{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/blocks/rcnn-transfer-pipeline.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"R-CNN 迁移的是已学会的视觉计算","url":"https://codingai-lec03.pages.dev/course/blocks/rcnn-transfer-pipeline.html","markdown":"https://codingai-lec03.pages.dev/course/blocks/rcnn-transfer-pipeline.md","pages":[{"id":"reading-opening","title":"R-CNN 迁移的是已学会的视觉计算","unit":"多个区域怎样复用同一份视觉计算？","unitId":"region-computation-reuse","chapter":"II","number":1}],"aliases":[{"oldId":"b-65a3a4988a889e8c39295dae","targetId":"b-65a3a4988a889e8c39295dae","page":"reading-opening","operation":"move","beforeSHA256":"3b8b8b419865456f71d786f07ba51eef94e7011438ecead077abe7acaec9f581","afterSHA256":"3b8b8b419865456f71d786f07ba51eef94e7011438ecead077abe7acaec9f581","reason":"Same source body in authored semantic page"},{"oldId":"b-100d2b561a639d0f25cc3f3f","targetId":"b-100d2b561a639d0f25cc3f3f","page":"reading-opening","operation":"move","beforeSHA256":"f5e46123e517748cb0b2feb64ecff80b4cc4c546f33d59cc42356db94db6d2c9","afterSHA256":"f5e46123e517748cb0b2feb64ecff80b4cc4c546f33d59cc42356db94db6d2c9","reason":"Same source body in authored semantic page"},{"oldId":"b-f8b817f1bdc58cc7e1e3f9ab","targetId":"b-f8b817f1bdc58cc7e1e3f9ab","page":"reading-opening","operation":"move","beforeSHA256":"a620c6e0f57f9d21a737c4af613fe460cac7e0d69e0cf0248c6827ec83e987db","afterSHA256":"a620c6e0f57f9d21a737c4af613fe460cac7e0d69e0cf0248c6827ec83e987db","reason":"Same source body in authored semantic page"},{"oldId":"b-e55cc5881ea9b0b043489c2a","targetId":"b-e55cc5881ea9b0b043489c2a","page":"reading-opening","operation":"move","beforeSHA256":"861774ba5faaffdebcc9f839c46f4495cc31d8be666790b3138545d99ce00d44","afterSHA256":"861774ba5faaffdebcc9f839c46f4495cc31d8be666790b3138545d99ce00d44","reason":"Same source body in authored semantic page"},{"oldId":"b-e888eeb1223080921ba60816","targetId":"b-e888eeb1223080921ba60816","page":"reading-opening","operation":"move","beforeSHA256":"178bc2569407866ae7371a893269c1c7e57bc6d16f7f1c044f037e1730c1062d","afterSHA256":"178bc2569407866ae7371a893269c1c7e57bc6d16f7f1c044f037e1730c1062d","reason":"Same source body in authored semantic page"},{"oldId":"b-5f63be66386a4a103f0665d3","targetId":"b-5f63be66386a4a103f0665d3","page":"reading-opening","operation":"rewrite","beforeSHA256":"ed9dce252be30f71b5a6c7ccc76e3a07557f616c4c8e407d2ca561cc0602e93a","afterSHA256":"7884d881b406d2bf34ec435ee6bd142637e4a25553b776f44d5e30e839c7762c","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-2c8b940b9531749b141efc2d","targetId":"b-2c8b940b9531749b141efc2d","page":"reading-opening","operation":"move","beforeSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","afterSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","reason":"Same source body in authored semantic page"}],"blocks":[{"id":"b-65a3a4988a889e8c39295dae","type":"h1","scope":"rcnn-transfer-pipeline","url":"https://codingai-lec03.pages.dev/course/blocks/rcnn-transfer-pipeline.html#b-65a3a4988a889e8c39295dae","content":"# R-CNN 迁移的是已学会的视觉计算","sha256":"3b8b8b419865456f71d786f07ba51eef94e7011438ecead077abe7acaec9f581","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"reading-opening","title":"R-CNN 迁移的是已学会的视觉计算","number":1},"sectionTitle":"R-CNN 迁移的是已学会的视觉计算","currentQuestion":"R-CNN 迁移的是已学会的视觉计算","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-100d2b561a639d0f25cc3f3f","prerequisites":[{"id":"b-100d2b561a639d0f25cc3f3f","url":"#b-100d2b561a639d0f25cc3f3f","role":"problem-context"}],"relatedBlocks":[{"id":"b-5f63be66386a4a103f0665d3","url":"#b-5f63be66386a4a103f0665d3","relation":"figure-and-explanation"}]},{"id":"b-100d2b561a639d0f25cc3f3f","type":"p","scope":"rcnn-transfer-pipeline","url":"https://codingai-lec03.pages.dev/course/blocks/rcnn-transfer-pipeline.html#b-100d2b561a639d0f25cc3f3f","content":"历史上 R-CNN 将监督分类预训练用于区域检测，支持了“先学表示，再适配新任务”的正面判断。