{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"Fast R-CNN：把区域放回共享计算中","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html","markdown":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.md","pages":[{"id":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","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":"explanation","targetId":"region-two-head-feedback","operation":"split-explanation-range"},{"oldId":"b-13547ae69f2f845ac369d54f","targetId":"b-13547ae69f2f845ac369d54f","page":"region-two-head-feedback","operation":"move","beforeSHA256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","afterSHA256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","reason":"Same source body in authored semantic page"},{"oldId":"b-8a1a1cc9dac8099ea1e6ab45","targetId":"b-8a1a1cc9dac8099ea1e6ab45","page":"region-two-head-feedback","operation":"move","beforeSHA256":"c3812e93813e9f6fcbbf264e82199bdaca19374c2778e1ffef7a45b794f26ac3","afterSHA256":"c3812e93813e9f6fcbbf264e82199bdaca19374c2778e1ffef7a45b794f26ac3","reason":"Same source body in authored semantic page"},{"oldId":"b-574f54c65a4f57c17b308651","targetId":"b-574f54c65a4f57c17b308651","page":"region-two-head-feedback","operation":"move","beforeSHA256":"a10e5e2e6d482cbef9ce4f17a89bba1ad5570d21cbc0682a29ab29fd083db01d","afterSHA256":"a10e5e2e6d482cbef9ce4f17a89bba1ad5570d21cbc0682a29ab29fd083db01d","reason":"Same source body in authored semantic page"},{"oldId":"b-1c3aa73655ee20eec556dd51","targetId":"b-1c3aa73655ee20eec556dd51","page":"region-two-head-feedback","operation":"rewrite","beforeSHA256":"3edf5ce22c1dbc2fe8a8f9a619c3483c39051146a324e0216711d3bb1e767512","afterSHA256":"f296afd233da986339bfbdfc4a809a81d636d632a2b98a34581b0a43af54ae17","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-fbba444623f74bd65c6caf15","targetId":"b-fbba444623f74bd65c6caf15","page":"region-two-head-feedback","operation":"move","beforeSHA256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","afterSHA256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","reason":"Same source body in authored semantic page"},{"oldId":"b-c58a978443fe2763df8a5b3c","targetId":"b-c58a978443fe2763df8a5b3c","page":"region-two-head-feedback","operation":"move","beforeSHA256":"8eef9dd1bd3e32d093fde478c1e0904aff097856a2a2661fbf1d0241d8b3540e","afterSHA256":"8eef9dd1bd3e32d093fde478c1e0904aff097856a2a2661fbf1d0241d8b3540e","reason":"Same source body in authored semantic page"},{"oldId":"b-f552dc1367db29b96e283ff2","targetId":"b-f552dc1367db29b96e283ff2","page":"region-two-head-feedback","operation":"move","beforeSHA256":"f918b89ffda44d7cfc0b97cf724291e6a3ac6980f8aba2d8fd881731798075fc","afterSHA256":"f918b89ffda44d7cfc0b97cf724291e6a3ac6980f8aba2d8fd881731798075fc","reason":"Same source body in authored semantic page"},{"oldId":"b-84ec4dff70a2be659c5734cb","targetId":"b-84ec4dff70a2be659c5734cb","page":"region-two-head-feedback","operation":"move","beforeSHA256":"87e1ddc05e492c9328d3500ad1c8e4b58ae471d196f166627f04c93bbb95942a","afterSHA256":"87e1ddc05e492c9328d3500ad1c8e4b58ae471d196f166627f04c93bbb95942a","reason":"Same