44ed4e4529beedf33817f14d3f--沿一次真实训练追踪从图像到框","targetId":"b-1cf8c4a92267dce4cda866f6","operation":"retain-source-identity"},{"oldId":"block-8f46ef44ed4e4529beedf33817f14d3f--共享特征减少重复计算但对象身份仍要保留","targetId":"b-e9345df2b7ee301512a859cb","operation":"retain-source-identity"}],"blocks":[{"id":"b-affe83e634aec87ec35c794f","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-affe83e634aec87ec35c794f","content":"# CLS 07 · 检测几何、匹配与评价","sha256":"5d6e7b33e278ad7f7a59a8a116727d408295a4963580ecb0cc72f0b51184d970","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-opening","title":"CLS 07 · 检测几何、匹配与评价","number":1},"sectionTitle":"CLS 07 · 检测几何、匹配与评价","currentQuestion":"CLS 07 · 检测几何、匹配与评价","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-3e5831dd77a73a3aa533b5e2","prerequisites":[{"id":"b-3e5831dd77a73a3aa533b5e2","url":"#b-3e5831dd77a73a3aa533b5e2","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-3e5831dd77a73a3aa533b5e2","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-3e5831dd77a73a3aa533b5e2","content":"按问题继续","sha256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-opening","title":"CLS 07 · 检测几何、匹配与评价","number":1},"sectionTitle":"CLS 07 · 检测几何、匹配与评价","currentQuestion":"CLS 07 · 检测几何、匹配与评价","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-affe83e634aec87ec35c794f","nextBlock":"b-4a192b0514374d5df100182a","prerequisites":[{"id":"b-affe83e634aec87ec35c794f","url":"#b-affe83e634aec87ec35c794f","role":"section-title"}],"relatedBlocks":[]},{"id":"b-4a192b0514374d5df100182a","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-4a192b0514374d5df100182a","content":"坐标、锚框偏移、训练/评价匹配、NMS/AP和真实检测训练连接","sha256":"88216e858779670a15bd42f20a06f136275ceee6a8685216f5821af4b1faf48f","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-opening","title":"CLS 07 · 检测几何、匹配与评价","number":1},"sectionTitle":"CLS 07 · 检测几何、匹配与评价","currentQuestion":"CLS 07 · 检测几何、匹配与评价","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-3e5831dd77a73a3aa533b5e2","nextBlock":"b-08b237d474ea6fe9c89ad57c","prerequisites":[{"id":"b-affe83e634aec87ec35c794f","url":"#b-affe83e634aec87ec35c794f","role":"section-title"},{"id":"b-3e5831dd77a73a3aa533b5e2","url":"#b-3e5831dd77a73a3aa533b5e2","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-08b237d474ea6fe9c89ad57c","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-08b237d474ea6fe9c89ad57c","content":"可先阅读已有结果，再决定是否运行。重算时请核对各节记录的数据、参数、环境与日期。","sha256":"f22152fd55dcdb0090b5653d37145694fd19388d31e10226549464fe8b483578","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-opening","title":"CLS 07 · 检测几何、匹配与评价","number":1},"sectionTitle":"CLS 07 · 检测几何、匹配与评价","currentQuestion":"CLS 07 · 检测几何、匹配与评价","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-4a192b0514374d5df100182a","nextBlock":null,"prerequisites":[{"id":"b-affe83e634aec87ec35c794f","url":"#b-affe83e634aec87ec35c794f","role":"section-title"},{"id":"b-3e5831dd77a73a3aa533b5e2","url":"#b-3e5831dd77a73a3aa533b5e2","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-18e139e5d97eaba63fe7c737","type":"p","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-18e139e5d97eaba63fe7c737","content":"**学习路径：** [一张图怎样变成一组框](/course/study/image-detection.html)","sha256":"93899883c4c044ee5171cc6d4515e67b10961c3d3e67eb507245ef30c5e625ce","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-059f9ebe54b6a66a5a81d0dc","prerequisites":[],"relatedBlocks":[]},{"id":"b-059f9ebe54b6a66a5a81d0dc","type":"p","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-059f9ebe54b6a66a5a81d0dc","content":"本册是检测单元的公式与结果解释，连接 [P27–35](/course/pages/l3-r2-p27.html) 和 [LAB 10 的图像到框训练](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/a14237bfa9dd413d959ab4e458e11f42)。