按区域提取特征、学习前景/背景和位置修正三个动作都需要具体输入。它不是 MoCo 或 MAE 的历史后继。","sha256":"f5e46123e517748cb0b2feb64ecff80b4cc4c546f33d59cc42356db94db6d2c9","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"reading-opening","title":"R-CNN 迁移的是已学会的视觉计算","number":1},"sectionTitle":"R-CNN 迁移的是已学会的视觉计算","currentQuestion":"R-CNN 迁移的是已学会的视觉计算","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-65a3a4988a889e8c39295dae","nextBlock":"b-f8b817f1bdc58cc7e1e3f9ab","prerequisites":[{"id":"b-65a3a4988a889e8c39295dae","url":"#b-65a3a4988a889e8c39295dae","role":"section-title"}],"relatedBlocks":[{"id":"b-5f63be66386a4a103f0665d3","url":"#b-5f63be66386a4a103f0665d3","relation":"figure-and-explanation"}]},{"id":"b-f8b817f1bdc58cc7e1e3f9ab","type":"p","scope":"rcnn-transfer-pipeline","url":"https://codingai-lec03.pages.dev/course/blocks/rcnn-transfer-pipeline.html#b-f8b817f1bdc58cc7e1e3f9ab","content":"R-CNN 先产生可能包含对象的区域，将每个区域变换为 CNN 接受的图像大小，再用 CNN 特征作区域分类与框修正。分类预训练提供的是卷积网络的参数；检测适配需要新的区域样本、前景/背景关系和位置目标。原始实现包含多个训练阶段，因此课堂画面中简洁的两项输出，不应被误读成原论文从一开始就是一个统一损失的端到端系统。","sha256":"a620c6e0f57f9d21a737c4af613fe460cac7e0d69e0cf0248c6827ec83e987db","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"reading-opening","title":"R-CNN 迁移的是已学会的视觉计算","number":1},"sectionTitle":"R-CNN 迁移的是已学会的视觉计算","currentQuestion":"R-CNN 迁移的是已学会的视觉计算","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-100d2b561a639d0f25cc3f3f","nextBlock":"b-e55cc5881ea9b0b043489c2a","prerequisites":[{"id":"b-65a3a4988a889e8c39295dae","url":"#b-65a3a4988a889e8c39295dae","role":"section-title"},{"id":"b-100d2b561a639d0f25cc3f3f","url":"#b-100d2b561a639d0f25cc3f3f","role":"problem-context"}],"relatedBlocks":[{"id":"b-5f63be66386a4a103f0665d3","url":"#b-5f63be66386a4a103f0665d3","relation":"figure-and-explanation"}]},{"id":"b-e55cc5881ea9b0b043489c2a","type":"p","scope":"rcnn-transfer-pipeline","url":"https://codingai-lec03.pages.dev/course/blocks/rcnn-transfer-pipeline.html#b-e55cc5881ea9b0b043489c2a","content":"为什么把区域当作一张小图很有力量？因为原来“整图里有什么”的视觉计算，现在可在许多局部窗口上复用。杯子只占整图一小部分时，候选区域让它占据输入的大部分，分类网络就能在更合适的尺度查看它。代价是裁剪、缩放与逐区域计算；区域边缘的上下文也会改变。","sha256":"861774ba5faaffdebcc9f839c46f4495cc31d8be666790b3138545d99ce00d44","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"reading-opening","title":"R-CNN 迁移的是已学会的视觉计算","number":1},"sectionTitle":"R-CNN 迁移的是已学会的视觉计算","currentQuestion":"R-CNN 迁移的是已学会的视觉计算","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-f8b817f1bdc58cc7e1e3f9ab","nextBlock":"b-e888eeb1223080921ba60816","prerequisites":[{"id":"b-65a3a4988a889e8c39295dae","url":"#b-65a3a4988a889e8c39295dae","role":"section-title"},{"id":"b-100d2b561a639d0f25cc3f3f","url":"#b-100d2b561a639d0f25cc3f3f","role":"problem-context"}],"relatedBlocks":[{"id":"b-5f63be66386a4a103f0665d3","url":"#b-5f63be66386a4a103f0665d3","relation":"figure-and-explanation"}]},{"id":"b-e888eeb1223080921ba60816","type":"p","scope":"rcnn-transfer-pipeline","url":"https://codingai-lec03.pages.dev/course/blocks/rcnn-transfer-pipeline.html#b-e888eeb1223080921ba60816","content":"原论文摘要报告 VOC 2012 检测 mAP 为 53.3%，相对之前最好结果提高超过 30%。这个同期结果支持“分类预训练与检测适配能够产生有竞争力的检测器”，不是自监督预训练的实验。