{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/blocks/rpn-three-boxes.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"参考框、候选框与最终检测框","url":"https://codingai-lec03.pages.dev/course/blocks/rpn-three-boxes.html","markdown":"https://codingai-lec03.pages.dev/course/blocks/rpn-three-boxes.md","pages":[{"id":"reading-opening","title":"参考框、候选框与最终检测框","unit":"类别与位置怎样成为可学习的区域目标？","unitId":"learnable-box-prediction","chapter":"II","number":1}],"aliases":[{"oldId":"b-97e10c138b60ee1074393220","targetId":"b-97e10c138b60ee1074393220","page":"reading-opening","operation":"move","beforeSHA256":"eec56a90c171ef4302d769b8084c62b15762d4590057691cc335cc1151bf5932","afterSHA256":"eec56a90c171ef4302d769b8084c62b15762d4590057691cc335cc1151bf5932","reason":"Same source body in authored semantic page"},{"oldId":"b-d2fc47f5c96c3e18ebc25078","targetId":"b-d2fc47f5c96c3e18ebc25078","page":"reading-opening","operation":"move","beforeSHA256":"6b45cf47a23e9eec2f15663b9a1624db57adbc4f2a6825dd6bc21c96bfb93aa4","afterSHA256":"6b45cf47a23e9eec2f15663b9a1624db57adbc4f2a6825dd6bc21c96bfb93aa4","reason":"Same source body in authored semantic page"},{"oldId":"b-0d84472b9472485950125102","targetId":"b-0d84472b9472485950125102","page":"reading-opening","operation":"move","beforeSHA256":"ff6018e332450f304b746defc546d270e531208c2bd0354e7e32a269661363c1","afterSHA256":"ff6018e332450f304b746defc546d270e531208c2bd0354e7e32a269661363c1","reason":"Same source body in authored semantic page"},{"oldId":"b-c759c065ee574f690e179937","targetId":"b-c759c065ee574f690e179937","page":"reading-opening","operation":"move","beforeSHA256":"ec5bbb4f56f3be17858a079f938ad06774395a07f9f46cda16ebf9c2538d0144","afterSHA256":"ec5bbb4f56f3be17858a079f938ad06774395a07f9f46cda16ebf9c2538d0144","reason":"Same source body in authored semantic page"},{"oldId":"b-662284b08f6794c1df56197f","targetId":"b-662284b08f6794c1df56197f","page":"reading-opening","operation":"move","beforeSHA256":"87e8f6e090d6b2bb0be5d6005c6ef7fae57d2b05e1c4df5c825928472ed755bf","afterSHA256":"87e8f6e090d6b2bb0be5d6005c6ef7fae57d2b05e1c4df5c825928472ed755bf","reason":"Same source body in authored semantic page"},{"oldId":"b-34bbe0ec19a0a6cf13d2e3a2","targetId":"b-34bbe0ec19a0a6cf13d2e3a2","page":"reading-opening","operation":"move","beforeSHA256":"d39d8ade1a97f74de22f5548d14b4a3ca83a4cff593b2c987f0b8ead3d6a7d6e","afterSHA256":"d39d8ade1a97f74de22f5548d14b4a3ca83a4cff593b2c987f0b8ead3d6a7d6e","reason":"Same source body in authored semantic page"},{"oldId":"b-45b32a886a77dca6404c2811","targetId":"b-45b32a886a77dca6404c2811","page":"reading-opening","operation":"rewrite","beforeSHA256":"4686233cd15e1445f7884588c004d9f4b85d7b582e04f9252bdb4c1c97773996","afterSHA256":"c96da081bf7874a144c100311a622d5788f5cf7663428ead2beb7c54db768660","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-d85c7ecf572b6002a1c67a91","targetId":"b-d85c7ecf572b6002a1c67a91","page":"reading-opening","operation":"move","beforeSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","afterSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","reason":"Same