先按自己的问题读一段，不必在运行模型前读完全部。","sha256":"8560cc92f5ba742f1c60ed74ecfdb4342af5e340aef761676c2308d0b54f08bc","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-18e139e5d97eaba63fe7c737","nextBlock":"b-0f02087943e28c2d6cca375f","prerequisites":[{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-0f02087943e28c2d6cca375f","type":"ul","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-0f02087943e28c2d6cca375f","content":"- 框如何表示和修正：[四偏移往返](/course/blocks/box-offset-roundtrip.html)。\n\n- 谁提供训练目标：[正例、背景与忽略项](/course/blocks/anchor-training-assignment.html)。\n\n- 重复框如何影响结果：[NMS 与评价匹配](/course/blocks/nms-versus-matching.html)。","sha256":"d2c2e5495e53ef8a3f1ca587b170cd2deadae52ef91b4508655dad4f24e8253e","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-059f9ebe54b6a66a5a81d0dc","nextBlock":"b-4f7c40e926fa9bd65f967bc7","prerequisites":[{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-4f7c40e926fa9bd65f967bc7","type":"li","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-4f7c40e926fa9bd65f967bc7","content":"- 框如何表示和修正：[四偏移往返](/course/blocks/box-offset-roundtrip.html)。","sha256":"d075524e22eca70e822d1c51e8bd1fcec6ce17365859e6b7281d6ed011199d65","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-0f02087943e28c2d6cca375f","nextBlock":"b-c78feaaba53e5517019cbbdb","prerequisites":[{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-c78feaaba53e5517019cbbdb","type":"li","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-c78feaaba53e5517019cbbdb","content":"- 谁提供训练目标：[正例、背景与忽略项](/course/blocks/anchor-training-assignment.html)。","sha256":"dcd700347a57ddfff3e1b7965ed1d5138733ba426f6d770c0d667c4357403782","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-4f7c40e926fa9bd65f967bc7","nextBlock":"b-80fa79a87a93bde84716d8a6","prerequisites":[{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-80fa79a87a93bde84716d8a6","type":"li","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-80fa79a87a93bde84716d8a6","content":"- 重复框如何影响结果：[NMS 与评价匹配](/course/blocks/nms-versus-matching.html)。","sha256":"f242c7ee4328040c119d646550367cf3a3d8f45a7a7d5b769edcec25309f4975","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c78feaaba53e5517019cbbdb","nextBlock":"b-bb91601d0984113d77b094df","prerequisites":[{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-bb91601d0984113d77b094df","type":"p","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-bb91601d0984113d77b094df","content":"读完后选一个预测，指出它被判错是因为重复、定位不足还是背景，并写明阈值和匹配协议。要亲手改坐标/分数，到附属工作台 [LAB 09](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/90d213b4e1e04c7e889ee6fc042b2358)；要理解模型在何处出错，回到 [检测学习证据](/course/blocks/detector-learning-evidence.html)。","sha256":"5824845f2bd25d6b85eb2bedeb7fde8bbdc2adbd87bd830bf65b0450fb73702c","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-80fa79a87a93bde84716d8a6","nextBlock":"b-4a5e4f005ff180e5f638b5c3","prerequisites":[{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-4a5e4f005ff180e5f638b5c3","type":"p","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-4a5e4f005ff180e5f638b5c3","content":"“这张图有杯子”可以由一个整图类别分数回答；“两只杯子分别在哪里”需要对象数、类别和位置。分类器即便准确，也没有被要求把分数分配给具体区域。全局平均池化还可能把左右位置汇总掉。检测因此要改变输出参数化、训练目标和评价协议，而不是给分类结果补画一个框。","sha256":"4069e14c5ad08dc619a5598db00977a4c0e0de429f00c9ecd0091eb7d58b8a1c","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-bb91601d0984113d77b094df","nextBlock":"b-b0162f4941340f1772a93a21","prerequisites":[{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-b0162f4941340f1772a93a21","type":"h2","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-b0162f4941340f1772a93a21","content":"## 四个数怎样指向图像里的同一区域","sha256":"0fe74dbe6635492abd693709fe30045ffdc781ab4cdbe8186bf03ddf8cbaf1fc","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-4a5e4f005ff180e5f638b5c3","nextBlock":"b-7ac1f8a70ae52eac28ce1da6","prerequisites":[{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-7ac1f8a70ae52eac28ce1da6","type":"p","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-7ac1f8a70ae52eac28ce1da6","content":"轴对齐框可表示为 $(x_{\\min},y_{\\min},x_{\\max},y_{\\max})$，也可表示为中心与宽高 $(c_x,c_y,w,h)$。