source body in authored semantic page"},{"oldId":"b-7960cf114d94b4d293e5d04d","targetId":"b-7960cf114d94b4d293e5d04d","page":"region-two-head-feedback","operation":"move","beforeSHA256":"5d15802c85bd3afc462f6a460ea201d77b7264fb0fc52bb2487e477882bb056c","afterSHA256":"5d15802c85bd3afc462f6a460ea201d77b7264fb0fc52bb2487e477882bb056c","reason":"Same source body in authored semantic page"},{"oldId":"b-9d540512c7fef7ff9673efef","targetId":"b-9d540512c7fef7ff9673efef","page":"region-two-head-feedback","operation":"move","beforeSHA256":"279e8b655d392285cb663a9a1b5f511c83a2f241603a5f221d31282a48b80e01","afterSHA256":"279e8b655d392285cb663a9a1b5f511c83a2f241603a5f221d31282a48b80e01","reason":"Same source body in authored semantic page"},{"oldId":"b-870955997c0b18b95005e7ed","targetId":"b-870955997c0b18b95005e7ed","page":"region-two-head-feedback","operation":"move","beforeSHA256":"fd190ed6180e2e409aac31bdda229fafdfea61aff28409486ceabf54248099b1","afterSHA256":"fd190ed6180e2e409aac31bdda229fafdfea61aff28409486ceabf54248099b1","reason":"Same source body in authored semantic page"},{"oldId":"b-d87853bc8191b2c554a2ba8a","targetId":"b-d87853bc8191b2c554a2ba8a","page":"region-two-head-feedback","operation":"move","beforeSHA256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","afterSHA256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","reason":"Same source body in authored semantic page"},{"oldId":"b-2b3c79f47ba2d2e163822f93","targetId":"b-2b3c79f47ba2d2e163822f93","page":"region-two-head-feedback","operation":"move","beforeSHA256":"a3163e9b7618b50175088fce50e8c02bf38584dbcb466052e3f6745aa91110a4","afterSHA256":"a3163e9b7618b50175088fce50e8c02bf38584dbcb466052e3f6745aa91110a4","reason":"Same source body in authored semantic page"},{"oldId":"b-96de5e538b92d73c7194d1ec","targetId":"b-96de5e538b92d73c7194d1ec","page":"region-two-head-feedback","operation":"move","beforeSHA256":"5b80989076482697232771ebc2f272ec5c90129cc735bb40ff36afe6a9dd6ca2","afterSHA256":"5b80989076482697232771ebc2f272ec5c90129cc735bb40ff36afe6a9dd6ca2","reason":"Same source body in authored semantic page"},{"oldId":"region-two-head-feedback-title","targetId":"b-75e419af8e4c010fb32985f9","operation":"retain-source-identity"}],"blocks":[{"id":"b-75e419af8e4c010fb32985f9","type":"h2","scope":"region-two-head-feedback","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-75e419af8e4c010fb32985f9","content":"## 让多个区域共同训练共享骨干","sha256":"39b7f28cda35a46545ce0f1253ae9f13744fad69709246360882a8570794476a","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"让多个区域共同训练共享骨干","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-13547ae69f2f845ac369d54f","prerequisites":[{"id":"b-574f54c65a4f57c17b308651","url":"#b-574f54c65a4f57c17b308651","role":"problem-context"}],"relatedBlocks":[{"id":"b-1c3aa73655ee20eec556dd51","url":"#b-1c3aa73655ee20eec556dd51","relation":"figure-and-explanation"}]},{"id":"b-13547ae69f2f845ac369d54f","type":"h2","scope":"explanation","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-13547ae69f2f845ac369d54f","content":"## 