{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"Mask R-CNN：区域内部，也保留二维结构","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html","markdown":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.md","pages":[{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","unit":"怎样让每一份轮廓属于一个明确对象？","unitId":"instance-mask-identity","chapter":"II","number":1}],"aliases":[{"oldId":"explanation","targetId":"instance-roi-mask-head","operation":"split-explanation-range"},{"oldId":"b-4cfd57baa8b03d4c59eb7031","targetId":"b-4cfd57baa8b03d4c59eb7031","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","afterSHA256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","reason":"Same source body in authored semantic page"},{"oldId":"b-3dcaf90e5f5082a9d8ce9b23","targetId":"b-3dcaf90e5f5082a9d8ce9b23","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"9dc7597f8857b8155402804f5d21a941111d36edfe7c31eb55af59261d185d54","afterSHA256":"9dc7597f8857b8155402804f5d21a941111d36edfe7c31eb55af59261d185d54","reason":"Same source body in authored semantic page"},{"oldId":"b-845f534ac48e246cdc0eb27b","targetId":"b-845f534ac48e246cdc0eb27b","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"1e822f009dd59e8a958ba8e5d9d7d033f7c6f4b6d131abd012857f2401cb5f5c","afterSHA256":"1e822f009dd59e8a958ba8e5d9d7d033f7c6f4b6d131abd012857f2401cb5f5c","reason":"Same source body in authored semantic page"},{"oldId":"b-9d1f282ac12206dcc381fe1a","targetId":"b-9d1f282ac12206dcc381fe1a","page":"instance-roi-mask-head","operation":"rewrite","beforeSHA256":"79d3f3dba5bbad39e77207fa675fceaa4c753ae1e999e0f0792fc1b88cd668b4","afterSHA256":"e9db98425332aa4570fce538383efd5d224b1b90d5aca5ffc2cbed83bb5bae89","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-5491e5f19f9a71a13b968fb1","targetId":"b-5491e5f19f9a71a13b968fb1","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","afterSHA256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","reason":"Same source body in authored semantic page"},{"oldId":"b-ac15c054e0011d48adb093f7","targetId":"b-ac15c054e0011d48adb093f7","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"ed61d718051d81e0362abee888390853ebbb272f2c3b9f475b8b4b09777b8659","afterSHA256":"ed61d718051d81e0362abee888390853ebbb272f2c3b9f475b8b4b09777b8659","reason":"Same source body in authored semantic page"},{"oldId":"b-bb8710bf286dada57448d53e","targetId":"b-bb8710bf286dada57448d53e","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"b2fa9c6eee329eb81acaad0b52114a3d6c31549bc70beb3f8229c7e1b79df5f4","afterSHA256":"b2fa9c6eee329eb81acaad0b52114a3d6c31549bc70beb3f8229c7e1b79df5f4","reason":"Same source body in authored semantic page"},{"oldId":"b-e2cad64d8b9c2afc84baf980","targetId":"b-e2cad64d8b9c2afc84baf980","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"0b511a4330a11c16ba46f599835b5c88e40c9d998df27f4bf90d4c02cb2264ca","afterSHA256":"0b511a4330a11c16ba46f599835b5c88e40c9d998df27f4bf90d4c02cb2264ca","reason":"Same source body in authored semantic page"},{"oldId":"b-eda6f32608e21f64766f92e5","targetId":"b-eda6f32608e21f64766f92e5","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"157aeac42b5c38f069caaedc9a6700d329e9f40f7dc6c00ff43a490b1a1c1683","afterSHA256":"157aeac42b5c38f069caaedc9a6700d329e9f40f7dc6c00ff43a490b1a1c1683","reason":"Same source body in authored semantic page"},{"oldId":"b-27a8a2ad7577d0e0cae04f2a","targetId":"b-27a8a2ad7577d0e0cae04f2a","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"04214673903fd768071a053ae57e5f39eda5e720d70baaf627515e96a7369fb7","afterSHA256":"04214673903fd768071a053ae57e5f39eda5e720d70baaf627515e96a7369fb7","reason":"Same source body in authored semantic