{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f.html","markdown":"https://codingai-lec03.pages.dev/course/notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f.md","pages":[{"id":"reading-opening","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","unit":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","unitId":"notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f","chapter":"course","number":1}],"aliases":[{"oldId":"b-04bce2d09d2997cef256cff8","targetId":"b-04bce2d09d2997cef256cff8","page":"reading-opening","operation":"move","beforeSHA256":"923e066494e483e0d0fdd514ec12b59fe7659670980fdf4639d15a488eccad0e","afterSHA256":"923e066494e483e0d0fdd514ec12b59fe7659670980fdf4639d15a488eccad0e","reason":"Same source body in authored semantic page"},{"oldId":"b-c16c62d50900e0c208590fb9","targetId":"b-c16c62d50900e0c208590fb9","page":"reading-opening","operation":"move","beforeSHA256":"98977fa9b22b0e395943dc9af599ff62039c8c900870dfdb035e47789e9205dc","afterSHA256":"98977fa9b22b0e395943dc9af599ff62039c8c900870dfdb035e47789e9205dc","reason":"Same source body in authored semantic page"},{"oldId":"b-372065426ea0ee91b068ae1e","targetId":"b-372065426ea0ee91b068ae1e","page":"reading-opening","operation":"move","beforeSHA256":"42cba31deccba2a350b9b23bd0ca1dcaa4c4c5cb9d35c773e1fcb89124788ccd","afterSHA256":"42cba31deccba2a350b9b23bd0ca1dcaa4c4c5cb9d35c773e1fcb89124788ccd","reason":"Same source body in authored semantic page"},{"oldId":"b-815bcddd3d8c4fa1e74ba83b","targetId":"b-815bcddd3d8c4fa1e74ba83b","page":"reading-opening","operation":"move","beforeSHA256":"e0584a8ed4ddfb5b9c9abb132012919eb506480c39698c3ddd67fccd73d0a12e","afterSHA256":"e0584a8ed4ddfb5b9c9abb132012919eb506480c39698c3ddd67fccd73d0a12e","reason":"Same source body in authored semantic page"},{"oldId":"b-46eb4866f4693b366a0fe87b","targetId":"b-46eb4866f4693b366a0fe87b","page":"reading-opening","operation":"move","beforeSHA256":"4082a61472678ab5d2f8f6ad434aac5a7a26a7755da80e702be3f8e24510f2db","afterSHA256":"4082a61472678ab5d2f8f6ad434aac5a7a26a7755da80e702be3f8e24510f2db","reason":"Same source body in authored semantic page"},{"oldId":"b-8190532a5913f4f662737e8f","targetId":"b-8190532a5913f4f662737e8f","page":"reading-opening","operation":"move","beforeSHA256":"16853b7a9385ca2afae17284eead500f1e6d68ebbaaf142c6480655034c23ff8","afterSHA256":"16853b7a9385ca2afae17284eead500f1e6d68ebbaaf142c6480655034c23ff8","reason":"Same source body in authored semantic page"},{"oldId":"b-c2a03bc97dbff04186d54159","targetId":"b-c2a03bc97dbff04186d54159","page":"reading-opening","operation":"move","beforeSHA256":"617ff24cec08fa67d7163ca56eb019a7b608d8159bd669d789b12f50a29cf487","afterSHA256":"617ff24cec08fa67d7163ca56eb019a7b608d8159bd669d789b12f50a29cf487","reason":"Same source body in authored semantic page"}],"blocks":[{"id":"b-04bce2d09d2997cef256cff8","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f.html#b-04bce2d09d2997cef256cff8","content":"[CLS 02 · 从类别到区域与掩码](/course/notebooks/f9ab632ba16b4f2084669791cba6ba85.html)","sha256":"923e066494e483e0d0fdd514ec12b59fe7659670980fdf4639d15a488eccad0e","chapter":{"id":"course","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框"},"unit":{"id":"notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框"},"page":{"id":"reading-opening","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","number":1},"sectionTitle":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","currentQuestion":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-c16c62d50900e0c208590fb9","prerequisites":[],"relatedBlocks":[]},{"id":"b-c16c62d50900e0c208590fb9","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f.html#b-c16c62d50900e0c208590fb9","content":"# R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","sha256":"98977fa9b22b0e395943dc9af599ff62039c8c900870dfdb035e47789e9205dc","chapter":{"id":"course","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框"},"unit":{"id":"notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框"},"page":{"id":"reading-opening","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","number":1},"sectionTitle":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","currentQuestion":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-04bce2d09d2997cef256cff8","nextBlock":"b-372065426ea0ee91b068ae1e","prerequisites":[{"id":"b-04bce2d09d2997cef256cff8","url":"#b-04bce2d09d2997cef256cff8","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-372065426ea0ee91b068ae1e","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f.html#b-372065426ea0ee91b068ae1e","content":"主线阅读与优先实验","sha256":"42cba31deccba2a350b9b23bd0ca1dcaa4c4c5cb9d35c773e1fcb89124788ccd","chapter":{"id":"course","