解释、算例与证据","sha256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"让多个区域共同训练共享骨干","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-75e419af8e4c010fb32985f9","nextBlock":"b-8a1a1cc9dac8099ea1e6ab45","prerequisites":[{"id":"b-75e419af8e4c010fb32985f9","url":"#b-75e419af8e4c010fb32985f9","role":"section-title"},{"id":"b-574f54c65a4f57c17b308651","url":"#b-574f54c65a4f57c17b308651","role":"problem-context"}],"relatedBlocks":[{"id":"b-1c3aa73655ee20eec556dd51","url":"#b-1c3aa73655ee20eec556dd51","relation":"figure-and-explanation"}]},{"id":"b-8a1a1cc9dac8099ea1e6ab45","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-8a1a1cc9dac8099ea1e6ab45","content":"# P31 · Fast R-CNN：把区域放回共享计算中","sha256":"c3812e93813e9f6fcbbf264e82199bdaca19374c2778e1ffef7a45b794f26ac3","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"让多个区域共同训练共享骨干","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-13547ae69f2f845ac369d54f","nextBlock":"b-574f54c65a4f57c17b308651","prerequisites":[{"id":"b-13547ae69f2f845ac369d54f","url":"#b-13547ae69f2f845ac369d54f","role":"section-title"},{"id":"b-574f54c65a4f57c17b308651","url":"#b-574f54c65a4f57c17b308651","role":"problem-context"}],"relatedBlocks":[{"id":"b-1c3aa73655ee20eec556dd51","url":"#b-1c3aa73655ee20eec556dd51","relation":"figure-and-explanation"}]},{"id":"b-574f54c65a4f57c17b308651","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-574f54c65a4f57c17b308651","content":"Fast R-CNN 的改动包括共享卷积、区域读取和类别/位置反馈共同训练。图上的多个区域仍各有答案，合并的是可共用计算。先指出损失从哪里回到同一骨干，再理解为什么外部候选成为下一项瓶颈。","sha256":"a10e5e2e6d482cbef9ce4f17a89bba1ad5570d21cbc0682a29ab29fd083db01d","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"让多个区域共同训练共享骨干","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-8a1a1cc9dac8099ea1e6ab45","nextBlock":"b-1c3aa73655ee20eec556dd51","prerequisites":[{"id":"b-8a1a1cc9dac8099ea1e6ab45","url":"#b-8a1a1cc9dac8099ea1e6ab45","role":"section-title"}],"relatedBlocks":[{"id":"b-1c3aa73655ee20eec556dd51","url":"#b-1c3aa73655ee20eec556dd51","relation":"figure-and-explanation"}]},{"id":"b-1c3aa73655ee20eec556dd51","type":"figure","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-1c3aa73655ee20eec556dd51","content":"图内标签与数值：Lecture 03 课堂图解 P31；共享卷积特征；RoI Pool；类别；框修正；类别；框修正；两项任务反馈到共享骨干\n\n整图特征、两个不同位置的 RoI Pool、各自类别及框输出；第二状态显示任务反馈到共享骨干的路径。\n\n- 共享卷积特征\n- RoI Pool\n- 类别\n- 框修正\n- 类别\n- 框修正\n- 两项任务反馈到共享骨干整图特征、两个不同位置的 RoI Pool、各自类别及框输出；第二状态显示任务反馈到共享骨干的路径。 [查看对应图解状态](/#/l3-r2-p31/1)[查看完整图解](/course/figures/l3-r2-p31-1.html)","sha256":"f296afd233da986339bfbdfc4a809a81d636d632a2b98a34581b0a43af54ae17","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"让多个区域共同训练共享骨干","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":null,"formulas":[],"figure":{"caption":"整图特征、两个不同位置的 RoI Pool、各自类别及框输出；第二状态显示任务反馈到共享骨干的路径。 查看对应图解状态","labels":["共享卷积特征","RoI Pool","类别","框修正","类别","框修正","两项任务反馈到共享骨干"],"description":"整图特征、两个不同位置的 RoI Pool、各自类别及框输出；第二状态显示任务反馈到共享骨干的路径。