{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/blocks/pixel-cross-entropy.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"每个有效像素提供一项监督","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-cross-entropy.html","markdown":"https://codingai-lec03.pages.dev/course/blocks/pixel-cross-entropy.md","pages":[{"id":"reading-opening","title":"每个有效像素提供一项监督","unit":"怎样直接学习一张像素类别地图？","unitId":"pixel-prediction-objective","chapter":"II","number":1}],"aliases":[{"oldId":"b-0c7cba43a95b7b0f276f0713","targetId":"b-0c7cba43a95b7b0f276f0713","page":"reading-opening","operation":"move","beforeSHA256":"8f81f3302260d979fd85bedcb932a9da02ed17845724da5c61a67db68acd306b","afterSHA256":"8f81f3302260d979fd85bedcb932a9da02ed17845724da5c61a67db68acd306b","reason":"Same source body in authored semantic page"},{"oldId":"b-af28245ff7c5891e33b674ff","targetId":"b-af28245ff7c5891e33b674ff","page":"reading-opening","operation":"move","beforeSHA256":"183d41f9f2dd4315d6651bd668ce69b215479fec798b3e287bba7d446b7f4a31","afterSHA256":"183d41f9f2dd4315d6651bd668ce69b215479fec798b3e287bba7d446b7f4a31","reason":"Same source body in authored semantic page"},{"oldId":"b-1b3419b2595778faae15d081","targetId":"b-1b3419b2595778faae15d081","page":"reading-opening","operation":"move","beforeSHA256":"eb0dfe6cebaefe68f519f69350ef606a2c687649403ed826851930d7434e55db","afterSHA256":"eb0dfe6cebaefe68f519f69350ef606a2c687649403ed826851930d7434e55db","reason":"Same source body in authored semantic page"},{"oldId":"b-bd1b02f99081b068da06d195","targetId":"b-bd1b02f99081b068da06d195","page":"reading-opening","operation":"move","beforeSHA256":"ab380c1e0f9d1f36683516a23e6ae4f936078b69369c6e93f7772b9db9492422","afterSHA256":"ab380c1e0f9d1f36683516a23e6ae4f936078b69369c6e93f7772b9db9492422","reason":"Same source body in authored semantic page"},{"oldId":"b-05556ad03eab79c710d9d82c","targetId":"b-05556ad03eab79c710d9d82c","page":"reading-opening","operation":"move","beforeSHA256":"a4c0dbd56af877d0b5e594f748f6f8d9603d840d6fc9a5838318be6e455ab3d5","afterSHA256":"a4c0dbd56af877d0b5e594f748f6f8d9603d840d6fc9a5838318be6e455ab3d5","reason":"Same source body in authored semantic page"},{"oldId":"b-bfa7f45f9031dc10c52b3a8f","targetId":"b-bfa7f45f9031dc10c52b3a8f","page":"reading-opening","operation":"move","beforeSHA256":"fb67cbca43315249869971d864733783ccb53739e70b92e40c20d6581533bb82","afterSHA256":"fb67cbca43315249869971d864733783ccb53739e70b92e40c20d6581533bb82","reason":"Same source body in authored semantic page"},{"oldId":"b-7cb961a56fa7777568b6bcbd","targetId":"b-7cb961a56fa7777568b6bcbd","page":"reading-opening","operation":"move","beforeSHA256":"75a26f3c334073d2322457d1f81001561d6c818c8f83b13b0f038ae1d77e5cc7","afterSHA256":"75a26f3c334073d2322457d1f81001561d6c818c8f83b13b0f038ae1d77e5cc7","reason":"Same source body in authored semantic page"},{"oldId":"b-c8a8c7c0eddb664c190ef9d8","targetId":"b-c8a8c7c0eddb664c190ef9d8","page":"reading-opening","operation":"move","beforeSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","afterSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","reason":"Same source body in authored semantic page"}],"blocks":[{"id":"b-0c7cba43a95b7b0f276f0713","type":"h1","scope":"pixel-cross-entropy","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-cross-entropy.html#b-0c7cba43a95b7b0f276f0713","content":"# 每个有效像素提供一项监督","sha256":"8f81f3302260d979fd85bedcb932a9da02ed17845724da5c61a67db68acd306b","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"pixel-prediction-objective","title":"怎样直接学习一张像素类别地图？"