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框"},"unit":{"id":"notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框"},"page":{"id":"reading-opening","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","number":1},"sectionTitle":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","currentQuestion":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c16c62d50900e0c208590fb9","nextBlock":"b-815bcddd3d8c4fa1e74ba83b","prerequisites":[{"id":"b-c16c62d50900e0c208590fb9","url":"#b-c16c62d50900e0c208590fb9","role":"section-title"},{"id":"b-04bce2d09d2997cef256cff8","url":"#b-04bce2d09d2997cef256cff8","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-815bcddd3d8c4fa1e74ba83b","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f.html#b-815bcddd3d8c4fa1e74ba83b","content":"同图四输出、共享计算、FCN、RoIAlign、信息碰撞与SAM接口","sha256":"e0584a8ed4ddfb5b9c9abb132012919eb506480c39698c3ddd67fccd73d0a12e","chapter":{"id":"course","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框"},"unit":{"id":"notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框"},"page":{"id":"reading-opening","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","number":1},"sectionTitle":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","currentQuestion":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-372065426ea0ee91b068ae1e","nextBlock":"b-46eb4866f4693b366a0fe87b","prerequisites":[{"id":"b-c16c62d50900e0c208590fb9","url":"#b-c16c62d50900e0c208590fb9","role":"section-title"},{"id":"b-04bce2d09d2997cef256cff8","url":"#b-04bce2d09d2997cef256cff8","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-46eb4866f4693b366a0fe87b","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f.html#b-46eb4866f4693b366a0fe87b","content":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框。预训练来自有监督图像分类，然后用检测数据适配。[R-CNN](https://arxiv.org/abs/1311.2524)","sha256":"4082a61472678ab5d2f8f6ad434aac5a7a26a7755da80e702be3f8e24510f2db","chapter":{"id":"course","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框"},"unit":{"id":"notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框"},"page":{"id":"reading-opening","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","number":1},"sectionTitle":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","currentQuestion":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-815bcddd3d8c4fa1e74ba83b","nextBlock":"b-8190532a5913f4f662737e8f","prerequisites":[{"id":"b-c16c62d50900e0c208590fb9","url":"#b-c16c62d50900e0c208590fb9","role":"section-title"},{"id":"b-04bce2d09d2997cef256cff8","url":"#b-04bce2d09d2997cef256cff8","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-8190532a5913f4f662737e8f","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f.html#b-8190532a5913f4f662737e8f","content":"用手沿图中的杯子画三个重叠候选框。它们包含了许多相同像素；如果每个候选都单独跑一遍卷积，就重复计算了共有的杯身和桌面。框越多，重复越明显。","sha256":"16853b7a9385ca2afae17284eead500f1e6d68ebbaaf142c6480655034c23ff8","chapter":{"id":"course","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框"},"unit":{"id":"notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框"},"page":{"id":"reading-opening","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","number":1},"sectionTitle":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","currentQuestion":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-46eb4866f4693b366a0fe87b","nextBlock":"b-c2a03bc97dbff04186d54159","prerequisites":[{"id":"b-c16c62d50900e0c208590fb9","url":"#b-c16c62d50900e0c208590fb9","role":"section-title"},{"id":"b-04bce2d09d2997cef256cff8","url":"#b-04bce2d09d2997cef256cff8","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-c2a03bc97dbff04186d54159","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f.html#b-c2a03bc97dbff04186d54159","content":"这一步的关键不是“发现了更多模型”，而是发现计算可以复用：**能否先对整张图计算一次，再从共享特征图上读取每个区域？** 框的位置仍必须传给区域读取环节，整图共享并不等于所有对象混成一个向量。","sha256":"617ff24cec08fa67d7163ca56eb019a7b608d8159bd669d789b12f50a29cf487","chapter":{"id":"course","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框"},"unit":{"id":"notebook-blocks/baa2d848a3684e9fa7adcc551b6b653f","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框"},"page":{"id":"reading-opening","title":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","number":1},"sectionTitle":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","currentQuestion":"R-CNN 把可能含有对象的区域提出来，将每个区域变成网络可以处理的输入，再抽取特征、分类与修正框","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-8190532a5913f4f662737e8f","nextBlock":null,"prerequisites":[{"id":"b-c16c62d50900e0c208590fb9","url":"#b-c16c62d50900e0c208590fb9","role":"section-title"},{"id":"b-04bce2d09d2997cef256cff8","url":"#b-04bce2d09d2997cef256cff8","role":"problem-context"}],"relatedBlocks":[]}]}