共享卷积特征RoI Pool类别框修正类别框修正两项任务反馈到共享骨干","evidenceIdentity":"整图特征、两个不同位置的 RoI Pool、各自类别及框输出；第二状态显示任务反馈到共享骨干的路径。 查看对应图解状态","images":[{"src":"/assets/display/8ced939a1fd55047fbf73ea5.webp","alt":""}]},"previousBlock":"b-574f54c65a4f57c17b308651","nextBlock":"b-fbba444623f74bd65c6caf15","prerequisites":[{"id":"b-8a1a1cc9dac8099ea1e6ab45","url":"#b-8a1a1cc9dac8099ea1e6ab45","role":"section-title"},{"id":"b-574f54c65a4f57c17b308651","url":"#b-574f54c65a4f57c17b308651","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-fbba444623f74bd65c6caf15","type":"h2","scope":"narrative","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-fbba444623f74bd65c6caf15","content":"## 课堂讲述","sha256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"让多个区域共同训练共享骨干","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-1c3aa73655ee20eec556dd51","nextBlock":"b-c58a978443fe2763df8a5b3c","prerequisites":[{"id":"b-8a1a1cc9dac8099ea1e6ab45","url":"#b-8a1a1cc9dac8099ea1e6ab45","role":"section-title"},{"id":"b-574f54c65a4f57c17b308651","url":"#b-574f54c65a4f57c17b308651","role":"problem-context"}],"relatedBlocks":[{"id":"b-1c3aa73655ee20eec556dd51","url":"#b-1c3aa73655ee20eec556dd51","relation":"figure-and-explanation"}]},{"id":"b-c58a978443fe2763df8a5b3c","type":"p","scope":"narrative","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-c58a978443fe2763df8a5b3c","content":"Fast R-CNN 把整图特征、区域读取以及类别和框回归组织在同一个较紧凑的训练流程里。图片主要计算一次，候选区域在共享特征上读取，然后各自产生预测。\n学生容易只记住“更快”两个字。真正应该看到的是：系统把原来重复的计算搬到了共同部分，同时让区域任务的反馈回到共享表示。训练怎样组织，和速度怎样改善，在这里是连在一起的。\n但此时的候选区域仍然需要产生。当前面主要计算变得更高效，原来相对不显眼的步骤就可能变成新的瓶颈。接下来研究者问的不是再给这个网络起一个更快的名字，而是：候选本身能不能也从共享特征中学出来？","sha256":"8eef9dd1bd3e32d093fde478c1e0904aff097856a2a2661fbf1d0241d8b3540e","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"让多个区域共同训练共享骨干","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-fbba444623f74bd65c6caf15","nextBlock":"b-f552dc1367db29b96e283ff2","prerequisites":[{"id":"b-fbba444623f74bd65c6caf15","url":"#b-fbba444623f74bd65c6caf15","role":"section-title"},{"id":"b-574f54c65a4f57c17b308651","url":"#b-574f54c65a4f57c17b308651","role":"problem-context"}],"relatedBlocks":[{"id":"b-1c3aa73655ee20eec556dd51","url":"#b-1c3aa73655ee20eec556dd51","relation":"figure-and-explanation"}]},{"id":"b-f552dc1367db29b96e283ff2","type":"h3","scope":"fast-rcnn-two-heads","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-f552dc1367db29b96e283ff2","content":"### 共享特征怎样收到每个区域的训练反馈","sha256":"f918b89ffda44d7cfc0b97cf724291e6a3ac6980f8aba2d8fd881731798075fc","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"共享特征怎样收到每个区域的训练反馈","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":"fast-rcnn-two-heads","formulas":[],"previousBlock":"b-c58a978443fe2763df8a5b3c","nextBlock":"b-84ec4dff70a2be659c5734cb","prerequisites":[{"id":"b-fbba444623f74bd65c6caf15","url":"#b-fbba444623f74bd65c6caf15","role":"section-title"},{"id":"b-574f54c65a4f57c17b308651","url":"#b-574f54c65a4f57c17b308651","role":"problem-context"}],"relatedBlocks":[{"id":"b-1c3aa73655ee20eec556