本册用连续边界、右下边界不额外加 1 的约定，并把坐标除以图像宽高归一化到 $[0,1]$。这与“像素编号含两端”的面积算法不同，混用会让小框 IoU 尤其失真。","sha256":"07781e6961e60ab655d9b67933a7916b6b48aaf8b55e0c0f2f00f89651cfd9a7","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"(x_{\\min},y_{\\min},x_{\\max},y_{\\max})","definitionContext":"四个数怎样指向图像里的同一区域"},{"latex":"(c_x,c_y,w,h)","definitionContext":"四个数怎样指向图像里的同一区域"},{"latex":"[0,1]","definitionContext":"四个数怎样指向图像里的同一区域"}],"previousBlock":"b-b0162f4941340f1772a93a21","nextBlock":"b-d96604e375b8a2fde3d8a3c0","prerequisites":[{"id":"b-b0162f4941340f1772a93a21","url":"#b-b0162f4941340f1772a93a21","role":"section-title"},{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-d96604e375b8a2fde3d8a3c0","type":"p","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-d96604e375b8a2fde3d8a3c0","content":"真值 $G=(0.2,0.2,0.6,0.6)$，预测 $P=(0.3,0.2,0.7,0.6)$。二者面积都为 0.16，交集宽 0.3、高 0.4，面积 0.12；并集为 $0.16+0.16-0.12=0.20$，所以","sha256":"e025ab4ea87ddd435bebe1613d16b6b5d848975ad5bc92e5256735cf2cb53561","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"G=(0.2,0.2,0.6,0.6)","definitionContext":"四个数怎样指向图像里的同一区域"},{"latex":"P=(0.3,0.2,0.7,0.6)","definitionContext":"四个数怎样指向图像里的同一区域"},{"latex":"0.16+0.16-0.12=0.20","definitionContext":"四个数怎样指向图像里的同一区域"}],"previousBlock":"b-7ac1f8a70ae52eac28ce1da6","nextBlock":"b-dd5256ac58115ed6a0d5160e","prerequisites":[{"id":"b-b0162f4941340f1772a93a21","url":"#b-b0162f4941340f1772a93a21","role":"section-title"},{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-dd5256ac58115ed6a0d5160e","type":"p","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-dd5256ac58115ed6a0d5160e","content":"$\\mathrm{IoU}(P,G)=\\frac{|P\\cap G|}{|P\\cup G|}=0.6.$","sha256":"ad1f7ec36605770f96dde71af12439c5a535f3e77cb01cdaabb9a196ed5f5ff6","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"\\mathrm{IoU}(P,G)=\\frac{|P\\cap G|}{|P\\cup G|}=0.6.","definitionContext":"四个数怎样指向图像里的同一区域"}],"previousBlock":"b-d96604e375b8a2fde3d8a3c0","nextBlock":"b-ce80e8e778aebeb5ed497f5a","prerequisites":[{"id":"b-b0162f4941340f1772a93a21","url":"#b-b0162f4941340f1772a93a21","role":"section-title"},{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-ce80e8e778aebeb5ed497f5a","type":"p","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-ce80e8e778aebeb5ed497f5a","content":"IoU 对同时缩放两框不变，却对相对位移敏感。把同样 0.1 的横移施加到宽 0.2 的小框，重叠比例会比宽 0.4 的框更差。检测中“小目标难”有一部分来自几何尺度，不能只归因于类别识别。","sha256":"b1a8121bb9b156ad2aebe396b09b8e6178ff5376ae3b0d39fd29ae93f53af5aa","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-dd5256ac58115ed6a0d5160e","nextBlock":"b-8e7f196c91b64b90b1cf135e","prerequisites":[{"id":"b-b0162f4941340f1772a93a21","url":"#b-b0162f4941340f1772a93a21","role":"section-title"},{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-8e7f196c91b64b90b1cf135e","type":"p","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-8e7f196c91b64b90b1cf135e","content":"![框重叠的精确几何与检测排序](/course/repair/assets/cls08-box-ap.svg)","sha256":"90b1e23eb5816cc872991feea47f4be353693efb37b7aabcb3cb413d74365fca","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-ce80e8e778aebeb5ed497f5a","nextBlock":"b-9ba22d63bb9aa7f560ed915f","prerequisites":[{"id":"b-b0162f4941340f1772a93a21","url":"#b-b0162f4941340f1772a93a21","role":"section-title"},{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-9ba22d63bb9aa7f560ed915f","type":"p","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-9ba22d63bb9aa7f560ed915f","content":"左图以相同坐标比例展示 