source body in authored semantic page"}],"blocks":[{"id":"b-97e10c138b60ee1074393220","type":"h1","scope":"rpn-three-boxes","url":"https://codingai-lec03.pages.dev/course/blocks/rpn-three-boxes.html#b-97e10c138b60ee1074393220","content":"# 参考框、候选框与最终检测框","sha256":"eec56a90c171ef4302d769b8084c62b15762d4590057691cc335cc1151bf5932","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"learnable-box-prediction","title":"类别与位置怎样成为可学习的区域目标？"},"page":{"id":"reading-opening","title":"参考框、候选框与最终检测框","number":1},"sectionTitle":"参考框、候选框与最终检测框","currentQuestion":"参考框、候选框与最终检测框","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-d2fc47f5c96c3e18ebc25078","prerequisites":[{"id":"b-d2fc47f5c96c3e18ebc25078","url":"#b-d2fc47f5c96c3e18ebc25078","role":"problem-context"}],"relatedBlocks":[{"id":"b-45b32a886a77dca6404c2811","url":"#b-45b32a886a77dca6404c2811","relation":"figure-and-explanation"}]},{"id":"b-d2fc47f5c96c3e18ebc25078","type":"p","scope":"rpn-three-boxes","url":"https://codingai-lec03.pages.dev/course/blocks/rpn-three-boxes.html#b-d2fc47f5c96c3e18ebc25078","content":"按“参考框→RPN 候选→最终框”给图中的矩形分配身份，并把人工真值单独留在监督侧。RPN 让候选变为可学习预测。LAB10 采用共享原理的一阶段教学变体，不应把它说成原版 Faster R-CNN。","sha256":"6b45cf47a23e9eec2f15663b9a1624db57adbc4f2a6825dd6bc21c96bfb93aa4","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"learnable-box-prediction","title":"类别与位置怎样成为可学习的区域目标？"},"page":{"id":"reading-opening","title":"参考框、候选框与最终检测框","number":1},"sectionTitle":"参考框、候选框与最终检测框","currentQuestion":"参考框、候选框与最终检测框","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-97e10c138b60ee1074393220","nextBlock":"b-0d84472b9472485950125102","prerequisites":[{"id":"b-97e10c138b60ee1074393220","url":"#b-97e10c138b60ee1074393220","role":"section-title"}],"relatedBlocks":[{"id":"b-45b32a886a77dca6404c2811","url":"#b-45b32a886a77dca6404c2811","relation":"figure-and-explanation"}]},{"id":"b-0d84472b9472485950125102","type":"p","scope":"rpn-three-boxes","url":"https://codingai-lec03.pages.dev/course/blocks/rpn-three-boxes.html#b-0d84472b9472485950125102","content":"RPN 在共享特征图上铺一组参考框。参考框由网格位置、预设尺度和长宽比决定；它们在看图之前就可生成，不能当成检测结果。网络为每个参考框输出物体性分数和四个偏移，经解码、筛选后形成 proposals，再交给区域头作更具体的类别判断和位置修正。","sha256":"ff6018e332450f304b746defc546d270e531208c2bd0354e7e32a269661363c1","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"learnable-box-prediction","title":"类别与位置怎样成为可学习的区域目标？"},"page":{"id":"reading-opening","title":"参考框、候选框与最终检测框","number":1},"sectionTitle":"参考框、候选框与最终检测框","currentQuestion":"参考框、候选框与最终检测框","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-d2fc47f5c96c3e18ebc25078","nextBlock":"b-c759c065ee574f690e179937","prerequisites":[{"id":"b-97e10c138b60ee1074393220","url":"#b-97e10c138b60ee1074393220","role":"section-title"},{"id":"b-d2fc47f5c96c3e18ebc25078","url":"#b-d2fc47f5c96c3e18ebc25078","role":"problem-context"}],"relatedBlocks":[{"id":"b-45b32a886a77dca6404c2811","url":"#b-45b32a886a77dca6404c2811","relation":"figure-and-explanation"}]},{"id":"b-c759c065ee574f690e179937","type":"p","scope":"rpn-three-boxes","url":"https://codingai-lec03.pages.dev/course/blocks/rpn-three-boxes.html#b-c759c065ee574f690e179937","content":"因此画面中的三个矩形有不同身份：anchor 是几何起点，proposal 是 RPN 的预测，最终 box 是整个检测流程交付的结果。