dd51","url":"#b-1c3aa73655ee20eec556dd51","relation":"figure-and-explanation"},{"id":"b-84ec4dff70a2be659c5734cb","url":"#b-84ec4dff70a2be659c5734cb","relation":"same-reasoning"},{"id":"b-7960cf114d94b4d293e5d04d","url":"#b-7960cf114d94b4d293e5d04d","relation":"same-reasoning"},{"id":"b-9d540512c7fef7ff9673efef","url":"#b-9d540512c7fef7ff9673efef","relation":"same-reasoning"},{"id":"b-870955997c0b18b95005e7ed","url":"#b-870955997c0b18b95005e7ed","relation":"same-reasoning"}]},{"id":"b-84ec4dff70a2be659c5734cb","type":"p","scope":"fast-rcnn-two-heads","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-84ec4dff70a2be659c5734cb","content":"从整图特征取到每个 RoI 后，分类分支判断类别或背景，框分支预测相对位置修正。设某个区域属于杯子，损失一部分来自“杯子”概率是否足够大，另一部分来自中心、宽高偏移是否接近匹配目标。背景区域参与分类，通常不承担指向某个对象的位置回归。","sha256":"87e1ddc05e492c9328d3500ad1c8e4b58ae471d196f166627f04c93bbb95942a","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"共享特征怎样收到每个区域的训练反馈","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":"fast-rcnn-two-heads","formulas":[],"previousBlock":"b-f552dc1367db29b96e283ff2","nextBlock":"b-7960cf114d94b4d293e5d04d","prerequisites":[{"id":"b-f552dc1367db29b96e283ff2","url":"#b-f552dc1367db29b96e283ff2","role":"section-title"},{"id":"b-574f54c65a4f57c17b308651","url":"#b-574f54c65a4f57c17b308651","role":"problem-context"}],"relatedBlocks":[{"id":"b-1c3aa73655ee20eec556dd51","url":"#b-1c3aa73655ee20eec556dd51","relation":"figure-and-explanation"},{"id":"b-f552dc1367db29b96e283ff2","url":"#b-f552dc1367db29b96e283ff2","relation":"same-reasoning"},{"id":"b-7960cf114d94b4d293e5d04d","url":"#b-7960cf114d94b4d293e5d04d","relation":"same-reasoning"},{"id":"b-9d540512c7fef7ff9673efef","url":"#b-9d540512c7fef7ff9673efef","relation":"same-reasoning"},{"id":"b-870955997c0b18b95005e7ed","url":"#b-870955997c0b18b95005e7ed","relation":"same-reasoning"}]},{"id":"b-7960cf114d94b4d293e5d04d","type":"p","scope":"fast-rcnn-two-heads","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-7960cf114d94b4d293e5d04d","content":"把一批区域的损失相加后，梯度沿各自的区域读出回到共享骨干。共享的参数会收到来自不同区域、不同任务的反馈；它并不要求所有区域有相同答案。正如同一把尺子测不同物体可以得到不同长度，同一个预测器处理不同区域特征也会给出不同结果。","sha256":"5d15802c85bd3afc462f6a460ea201d77b7264fb0fc52bb2487e477882bb056c","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"共享特征怎样收到每个区域的训练反馈","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":"fast-rcnn-two-heads","formulas":[],"previousBlock":"b-84ec4dff70a2be659c5734cb","nextBlock":"b-9d540512c7fef7ff9673efef","prerequisites":[{"id":"b-f552dc1367db29b96e283ff2","url":"#b-f552dc1367db29b96e283ff2","role":"section-title"},{"id":"b-574f54c65a4f57c17b308651","url":"#b-574f54c65a4f57c17b308651","role":"problem-context"}],"relatedBlocks":[{"id":"b-1c3aa73655ee20eec556dd51","url":"#b-1c3aa73655ee20eec556dd51","relation":"figure-and-explanation"},{"id":"b-f552dc1367db29b96e283ff2","url":"#b-f552dc1367db29b96e283ff2","relation":"same-reasoning"},{"id":"b-84ec4dff70a2be659c5734cb","url":"#b-84ec4dff70a2be659c5734cb","relation":"same-reasoning"},{"id":"b-9d540512c7fef7ff9673efef","url":"#b-9d540512c7fef7ff9673efef","relation":"same-reasoning"},{"id":"b-870955997c0b18b95005e7ed","url":"#b-870955997c0b18b95005e7ed","relation":"same-reasoning"}]},{"id":"b-9d540512c7fef7ff9673efef","type":"p","scope":"fast-rcnn-two-heads","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-9d540512c7fef7ff9673efef","content":"Fast R-CNN 保留了外部候选来源，但把共享卷积、RoI 读取、分类和框回归组织进更紧凑的训练流程。