作者是 Girshick、Donahue、Darrell、Malik；此处从 MoCo/MAE 接到 R-CNN，是课程按问题重组，不能反写历史先后。","sha256":"178bc2569407866ae7371a893269c1c7e57bc6d16f7f1c044f037e1730c1062d","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"reading-opening","title":"R-CNN 迁移的是已学会的视觉计算","number":1},"sectionTitle":"R-CNN 迁移的是已学会的视觉计算","currentQuestion":"R-CNN 迁移的是已学会的视觉计算","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-e55cc5881ea9b0b043489c2a","nextBlock":"b-5f63be66386a4a103f0665d3","prerequisites":[{"id":"b-65a3a4988a889e8c39295dae","url":"#b-65a3a4988a889e8c39295dae","role":"section-title"},{"id":"b-100d2b561a639d0f25cc3f3f","url":"#b-100d2b561a639d0f25cc3f3f","role":"problem-context"}],"relatedBlocks":[{"id":"b-5f63be66386a4a103f0665d3","url":"#b-5f63be66386a4a103f0665d3","relation":"figure-and-explanation"}]},{"id":"b-5f63be66386a4a103f0665d3","type":"figure","scope":"rcnn-transfer-pipeline","url":"https://codingai-lec03.pages.dev/course/blocks/rcnn-transfer-pipeline.html#b-5f63be66386a4a103f0665d3","content":"图内标签与数值：Lecture 03 课堂图解 P27；候选区域 → 裁出并缩放；红杯候选；预训练 CNN；ImageNet 分类权重；区域类别：杯子；位置：相对框修正；Girshick · Donahue · Darrell · Malik；迁移参数，还要学习检测任务；R-CNN · 2013 预印本 / CVPR 2014。课堂顺序不是历史发生顺序。\n\n裁剪区域进入预训练 CNN，再适配类别与位置输出。 查看对应图解状态\n\n- 候选区域 → 裁出并缩放\n- 红杯候选\n- 预训练 CNN\n- ImageNet 分类权重\n- 区域类别：杯子\n- 位置：相对框修正\n- Girshick · Donahue · Darrell · Malik\n- 迁移参数，还要学习检测任务\n- R-CNN · 2013 预印本 / CVPR 2014。课堂顺序不是历史发生顺序。裁剪区域进入预训练 CNN，再适配类别与位置输出。 [查看对应图解状态](/#/l3-r2-p27/1)[查看完整图解](/#/l3-r2-p27/1)","sha256":"7884d881b406d2bf34ec435ee6bd142637e4a25553b776f44d5e30e839c7762c","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"reading-opening","title":"R-CNN 迁移的是已学会的视觉计算","number":1},"sectionTitle":"R-CNN 迁移的是已学会的视觉计算","currentQuestion":"R-CNN 迁移的是已学会的视觉计算","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"figure":{"caption":"裁剪区域进入预训练 CNN，再适配类别与位置输出。 查看对应图解状态","labels":["候选区域 → 裁出并缩放","红杯候选","预训练 CNN","ImageNet 分类权重","区域类别：杯子","位置：相对框修正","Girshick · Donahue · Darrell · Malik","迁移参数，还要学习检测任务","R-CNN · 2013 预印本 / CVPR 2014。课堂顺序不是历史发生顺序。"],"description":"裁剪区域进入预训练 CNN，再适配类别与位置输出。 查看对应图解状态候选区域 → 裁出并缩放红杯候选预训练 CNNImageNet 分类权重区域类别：杯子位置：相对框修正Girshick · Donahue · Darrell · Malik迁移参数，还要学习检测任务R-CNN · 2013 预印本 / CVPR 2014。课堂顺序不是历史发生顺序。","evidenceIdentity":"裁剪区域进入预训练 CNN，再适配类别与位置输出。 查看对应图解状态","images":[{"src":"/assets/display/8ced939a1fd55047fbf73ea5.webp","alt":""},{"src":"/assets/display/8ced939a1fd55047fbf73ea5.webp","alt":""}]},"previousBlock":"b-e888eeb1223080921ba60816","nextBlock":"b-2c8b940b9531749b141efc2d","prerequisites":[{"id":"b-65a3a4988a889e8c39295dae","url":"#b-65a3a4988a889e8c39295dae","role":"section-title"},{"id":"b-100d2b561a639d0f25cc3f3f","url":"#b-100d2b561a639d0f25cc3f3f","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-2c8b940b9531749b141efc2d","type":"h2","scope":"rcnn-transfer-pipeline","url":"https://codingai-lec03.pages.dev/course/blocks/rcnn-transfer-pipeline.html#b-2c8b940b9531749b141efc2d","content":"## 继续阅读与实验","sha256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"reading-opening","title":"R-CNN 迁移的是已学会的视觉计算","number":1},"sectionTitle":"R-CNN 迁移的是已学会的视觉计算","currentQuestion":"R-CNN 迁移的是已学会的视觉计算","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-5f63be66386a4a103f0665d3","nextBlock":null,"prerequisites":[{"id":"b-65a3a4988a889e8c39295dae","url":"#b-65a3a4988a889e8c39295dae","role":"section-title"},{"id":"b-100d2b561a639d0f25cc3f3f","url":"#b-100d2b561a639d0f25cc3f3f","role":"problem-context"}],"relatedBlocks":[{"id":"b-5f63be66386a4a103f0665d3","url":"#b-5f63be66386a4a103f0665d3","relation":"figure-and-explanation"}]}]}