IoU=0.6；右表是一类、两个真值对象的排序算例，第二个预测虽也盖住对象，却因重复而算假阳性。","sha256":"9098789457bdb4955ec4f933dc2f9e3ba5366f7356882e102ee2309a78302040","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-8e7f196c91b64b90b1cf135e","nextBlock":"b-e37758de87cc17697aab140d","prerequisites":[{"id":"b-b0162f4941340f1772a93a21","url":"#b-b0162f4941340f1772a93a21","role":"section-title"},{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-e37758de87cc17697aab140d","type":"h2","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-e37758de87cc17697aab140d","content":"## 锚框先给坐标一个起点，匹配再赋予训练身份","sha256":"8e6b0c8407cefcaeb3b9a235195b40b038695cddb1c3461c819b19ec767f0650","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-9ba22d63bb9aa7f560ed915f","nextBlock":"b-beebf75e778f4bef3a5f419a","prerequisites":[{"id":"b-b0162f4941340f1772a93a21","url":"#b-b0162f4941340f1772a93a21","role":"section-title"},{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-beebf75e778f4bef3a5f419a","type":"p","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-beebf75e778f4bef3a5f419a","content":"锚框是预设位置、尺度和长宽比的参考框，不是模型已经检测出的结果。输出头为每个锚框预测类别或 objectness，以及相对坐标改变量。一种不加额外尺度系数的参数化是","sha256":"39e9d99c41a61b16e0edc703daef9d3075d4e5c7b18009e2090e7a4dfc508443","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-e37758de87cc17697aab140d","nextBlock":"b-9d7d0dd4d72116403e38b7d2","prerequisites":[{"id":"b-e37758de87cc17697aab140d","url":"#b-e37758de87cc17697aab140d","role":"section-title"},{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-9d7d0dd4d72116403e38b7d2","type":"p","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-9d7d0dd4d72116403e38b7d2","content":"$t_x=(c_x^g-c_x^a)/w^a,\\quad t_y=(c_y^g-c_y^a)/h^a,\\quad\nt_w=\\log(w^g/w^a),\\quad t_h=\\log(h^g/h^a).$","sha256":"f2e92753771abbafbcb6b78618e928db0d86d334070ed63cd8ae3c084e0a19f5","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"t_x=(c_x^g-c_x^a)/w^a,\\quad t_y=(c_y^g-c_y^a)/h^a,\\quad\nt_w=\\log(w^g/w^a),\\quad t_h=\\log(h^g/h^a).","definitionContext":"四个数怎样指向图像里的同一区域"}],"previousBlock":"b-beebf75e778f4bef3a5f419a","nextBlock":"b-d76fae133d2523ee907cd725","prerequisites":[{"id":"b-e37758de87cc17697aab140d","url":"#b-e37758de87cc17697aab140d","role":"section-title"},{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-d76fae133d2523ee907cd725","type":"p","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-d76fae133d2523ee907cd725","content":"锚框中心 $(0.4,0.4)$、宽高均为 0.4，真值中心移到 $(0.45,0.4)$ 且大小不变，目标就是 $(0.125,0,0,0)$。解码时用 $c_x^p=c_x^a+w^at_x$、$w^p=w^a\\exp(t_w)$ 等逆运算。具体实现可能乘 10、5 等尺度因子，例如 D2L 的一套约定；编码和解码必须成对，不可只复制一边公式。","sha256":"40858ec0f600a17096cc7466a4f922b9540d5c1a169226cf8af91498a8a711e3","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"(0.4,0.4)","definitionContext":"四个数怎样指向图像里的同一区域"},{"latex":"(0.45,0.4)","definitionContext":"四个数怎样指向图像里的同一区域"},{"latex":"(0.125,0,0,0)","definitionContext":"四个数怎样指向图像里的同一区域"},{"latex":"c_x^p=c_x^a+w^at_x","definitionContext":"四个数怎样指向图像里的同一区域"},{"latex":"w^p=w^a\\exp(t_w)","definitionContext":"四个数怎样指向图像里的同一区域"}],"previousBlock":"b-9d7d0dd4d72116403e38b7d2","nextBlock":"b-84cf6883740bdc3d1748e68d","prerequisites":[{"id":"b-e37758de87cc17697aab140d","url":"#b-e37758de87cc17697aab140d","role":"section-title"},{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-84cf6883740bdc3d1748e68d","type":"p","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-84cf6883740bdc3d1748e68d","content":"训练时要把锚框分配为正例、负例或忽略项。常见方案先保障每个真值有匹配，再按 IoU 阈值增加正例，把远离真值的锚框设为背景，中间区域可忽略。两阈值的具体值是协议选择，不是 IoU 的定义。位置回归只对正例计损失；背景没有“应该回归到哪里”的真值。