page"},{"oldId":"b-f99a33e0ae5fe7926e7a1a9b","targetId":"b-f99a33e0ae5fe7926e7a1a9b","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"ff964f198c4004d329d895e4061f11e6b2f004b82f8f73aa6eb186ea34262bd3","afterSHA256":"ff964f198c4004d329d895e4061f11e6b2f004b82f8f73aa6eb186ea34262bd3","reason":"Same source body in authored semantic page"},{"oldId":"b-3b5243f1a38ccc1850b7e4b5","targetId":"b-3b5243f1a38ccc1850b7e4b5","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"5cea2ec3cb0e60da6a03638a2746272cc387521d0c8b748c2e2c06412a9e1640","afterSHA256":"5cea2ec3cb0e60da6a03638a2746272cc387521d0c8b748c2e2c06412a9e1640","reason":"Same source body in authored semantic page"},{"oldId":"b-a834f80b3e3bff8a5f931aa9","targetId":"b-a834f80b3e3bff8a5f931aa9","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"f16be0424762a44344338696f06ef266c4571d15367ae2cbbd28fac98ac7e9cb","afterSHA256":"f16be0424762a44344338696f06ef266c4571d15367ae2cbbd28fac98ac7e9cb","reason":"Same source body in authored semantic page"},{"oldId":"b-95712553f4a1077f129ee97a","targetId":"b-95712553f4a1077f129ee97a","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"b25685f04cd1b885f3a5999e6a912460c7424021cdc5b2190584242086a82681","afterSHA256":"b25685f04cd1b885f3a5999e6a912460c7424021cdc5b2190584242086a82681","reason":"Same source body in authored semantic page"},{"oldId":"b-673d72ce2887a7323cb723ed","targetId":"b-673d72ce2887a7323cb723ed","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"5b48775ae7fd8fa08f842c069963bd28d82e07c289193b465137eec0416689e8","afterSHA256":"5b48775ae7fd8fa08f842c069963bd28d82e07c289193b465137eec0416689e8","reason":"Same source body in authored semantic page"},{"oldId":"b-95ec6c1178709a89096220f6","targetId":"b-95ec6c1178709a89096220f6","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"44f17f66694339f3941506852e2533d2d9d284791343964e2122dff4993e6375","afterSHA256":"44f17f66694339f3941506852e2533d2d9d284791343964e2122dff4993e6375","reason":"Same source body in authored semantic page"},{"oldId":"b-b0a3da4b7db4bd5d1391b9ec","targetId":"b-b0a3da4b7db4bd5d1391b9ec","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","afterSHA256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","reason":"Same source body in authored semantic page"},{"oldId":"b-cb89a907e6708cf35bec0307","targetId":"b-cb89a907e6708cf35bec0307","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"30703647a4a582d742b4724deb64e595de3208a7897216648bca33414cceabd0","afterSHA256":"30703647a4a582d742b4724deb64e595de3208a7897216648bca33414cceabd0","reason":"Same source body in authored semantic page"},{"oldId":"b-099b7d66a65defd7dfc4a75a","targetId":"b-099b7d66a65defd7dfc4a75a","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"e73d190332b7077e63d983c09848058cabb628eb81bd55af0097760b571651ab","afterSHA256":"e73d190332b7077e63d983c09848058cabb628eb81bd55af0097760b571651ab","reason":"Same source body in authored semantic page"},{"oldId":"b-fb8c863b5ec02d943248a161","targetId":"b-fb8c863b5ec02d943248a161","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"abba6e46583bde30a6ed7cbd1ddf57f4ed3d3f3a382d99060f07097487d9eb36","afterSHA256":"abba6e46583bde30a6ed7cbd1ddf57f4ed3d3f3a382d99060f07097487d9eb36","reason":"Same source body in authored semantic page"},{"oldId":"b-4594e117e668de110d3e37be","targetId":"b-4594e117e668de110d3e37be","page":"instance-roi-mask-head","operation":"move","beforeSHA256":"abba6e46583bde30a6ed7cbd1ddf57f4ed3d3f3a382d99060f07097487d9eb36","afterSHA256":"abba6e46583bde30a6ed7cbd1ddf57f4ed3d3f3a382d99060f07097487d9eb36","reason":"Same source body in authored semantic