},"page":{"id":"reading-opening","title":"每个有效像素提供一项监督","number":1},"sectionTitle":"每个有效像素提供一项监督","currentQuestion":"每个有效像素提供一项监督","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-af28245ff7c5891e33b674ff","prerequisites":[{"id":"b-af28245ff7c5891e33b674ff","url":"#b-af28245ff7c5891e33b674ff","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-af28245ff7c5891e33b674ff","type":"p","scope":"pixel-cross-entropy","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-cross-entropy.html#b-af28245ff7c5891e33b674ff","content":"FCN 从逐像素类别图出发，独立于区域检测路线。将 `[B,C,H,W]` logits 与 `[B,H,W]` 标签一一对应，理解分割损失和 ignore 像素。LAB11 从图像预测 mask；已知 mask 的缩放练习只检验采样，不检验视觉预测。","sha256":"183d41f9f2dd4315d6651bd668ce69b215479fec798b3e287bba7d446b7f4a31","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"pixel-prediction-objective","title":"怎样直接学习一张像素类别地图？"},"page":{"id":"reading-opening","title":"每个有效像素提供一项监督","number":1},"sectionTitle":"每个有效像素提供一项监督","currentQuestion":"每个有效像素提供一项监督","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-0c7cba43a95b7b0f276f0713","nextBlock":"b-1b3419b2595778faae15d081","prerequisites":[{"id":"b-0c7cba43a95b7b0f276f0713","url":"#b-0c7cba43a95b7b0f276f0713","role":"section-title"}],"relatedBlocks":[]},{"id":"b-1b3419b2595778faae15d081","type":"p","scope":"pixel-cross-entropy","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-cross-entropy.html#b-1b3419b2595778faae15d081","content":"对目标类别为 $y_{uv}$ 的像素，交叉熵是 $-\\log p_{y_{uv},u,v}$。若有效像素集合为 Ω，平均损失写成","sha256":"eb0dfe6cebaefe68f519f69350ef606a2c687649403ed826851930d7434e55db","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"pixel-prediction-objective","title":"怎样直接学习一张像素类别地图？"},"page":{"id":"reading-opening","title":"每个有效像素提供一项监督","number":1},"sectionTitle":"每个有效像素提供一项监督","currentQuestion":"每个有效像素提供一项监督","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"y_{uv}","definitionContext":"每个有效像素提供一项监督"},{"latex":"-\\log p_{y_{uv},u,v}","definitionContext":"每个有效像素提供一项监督"}],"previousBlock":"b-af28245ff7c5891e33b674ff","nextBlock":"b-bd1b02f99081b068da06d195","prerequisites":[{"id":"b-0c7cba43a95b7b0f276f0713","url":"#b-0c7cba43a95b7b0f276f0713","role":"section-title"},{"id":"b-af28245ff7c5891e33b674ff","url":"#b-af28245ff7c5891e33b674ff","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-bd1b02f99081b068da06d195","type":"p","scope":"pixel-cross-entropy","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-cross-entropy.html#b-bd1b02f99081b068da06d195","content":"$\nL=-\\frac{1}{|\\Omega|}\\sum_{(u,v)\\in\\Omega}\\log p_{y_{uv},u,v}.\n$","sha256":"ab380c1e0f9d1f36683516a23e6ae4f936078b69369c6e93f7772b9db9492422","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"pixel-prediction-objective","title":"怎样直接学习一张像素类别地图？"},"page":{"id":"reading-opening","title":"每个有效像素提供一项监督","number":1},"sectionTitle":"每个有效像素提供一项监督","currentQuestion":"每个有效像素提供一项监督","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"\nL=-\\frac{1}{|\\Omega|}\\sum_{(u,v)\\in\\Omega}\\log p_{y_{uv},u,v}.\n","definitionContext":"每个有效像素提供一项监督"}],"previousBlock":"b-1b3419b2595778faae15d081","nextBlock":"b-05556ad03eab79c710d9d82c","prerequisites":[{"id":"b-0c7cba43a95b7b0f276f0713","url":"#b-0c7cba43a95b7b0f276f0713","role":"section-title"},{"id":"b-af28245ff7c5891e33b674ff","url":"#b-af28245ff7c5891e33b674ff","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-05556ad03eab79c710d9d82c","type":"p","scope":"pixel-cross-entropy","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-cross-entropy.html#b-05556ad03eab79c710d9d82c","content":"三类 logits `[B,3,H,W]` 与整数目标 `[B,H,W]` 相配；不要把类别 ID 当成连续数直接算回归平方误差。