若一幅图没有对象，仍应通过负例分类损失学会不报框，回归项则安全地归零，避免除以零。","sha256":"072c7fcb91f010db36447b5f7636ad481ddad26681f26397e6b7e7661508f502","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-d76fae133d2523ee907cd725","nextBlock":"b-8656bd9b4bd2d200803bcbb7","prerequisites":[{"id":"b-e37758de87cc17697aab140d","url":"#b-e37758de87cc17697aab140d","role":"section-title"},{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-8656bd9b4bd2d200803bcbb7","type":"p","scope":"block-9264aa76ab44410a94ba0c31731e50ef","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-8656bd9b4bd2d200803bcbb7","content":"写成 $L=L_{cls}+\\lambda L_{box}$ 仍不够：两个分母怎样归一化、负例是否筛选、ignore 如何 mask，都决定了优化问题。大量背景锚框可能淹没少数正例；hard-negative mining 和 focal loss 是改变贡献权重的不同办法。学生可以沿 [D2L §14.4 锚框分配/偏移](https://d2l.ai/chapter_computer-vision/anchor.html) 与 [§14.7 TinySSD](https://d2l.ai/chapter_computer-vision/ssd.html) 核对输出张量、分类标签和回归 mask。","sha256":"842f4305a9f2d8fe7fb3192c4b4b2ff23e7ab466c311abaad3889acebfb6dd58","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-9264aa76ab44410a94ba0c31731e50ef","title":"四个数怎样指向图像里的同一区域","number":2},"sectionTitle":"四个数怎样指向图像里的同一区域","currentQuestion":"四个数怎样指向图像里的同一区域","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"L=L_{cls}+\\lambda L_{box}","definitionContext":"四个数怎样指向图像里的同一区域"}],"previousBlock":"b-84cf6883740bdc3d1748e68d","nextBlock":null,"prerequisites":[{"id":"b-e37758de87cc17697aab140d","url":"#b-e37758de87cc17697aab140d","role":"section-title"},{"id":"b-18e139e5d97eaba63fe7c737","url":"#b-18e139e5d97eaba63fe7c737","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-6de54f04bd7c726ca86f63db","type":"h2","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-6de54f04bd7c726ca86f63db","content":"## NMS 与评价匹配解决两个不同的重复问题","sha256":"72d3c7834c3fafe18571e3ac3fc55ec05967604ca8e1c604fc52313b1c6b033b","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-741c18a315124e1fd3152db4","prerequisites":[{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-741c18a315124e1fd3152db4","type":"p","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-741c18a315124e1fd3152db4","content":"Non-maximum suppression（NMS）是推理后处理：通常按类把预测框按分数排序，保留最高分框，删除与之重叠超过阈值的其他框，再继续。它减少同一对象的重复报告，但不知道真值；挤在一起的不同对象也可能被误删。NMS 阈值影响交付的结果，不能在测试集上追着最好 AP 挑阈值。","sha256":"b117cf446d2ad20730ef6f594c83df176d7660e6471a02cd3f1b71e3e15edf9e","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-6de54f04bd7c726ca86f63db","nextBlock":"b-93bdc2a6600e835e5bc702df","prerequisites":[{"id":"b-6de54f04bd7c726ca86f63db","url":"#b-6de54f04bd7c726ca86f63db","role":"section-title"}],"relatedBlocks":[]},{"id":"b-93bdc2a6600e835e5bc702df","type":"p","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-93bdc2a6600e835e5bc702df","content":"评价匹配则拿最终预测与真值比较。在固定类别和 IoU 阈值下，按置信度从高到低，每个预测至多匹配一个尚未被使用的真值；同一真值只能被命中一次。于是“框覆盖了对象”还不保证是真阳性，它可能是已有更高分预测的重复框。","sha256":"315a10d85c66d1a4cc67a41fd3d34c08e2b14631a03e3d0174dc9b1de08132f4","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-741c18a315124e1fd3152db4","nextBlock":"b-3c41d9b855efb5e7618f0f81","prerequisites":[{"id":"b-6de54f04bd7c726ca86f63db","url":"#b-6de54f04bd7c726ca86f63db","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-3c41d9b855efb5e7618f0f81","type":"p","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-3c41d9b855efb5e7618f0f81","content":"设共有两个真值，四个预测排序后的身份依次为 TP、FP（重复）、FP（背景）、TP。累计 precision 为 $1,1/2,1/3,1/2$，recall 为 $1/2,1/2,1/2,1$。若采用本册的全点插值 precision envelope 与 recall 增量积分，AP 为 $0.5\\times1+0.5\\times0.5=0.75$。