卷积不再为每个重叠区域重复做，候选提取的时间便更加突出。下一步自然不是继续增加分类头，而是让候选也从这张 F 中学习。","sha256":"279e8b655d392285cb663a9a1b5f511c83a2f241603a5f221d31282a48b80e01","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"共享特征怎样收到每个区域的训练反馈","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":"fast-rcnn-two-heads","formulas":[],"previousBlock":"b-7960cf114d94b4d293e5d04d","nextBlock":"b-870955997c0b18b95005e7ed","prerequisites":[{"id":"b-f552dc1367db29b96e283ff2","url":"#b-f552dc1367db29b96e283ff2","role":"section-title"},{"id":"b-574f54c65a4f57c17b308651","url":"#b-574f54c65a4f57c17b308651","role":"problem-context"}],"relatedBlocks":[{"id":"b-1c3aa73655ee20eec556dd51","url":"#b-1c3aa73655ee20eec556dd51","relation":"figure-and-explanation"},{"id":"b-f552dc1367db29b96e283ff2","url":"#b-f552dc1367db29b96e283ff2","relation":"same-reasoning"},{"id":"b-84ec4dff70a2be659c5734cb","url":"#b-84ec4dff70a2be659c5734cb","relation":"same-reasoning"},{"id":"b-7960cf114d94b4d293e5d04d","url":"#b-7960cf114d94b4d293e5d04d","relation":"same-reasoning"},{"id":"b-870955997c0b18b95005e7ed","url":"#b-870955997c0b18b95005e7ed","relation":"same-reasoning"}]},{"id":"b-870955997c0b18b95005e7ed","type":"p","scope":"fast-rcnn-two-heads","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-870955997c0b18b95005e7ed","content":"理解这一页时，能够在图上指出“两个区域各自的输出”和“同一个被共同更新的骨干”，就已经抓住它的机制。历史多阶段实现细节与训练采样放在 CLS07，而不加回主讲页。","sha256":"fd190ed6180e2e409aac31bdda229fafdfea61aff28409486ceabf54248099b1","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"共享特征怎样收到每个区域的训练反馈","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":"fast-rcnn-two-heads","formulas":[],"previousBlock":"b-9d540512c7fef7ff9673efef","nextBlock":"b-d87853bc8191b2c554a2ba8a","prerequisites":[{"id":"b-f552dc1367db29b96e283ff2","url":"#b-f552dc1367db29b96e283ff2","role":"section-title"},{"id":"b-574f54c65a4f57c17b308651","url":"#b-574f54c65a4f57c17b308651","role":"problem-context"}],"relatedBlocks":[{"id":"b-1c3aa73655ee20eec556dd51","url":"#b-1c3aa73655ee20eec556dd51","relation":"figure-and-explanation"},{"id":"b-f552dc1367db29b96e283ff2","url":"#b-f552dc1367db29b96e283ff2","relation":"same-reasoning"},{"id":"b-84ec4dff70a2be659c5734cb","url":"#b-84ec4dff70a2be659c5734cb","relation":"same-reasoning"},{"id":"b-7960cf114d94b4d293e5d04d","url":"#b-7960cf114d94b4d293e5d04d","relation":"same-reasoning"},{"id":"b-9d540512c7fef7ff9673efef","url":"#b-9d540512c7fef7ff9673efef","relation":"same-reasoning"}]},{"id":"b-d87853bc8191b2c554a2ba8a","type":"h2","scope":"practice","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-d87853bc8191b2c554a2ba8a","content":"## 