训练真值是第四种对象，它负责给预测规定目标，并不在推理时送进模型告诉它答案。","sha256":"ec5bbb4f56f3be17858a079f938ad06774395a07f9f46cda16ebf9c2538d0144","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"learnable-box-prediction","title":"类别与位置怎样成为可学习的区域目标？"},"page":{"id":"reading-opening","title":"参考框、候选框与最终检测框","number":1},"sectionTitle":"参考框、候选框与最终检测框","currentQuestion":"参考框、候选框与最终检测框","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-0d84472b9472485950125102","nextBlock":"b-662284b08f6794c1df56197f","prerequisites":[{"id":"b-97e10c138b60ee1074393220","url":"#b-97e10c138b60ee1074393220","role":"section-title"},{"id":"b-d2fc47f5c96c3e18ebc25078","url":"#b-d2fc47f5c96c3e18ebc25078","role":"problem-context"}],"relatedBlocks":[{"id":"b-45b32a886a77dca6404c2811","url":"#b-45b32a886a77dca6404c2811","relation":"figure-and-explanation"}]},{"id":"b-662284b08f6794c1df56197f","type":"p","scope":"rpn-three-boxes","url":"https://codingai-lec03.pages.dev/course/blocks/rpn-three-boxes.html#b-662284b08f6794c1df56197f","content":"同一张 F 可同时供 RPN 与后级区域头使用。它把“在哪里值得再看”从外部算法变成从图像特征学习的任务。这是 Faster R-CNN 与 Fast R-CNN 相比发生改变的具体一段，不能只记一个更快的名字。","sha256":"87e8f6e090d6b2bb0be5d6005c6ef7fae57d2b05e1c4df5c825928472ed755bf","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"learnable-box-prediction","title":"类别与位置怎样成为可学习的区域目标？"},"page":{"id":"reading-opening","title":"参考框、候选框与最终检测框","number":1},"sectionTitle":"参考框、候选框与最终检测框","currentQuestion":"参考框、候选框与最终检测框","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c759c065ee574f690e179937","nextBlock":"b-34bbe0ec19a0a6cf13d2e3a2","prerequisites":[{"id":"b-97e10c138b60ee1074393220","url":"#b-97e10c138b60ee1074393220","role":"section-title"},{"id":"b-d2fc47f5c96c3e18ebc25078","url":"#b-d2fc47f5c96c3e18ebc25078","role":"problem-context"}],"relatedBlocks":[{"id":"b-45b32a886a77dca6404c2811","url":"#b-45b32a886a77dca6404c2811","relation":"figure-and-explanation"}]},{"id":"b-34bbe0ec19a0a6cf13d2e3a2","type":"p","scope":"rpn-three-boxes","url":"https://codingai-lec03.pages.dev/course/blocks/rpn-three-boxes.html#b-34bbe0ec19a0a6cf13d2e3a2","content":"LAB10 采用单层锚框直接分类与回归的一阶段教学检测器，不是原版两阶段 RPN。它与本页共享锚框、匹配、位置损失等原理，保留了可运行的学习闭环；读实验时应辨明哪些流程共用、哪些结构不同。","sha256":"d39d8ade1a97f74de22f5548d14b4a3ca83a4cff593b2c987f0b8ead3d6a7d6e","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"learnable-box-prediction","title":"类别与位置怎样成为可学习的区域目标？"