这是明确定义的简化 AP；COCO 的 101 个 recall 阈值采样、多个 IoU 阈值平均和面积分组是更具体的另一套协议，不能把一个数未经说明叫“COCO AP”。","sha256":"ac5d224648f4ebde30f06c9a67e260c09658b54cddb2f0ff694279ac674969bc","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"1,1/2,1/3,1/2","definitionContext":"NMS 与评价匹配解决两个不同的重复问题"},{"latex":"1/2,1/2,1/2,1","definitionContext":"NMS 与评价匹配解决两个不同的重复问题"},{"latex":"0.5\\times1+0.5\\times0.5=0.75","definitionContext":"NMS 与评价匹配解决两个不同的重复问题"}],"previousBlock":"b-93bdc2a6600e835e5bc702df","nextBlock":"b-04ce59162ab0695e1069f8fc","prerequisites":[{"id":"b-6de54f04bd7c726ca86f63db","url":"#b-6de54f04bd7c726ca86f63db","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-04ce59162ab0695e1069f8fc","type":"p","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-04ce59162ab0695e1069f8fc","content":"[LAB 09](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/90d213b4e1e04c7e889ee6fc042b2358?utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_content=90d213b4e1e04c7e889ee6fc042b2358&utm_term=get_notebook) 可改重复框分数，观察排序如何改变 AP；也可改 IoU 阈值，使一个原先合格的框失去匹配。准确率把大量背景当作普通样本时可能很高，却无法体现检测的定位与重复问题。","sha256":"fdf90e10bf1db95f074762e097e80c61cd82e960b2ac2808385f306853065c3c","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-3c41d9b855efb5e7618f0f81","nextBlock":"b-1cf8c4a92267dce4cda866f6","prerequisites":[{"id":"b-6de54f04bd7c726ca86f63db","url":"#b-6de54f04bd7c726ca86f63db","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-1cf8c4a92267dce4cda866f6","type":"h2","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-1cf8c4a92267dce4cda866f6","content":"## 沿一次真实训练追踪从图像到框","sha256":"9f8de4572c912ab854fcccf5932a372018412e10c44dbb237cf7260d10cfc308","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-04ce59162ab0695e1069f8fc","nextBlock":"b-7a322e6157d86fb9cf764251","prerequisites":[{"id":"b-6de54f04bd7c726ca86f63db","url":"#b-6de54f04bd7c726ca86f63db","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-7a322e6157d86fb9cf764251","type":"p","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-7a322e6157d86fb9cf764251","content":"[LAB 10](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/a14237bfa9dd413d959ab4e458e11f42?utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_content=a14237bfa9dd413d959ab4e458e11f42&utm_term=get_notebook) 的主实验应按以下对象检查：RGB 图像和类别/框标注进入数据集，卷积骨干产生空间特征，输出头产生类别与偏移，匹配提供监督，损失更新参数，解码及 NMS 产生最终框，最后仅用保留图像评价。真值坐标只用于目标与评价，不能作为图像模型的输入特征。","sha256":"79ae2f09f6174dcb8948392160784f39aefcad99854bef12aef60b9ed330c357","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-1cf8c4a92267dce4cda866f6","nextBlock":"b-c089aa1a6789d3dc45bd8114","prerequisites":[{"id":"b-1cf8c4a92267dce4cda866f6","url":"#b-1cf8c4a92267dce4cda866f6","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-c089aa1a6789d3dc45bd8114","type":"p","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-c089aa1a6789d3dc45bd8114","content":"主实验已在 2026-09-22 本地 CPU 完整运行。输入是 48×48 RGB 合成图，含红矩形、蓝椭圆、可变位置与尺度、多个对象及空图；颜色与形状相关，因此不是自然场景通用检测基准。按原始场景种子划分 768 训练、96 验证、192 保留图。45,806 参数的卷积检测器在单层网格上使用 288 个锚框，训练 28 epoch；正例阈值 0.45、忽略区间 [0.25,0.45)，并保障每个真值有独立强制匹配。分类使用 hard-negative CE，位置项为正例 smooth-L1，解码后按类 NMS。","sha256":"939c4761628c5da65993b40a6d844bfb7378ef29dbeeb37b428d08c76b31f013","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-7a322e6157d86fb9cf764251","nextBlock":"b-43334f1c1312bb8cffc0da33","prerequisites":[{"id":"b-1cf8c4a92267dce4cda866f6","url":"#b-1cf8c4a92267dce4cda866f6","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-43334f1c1312bb8cffc0da33","type":"table","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-43334f1c1312bb8cffc0da33","content":"| 同一保留协议 | 两类平均 AP@IoU0.50 | 两类平均 AP@IoU0.75 |\n| --- | --- | --- |\n| 随机初始化 | 0.00746 | 0.00003 |\n| 训练后，原尺度 | 0.97303 | 0.70367 |\n| 训练后，仅对象尺寸减半 | 0.61490 | 0.17987 |","sha256":"f24808c9a952fb28fdcdc02be7ca78258ce96c4f5df42045c6c1f671959e3cb0","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"table":{"headers":["同一保留协议","两类平均 AP@IoU0.50","两类平均 