二类分割也可只预测一个前景 logit，用 sigmoid 与二元交叉熵，这时标签与输出接口跟三类方案不同。","sha256":"a4c0dbd56af877d0b5e594f748f6f8d9603d840d6fc9a5838318be6e455ab3d5","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"pixel-prediction-objective","title":"怎样直接学习一张像素类别地图？"},"page":{"id":"reading-opening","title":"每个有效像素提供一项监督","number":1},"sectionTitle":"每个有效像素提供一项监督","currentQuestion":"每个有效像素提供一项监督","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-bd1b02f99081b068da06d195","nextBlock":"b-bfa7f45f9031dc10c52b3a8f","prerequisites":[{"id":"b-0c7cba43a95b7b0f276f0713","url":"#b-0c7cba43a95b7b0f276f0713","role":"section-title"},{"id":"b-af28245ff7c5891e33b674ff","url":"#b-af28245ff7c5891e33b674ff","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-bfa7f45f9031dc10c52b3a8f","type":"p","scope":"pixel-cross-entropy","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-cross-entropy.html#b-bfa7f45f9031dc10c52b3a8f","content":"未标注位置常用 ignore=255。它表示“不在这里计算监督和指标”，不表示第四类，也不等于背景。LAB11 的损失和混淆矩阵都排除这些位置；否则模型可能因为迎合不存在的背景标签而产生错误梯度。","sha256":"fb67cbca43315249869971d864733783ccb53739e70b92e40c20d6581533bb82","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"pixel-prediction-objective","title":"怎样直接学习一张像素类别地图？"},"page":{"id":"reading-opening","title":"每个有效像素提供一项监督","number":1},"sectionTitle":"每个有效像素提供一项监督","currentQuestion":"每个有效像素提供一项监督","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-05556ad03eab79c710d9d82c","nextBlock":"b-7cb961a56fa7777568b6bcbd","prerequisites":[{"id":"b-0c7cba43a95b7b0f276f0713","url":"#b-0c7cba43a95b7b0f276f0713","role":"section-title"},{"id":"b-af28245ff7c5891e33b674ff","url":"#b-af28245ff7c5891e33b674ff","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-7cb961a56fa7777568b6bcbd","type":"p","scope":"pixel-cross-entropy","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-cross-entropy.html#b-7cb961a56fa7777568b6bcbd","content":"图像翻转、裁剪或缩放时，目标地图必须接受同一几何变换。RGB 图能用连续插值，离散 mask 通常用最近邻；把类 1 与类 3 插出“类 2”没有语义依据。先保证一处图像像素仍对应同一处目标，训练曲线才有解释价值。","sha256":"75a26f3c334073d2322457d1f81001561d6c818c8f83b13b0f038ae1d77e5cc7","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"pixel-prediction-objective","title":"怎样直接学习一张像素类别地图？"},"page":{"id":"reading-opening","title":"每个有效像素提供一项监督","number":1},"sectionTitle":"每个有效像素提供一项监督","currentQuestion":"每个有效像素提供一项监督","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-bfa7f45f9031dc10c52b3a8f","nextBlock":"b-c8a8c7c0eddb664c190ef9d8","prerequisites":[{"id":"b-0c7cba43a95b7b0f276f0713","url":"#b-0c7cba43a95b7b0f276f0713","role":"section-title"},{"id":"b-af28245ff7c5891e33b674ff","url":"#b-af28245ff7c5891e33b674ff","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-c8a8c7c0eddb664c190ef9d8","type":"h2","scope":"pixel-cross-entropy","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-cross-entropy.html#b-c8a8c7c0eddb664c190ef9d8","content":"## 继续阅读与实验","sha256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"pixel-prediction-objective","title":"怎样直接学习一张像素类别地图？"},"page":{"id":"reading-opening","title":"每个有效像素提供一项监督","number":1},"sectionTitle":"每个有效像素提供一项监督","currentQuestion":"每个有效像素提供一项监督","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-7cb961a56fa7777568b6bcbd","nextBlock":null,"prerequisites":[{"id":"b-0c7cba43a95b7b0f276f0713","url":"#b-0c7cba43a95b7b0f276f0713","role":"section-title"},{"id":"b-af28245ff7c5891e33b674ff","url":"#b-af28245ff7c5891e33b674ff","role":"problem-context"}],"relatedBlocks":[]}]}