沿这个问题继续","sha256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"让多个区域共同训练共享骨干","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-870955997c0b18b95005e7ed","nextBlock":"b-2b3c79f47ba2d2e163822f93","prerequisites":[{"id":"b-f552dc1367db29b96e283ff2","url":"#b-f552dc1367db29b96e283ff2","role":"section-title"},{"id":"b-574f54c65a4f57c17b308651","url":"#b-574f54c65a4f57c17b308651","role":"problem-context"}],"relatedBlocks":[{"id":"b-1c3aa73655ee20eec556dd51","url":"#b-1c3aa73655ee20eec556dd51","relation":"figure-and-explanation"}]},{"id":"b-2b3c79f47ba2d2e163822f93","type":"p","scope":"practice","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-2b3c79f47ba2d2e163822f93","content":"沿 CLS07 的公式定位疑问，再选 LAB09 解析练习或 LAB10 完整训练。","sha256":"a3163e9b7618b50175088fce50e8c02bf38584dbcb466052e3f6745aa91110a4","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"让多个区域共同训练共享骨干","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-d87853bc8191b2c554a2ba8a","nextBlock":"b-0bbd2f6925e76138017ece55","prerequisites":[{"id":"b-d87853bc8191b2c554a2ba8a","url":"#b-d87853bc8191b2c554a2ba8a","role":"section-title"},{"id":"b-574f54c65a4f57c17b308651","url":"#b-574f54c65a4f57c17b308651","role":"problem-context"}],"relatedBlocks":[{"id":"b-1c3aa73655ee20eec556dd51","url":"#b-1c3aa73655ee20eec556dd51","relation":"figure-and-explanation"}]},{"id":"b-0bbd2f6925e76138017ece55","type":"h4","scope":"region-two-head-feedback","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-0bbd2f6925e76138017ece55","content":"#### 图示条件与公式约定","sha256":"ef7f915c1ae63ed67bd23fecf050249273ba0efa8965d2c8d1c1c6b2d3178122","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"让多个区域共同训练共享骨干","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-2b3c79f47ba2d2e163822f93","nextBlock":"b-96de5e538b92d73c7194d1ec","prerequisites":[{"id":"b-d87853bc8191b2c554a2ba8a","url":"#b-d87853bc8191b2c554a2ba8a","role":"section-title"},{"id":"b-574f54c65a4f57c17b308651","url":"#b-574f54c65a4f57c17b308651","role":"problem-context"}],"relatedBlocks":[{"id":"b-1c3aa73655ee20eec556dd51","url":"#b-1c3aa73655ee20eec556dd51","relation":"figure-and-explanation"}]},{"id":"b-96de5e538b92d73c7194d1ec","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p31.html#b-96de5e538b92d73c7194d1ec","content":"主图展示共享特征、区域读出及类别/回归的联合训练思想，候选仍来自外部方法。反馈线表示总损失更新共享计算，不表示多个区域同一个答案。","sha256":"5b80989076482697232771ebc2f272ec5c90129cc735bb40ff36afe6a9dd6ca2","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":"region-two-head-feedback","title":"让多个区域共同训练共享骨干","number":1},"sectionTitle":"让多个区域共同训练共享骨干","currentQuestion":"候选、训练匹配和框偏移分别承担哪一步？","sourceSlides":["l3-r2-p31"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-0bbd2f6925e76138017ece55","nextBlock":null,"prerequisites":[{"id":"b-0bbd2f6925e76138017ece55","url":"#b-0bbd2f6925e76138017ece55","role":"section-title"},{"id":"b-574f54c65a4f57c17b308651","url":"#b-574f54c65a4f57c17b308651","role":"problem-context"}],"relatedBlocks":[{"id":"b-1c3aa73655ee20eec556dd51","url":"#b-1c3aa73655ee20eec556dd51","relation":"figure-and-explanation"}]}]}