AP@IoU0.75"],"rows":[["同一保留协议","两类平均 AP@IoU0.50","两类平均 AP@IoU0.75"],["随机初始化","0.00746","0.00003"],["训练后，原尺度","0.97303","0.70367"],["训练后，仅对象尺寸减半","0.61490","0.17987"]]},"previousBlock":"b-c089aa1a6789d3dc45bd8114","nextBlock":"b-7c53bf4393cf6b1a1809d60f","prerequisites":[{"id":"b-1cf8c4a92267dce4cda866f6","url":"#b-1cf8c4a92267dce4cda866f6","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-7c53bf4393cf6b1a1809d60f","type":"p","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-7c53bf4393cf6b1a1809d60f","content":"AP 按各类全局分数排序、一个真值最多匹配一次、precision envelope 全点积分；不是 COCO 多 IoU 平均 AP。保留集有 33 张空图，展示阈值 0.4 下没有空图误报；这不代表所有阈值或所有背景都不会误报。AP50 与 AP75 的差距说明定位仍不够精细，尺度减半的下降说明训练分布外的小对象尤其困难。","sha256":"8a25829866cb20b99f921364d86c33657dae9e1547549d9560d15ceb1a6f9537","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-43334f1c1312bb8cffc0da33","nextBlock":"b-b594d04da5348f75d19fffc8","prerequisites":[{"id":"b-1cf8c4a92267dce4cda866f6","url":"#b-1cf8c4a92267dce4cda866f6","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-b594d04da5348f75d19fffc8","type":"p","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-b594d04da5348f75d19fffc8","content":"![真实模型的保留预测与失败](/assets/display/88df211a98edfb7b00515441.webp)","sha256":"09e76a6115a60db2f769dcaf9677a8e9e048c037ec149a8c650a05d73a12af9f","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-7c53bf4393cf6b1a1809d60f","nextBlock":"b-487abc79fa988f34f27f021f","prerequisites":[{"id":"b-1cf8c4a92267dce4cda866f6","url":"#b-1cf8c4a92267dce4cda866f6","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-487abc79fa988f34f27f021f","type":"p","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-487abc79fa988f34f27f021f","content":"![仅对象尺寸减半时的预测](/assets/display/735d18643da4042bf639ef2c.webp)","sha256":"9ab6840d67cc4c3ac9ff62f6ec9a8c36064f02605214b40349dbf9191fcbd589","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-b594d04da5348f75d19fffc8","nextBlock":"b-ccf3be989ba4a4ef634bae38","prerequisites":[{"id":"b-1cf8c4a92267dce4cda866f6","url":"#b-1cf8c4a92267dce4cda866f6","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-ccf3be989ba4a4ef634bae38","type":"p","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-ccf3be989ba4a4ef634bae38","content":"保持权重与输入不变，仅把 NMS IoU 阈值从 0.35 放宽到 0.65，AP50 变为 0.95866；展示阈值下 recall 从 0.98311 到 1.0，precision 则从 0.94175 降到 0.41926。更多预测覆盖了剩余真值，也留下大量重复框。这组结果把 NMS 的取舍落实到了同一模型，而非抽象宣称后处理总会改善成绩。","sha256":"d50b94eb722dee5ced832487e587531ad8fc0c7e1ea6a13ff5fe4d0906e93d7b","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-487abc79fa988f34f27f021f","nextBlock":"b-a81974d09a52b6dc4ee142c2","prerequisites":[{"id":"b-1cf8c4a92267dce4cda866f6","url":"#b-1cf8c4a92267dce4cda866f6","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-a81974d09a52b6dc4ee142c2","type":"p","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-a81974d09a52b6dc4ee142c2","content":"四维高斯样本上的逻辑回归仍能说明 sigmoid 分类损失和参数更新，但它没有图像和位置输出，应保留为分类热身。完整检测器给出的错误则可以进一步分解：漏检、错类、框偏、重复、空图误报、尺度外推。这些区分决定下一步该改表示、匹配、损失、解码还是数据。","sha256":"8b7d63701e0e921bee18b7f1c1e38d6db5d9a5db3cb2e40577eaa14a66c80021","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-ccf3be989ba4a4ef634bae38","nextBlock":"b-321aa9f046bea8cf21fdcac0","prerequisites":[{"id":"b-1cf8c4a92267dce4cda866f6","url":"#b-1cf8c4a92267dce4cda866f6","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-321aa9f046bea8cf21fdcac0","type":"p","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-321aa9f046bea8cf21fdcac0","content":"即使检测框全部正确，框里的杯把空洞和两杯接触边界仍未知。