},"page":{"id":"reading-opening","title":"参考框、候选框与最终检测框","number":1},"sectionTitle":"参考框、候选框与最终检测框","currentQuestion":"参考框、候选框与最终检测框","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-662284b08f6794c1df56197f","nextBlock":"b-45b32a886a77dca6404c2811","prerequisites":[{"id":"b-97e10c138b60ee1074393220","url":"#b-97e10c138b60ee1074393220","role":"section-title"},{"id":"b-d2fc47f5c96c3e18ebc25078","url":"#b-d2fc47f5c96c3e18ebc25078","role":"problem-context"}],"relatedBlocks":[{"id":"b-45b32a886a77dca6404c2811","url":"#b-45b32a886a77dca6404c2811","relation":"figure-and-explanation"}]},{"id":"b-45b32a886a77dca6404c2811","type":"figure","scope":"rpn-three-boxes","url":"https://codingai-lec03.pages.dev/course/blocks/rpn-three-boxes.html#b-45b32a886a77dca6404c2811","content":"图内标签与数值：Lecture 03 课堂图解 P32；共享 F；RPN；前景分数 + 框偏移；区域读出；分类 + 最终框修正；候选框 → 供后级使用；参考框；参考几何 → 预测候选 → 最终检测框\n\n从 reference 到 proposal，再到最终 detection；真值只用于训练和评价。 查看对应图解状态\n\n- 共享 F\n- RPN\n- 前景分数 + 框偏移\n- 区域读出\n- 分类 + 最终框修正\n- 候选框 → 供后级使用\n- 参考框\n- 参考几何 → 预测候选 → 最终检测框从 reference 到 proposal，再到最终 detection；真值只用于训练和评价。 [查看对应图解状态](/#/l3-r2-p32/1)[查看完整图解](/#/l3-r2-p32/1)","sha256":"c96da081bf7874a144c100311a622d5788f5cf7663428ead2beb7c54db768660","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"learnable-box-prediction","title":"类别与位置怎样成为可学习的区域目标？"},"page":{"id":"reading-opening","title":"参考框、候选框与最终检测框","number":1},"sectionTitle":"参考框、候选框与最终检测框","currentQuestion":"参考框、候选框与最终检测框","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"figure":{"caption":"从 reference 到 proposal，再到最终 detection；真值只用于训练和评价。 查看对应图解状态","labels":["共享 F","RPN","前景分数 + 框偏移","区域读出","分类 + 最终框修正","候选框 → 供后级使用","参考框","参考几何 → 预测候选 → 最终检测框"],"description":"从 reference 到 proposal，再到最终 detection；真值只用于训练和评价。 查看对应图解状态共享 FRPN前景分数 + 框偏移区域读出分类 + 最终框修正候选框 → 供后级使用参考框参考几何 → 预测候选 → 最终检测框","evidenceIdentity":"从 reference 到 proposal，再到最终 detection；真值只用于训练和评价。 查看对应图解状态","images":[{"src":"/assets/display/8ced939a1fd55047fbf73ea5.webp","alt":""}]},"previousBlock":"b-34bbe0ec19a0a6cf13d2e3a2","nextBlock":"b-d85c7ecf572b6002a1c67a91","prerequisites":[{"id":"b-97e10c138b60ee1074393220","url":"#b-97e10c138b60ee1074393220","role":"section-title"},{"id":"b-d2fc47f5c96c3e18ebc25078","url":"#b-d2fc47f5c96c3e18ebc25078","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-d85c7ecf572b6002a1c67a91","type":"h2","scope":"rpn-three-boxes","url":"https://codingai-lec03.pages.dev/course/blocks/rpn-three-boxes.html#b-d85c7ecf572b6002a1c67a91","content":"## 继续阅读与实验","sha256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"learnable-box-prediction","title":"类别与位置怎样成为可学习的区域目标？"},"page":{"id":"reading-opening","title":"参考框、候选框与最终检测框","number":1},"sectionTitle":"参考框、候选框与最终检测框","currentQuestion":"参考框、候选框与最终检测框","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-45b32a886a77dca6404c2811","nextBlock":null,"prerequisites":[{"id":"b-97e10c138b60ee1074393220","url":"#b-97e10c138b60ee1074393220","role":"section-title"},{"id":"b-d2fc47f5c96c3e18ebc25078","url":"#b-d2fc47f5c96c3e18ebc25078","role":"problem-context"}],"relatedBlocks":[{"id":"b-45b32a886a77dca6404c2811","url":"#b-45b32a886a77dca6404c2811","relation":"figure-and-explanation"}]}]}