下一册 [CLS 08](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/2d81e2a1936043349eacb71c53c0defa?utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_content=2d81e2a1936043349eacb71c53c0defa&utm_term=get_notebook) 从这个具体缺口走向逐像素和实例输出。","sha256":"47399423a3cd722f6cc81989106117ca9567f5f15f0b56d84fbad77bbc4db80c","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-a81974d09a52b6dc4ee142c2","nextBlock":"b-a8235fb384b29ef3aff3893a","prerequisites":[{"id":"b-1cf8c4a92267dce4cda866f6","url":"#b-1cf8c4a92267dce4cda866f6","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-a8235fb384b29ef3aff3893a","type":"h4","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-a8235fb384b29ef3aff3893a","content":"#### 回看：共享卷积、RPN 与一阶段检测的计算位置","sha256":"9d01ec975dd28a25f7030eff6e9ed5dac5ce031ab871645fb164d3f3b8ed1dfa","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-321aa9f046bea8cf21fdcac0","nextBlock":"b-e9345df2b7ee301512a859cb","prerequisites":[{"id":"b-1cf8c4a92267dce4cda866f6","url":"#b-1cf8c4a92267dce4cda866f6","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-e9345df2b7ee301512a859cb","type":"h2","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-e9345df2b7ee301512a859cb","content":"## 共享特征减少重复计算，但对象身份仍要保留","sha256":"11cc60ecac5230cad0208d185147fdee5e90412fc3a83335039e6be3103df597","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-a8235fb384b29ef3aff3893a","nextBlock":"b-ba9184d24dc136bdd439dde8","prerequisites":[{"id":"b-a8235fb384b29ef3aff3893a","url":"#b-a8235fb384b29ef3aff3893a","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-ba9184d24dc136bdd439dde8","type":"p","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-ba9184d24dc136bdd439dde8","content":"早期区域 CNN 为很多候选分别提特征，重复处理重叠像素。Fast R-CNN 把卷积移到整图共享阶段，再在区域内读取特征；Faster R-CNN 用 RPN 从共享特征提出候选；SSD 在多个尺度的特征图上直接预测类别和位置。这些变化围绕候选产生、特征复用和预测阶段数组织，并不意味着所有方法在任何设备或目标尺度上只有一个速度排序。[Fast R-CNN](https://arxiv.org/abs/1504.08083) · [Faster R-CNN](https://arxiv.org/abs/1506.01497) · [SSD](https://arxiv.org/abs/1512.02325)。","sha256":"bee32e87d504e0167d898e9c0f91a56e4c7b091c9b8feb49d487de4dab842691","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-e9345df2b7ee301512a859cb","nextBlock":"b-6c06d14c4c4cbc99b2434a6f","prerequisites":[{"id":"b-e9345df2b7ee301512a859cb","url":"#b-e9345df2b7ee301512a859cb","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-6c06d14c4c4cbc99b2434a6f","type":"p","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-6c06d14c4c4cbc99b2434a6f","content":"共享整图特征不应抹掉区域身份。同一张特征图上取左杯和右杯两个区域，仍要得到各自预测；把所有空间位置先平均成一个向量，再期待检测头重建任意细节，可能遇到表示碰撞。是否保留坐标、感受野与采样密度，应与表示的分类能力一起检查。","sha256":"4746f9c7084f10c1639a38d91b9ad149f85aba184e86b9c42f324db145121c64","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-ba9184d24dc136bdd439dde8","nextBlock":"b-ea119481db408e15090f7706","prerequisites":[{"id":"b-e9345df2b7ee301512a859cb","url":"#b-e9345df2b7ee301512a859cb","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-ea119481db408e15090f7706","type":"p","scope":"block-8f46ef44ed4e4529beedf33817f14d3f","url":"https://codingai-lec03.pages.dev/course/cls08-检测几何匹配与评价.html#b-ea119481db408e15090f7706","content":"[一阶段检测如何对应 LAB 10](/course/blocks/dense-detector-ssd-extension.html)","sha256":"609abc3860560a383fa699b0b71970390c0f3d0f80d59952beb64f4d04b0862a","chapter":{"id":"course","title":"CLS 07 · 检测几何、匹配与评价"},"unit":{"id":"cls08-检测几何匹配与评价","title":"CLS 07 · 检测几何、匹配与评价"},"page":{"id":"reading-block-8f46ef44ed4e4529beedf33817f14d3f","title":"NMS 与评价匹配解决两个不同的重复问题","number":3},"sectionTitle":"NMS 与评价匹配解决两个不同的重复问题","currentQuestion":"NMS 与评价匹配解决两个不同的重复问题","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-6c06d14c4c4cbc99b2434a6f","nextBlock":null,"prerequisites":[{"id":"b-e9345df2b7ee301512a859cb","url":"#b-e9345df2b7ee301512a859cb","role":"section-title"},{"id":"b-741c18a315124e1fd3152db4","url":"#b-741c18a315124e1fd3152db4","role":"problem-context"}],"relatedBlocks":[]}]}