page"},{"oldId":"instance-roi-mask-head-title","targetId":"b-2bdb10d8b2643ca7bdd08060","operation":"retain-source-identity"}],"blocks":[{"id":"b-2bdb10d8b2643ca7bdd08060","type":"h2","scope":"instance-roi-mask-head","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-2bdb10d8b2643ca7bdd08060","content":"## 用二维区域特征预测当前实例的掩码","sha256":"866e2558d5a3f70e06195aaab40d61f294b8b202e08cd5d57843d286f7e50774","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-4cfd57baa8b03d4c59eb7031","prerequisites":[{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-4cfd57baa8b03d4c59eb7031","type":"h2","scope":"explanation","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-4cfd57baa8b03d4c59eb7031","content":"## 解释、算例与证据","sha256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-2bdb10d8b2643ca7bdd08060","nextBlock":"b-3dcaf90e5f5082a9d8ce9b23","prerequisites":[{"id":"b-2bdb10d8b2643ca7bdd08060","url":"#b-2bdb10d8b2643ca7bdd08060","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-3dcaf90e5f5082a9d8ce9b23","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-3dcaf90e5f5082a9d8ce9b23","content":"# P40 · Mask R-CNN：区域内部，也保留二维结构","sha256":"9dc7597f8857b8155402804f5d21a941111d36edfe7c31eb55af59261d185d54","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-4cfd57baa8b03d4c59eb7031","nextBlock":"b-845f534ac48e246cdc0eb27b","prerequisites":[{"id":"b-4cfd57baa8b03d4c59eb7031","url":"#b-4cfd57baa8b03d4c59eb7031","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-845f534ac48e246cdc0eb27b","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-845f534ac48e246cdc0eb27b","content":"一个 RoI 的类别、框与局部 mask 有不同输出形状和监督。mask 分支需要二维布局，并以当前实例为目标，框中另一个同类对象不该被顺便选中。P41–43 接着检验区域读出与目标坐标是否真的一致。","sha256":"1e822f009dd59e8a958ba8e5d9d7d033f7c6f4b6d131abd012857f2401cb5f5c","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-3dcaf90e5f5082a9d8ce9b23","nextBlock":"b-9d1f282ac12206dcc381fe1a","prerequisites":[{"id":"b-3dcaf90e5f5082a9d8ce9b23","url":"#b-3dcaf90e5f5082a9d8ce9b23","role":"section-title"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-9d1f282ac12206dcc381fe1a","type":"figure","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-9d1f282ac12206dcc381fe1a","content":"图内标签与数值：Lecture 03 课堂图解 P40；原图 + 当前 RoI；共享 F；RoI 读取；二维区域特征 F_RoI；类别：杯子；框：位置修正；原图局部：仅用于对应对象；实例；二值 mask；教学目标示意 · 非模型预测；三项任务读取共享 F 中的区域特征；照片仅作参考，mask 保持二维局部布局。\n\n同源场景标出红杯RoI；整图计算共享特征F，再按RoI读取蓝色二维区域特征F_RoI，连接类别、框修正与实例mask三项分支。下方RGB裁剪独立标为对象参考，掩码标为教学目标示意。\n\n- 原图 + 当前 RoI\n- 共享 F\n- RoI 读取\n- 二维区域特征 F_RoI\n- 类别：杯子\n- 框：位置修正\n- 原图局部：仅用于对应对象\n- 实例\n- 二值 mask\n- 教学目标示意 · 非模型预测\n- 三项任务读取共享 F 中的区域特征；照片仅作参考，mask 保持二维局部布局。同源场景标出红杯RoI；整图计算共享特征F，再按RoI读取蓝色二维区域特征F_RoI，连接类别、框修正与实例mask三项分支。下方RGB裁剪独立标为对象参考，掩码标为教学目标示意。 [查看对应图解状态](/#/l3-r2-p40/1)[查看完整图解](/course/figures/l3-r2-p40-1.html)","sha256":"e9db98425332aa4570fce538383efd5d224b1b90d5aca5ffc2cbed83bb5bae89","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"figure":{"caption":"同源场景标出红杯RoI；整图计算共享特征F，再按RoI读取蓝色二维区域特征F_RoI，连接类别、框修正与实例mask三项分支。下方RGB裁剪独立标为对象参考，掩码标为教学目标示意。 查看对应图解状态","labels":["原图 + 当前 RoI","共享 F","RoI 读取","二维区域特征 F_RoI","类别：杯子","框：位置修正","原图局部：仅用于对应对象","实例","二值 mask","教学目标示意 · 非模型预测","三项任务读取共享 F 中的区域特征；照片仅作参考，mask 保持二维局部布局。"],"description":"同源场景标出红杯RoI；整图计算共享特征F，再按RoI读取蓝色二维区域特征F_RoI，连接类别、框修正与实例mask三项分支。下方RGB裁剪独立标为对象参考，掩码标为教学目标示意。原图 + 当前 RoI共享 FRoI 读取二维区域特征 F_RoI类别：杯子框：位置修正原图局部：仅用于对应对象实例二值 mask教学目标示意 · 非模型预测三项任务读取共享 F 中的区域特征；照片仅作参考，mask 保持二维局部布局。","evidenceIdentity":"同源场景标出红杯RoI；整图计算共享特征F，再按RoI读取蓝色二维区域特征F_RoI，连接类别、框修正与实例mask三项分支。下方RGB裁剪独立标为对象参考，掩码标为教学目标示意。 查看对应图解状态","images":[{"src":"/assets/display/8ced939a1fd55047fbf73ea5.webp","alt":""},{"src":"/assets/display/8ced939a1fd55047fbf73ea5.webp","alt":""}]},"previousBlock":"b-845f534ac48e246cdc0eb27b","nextBlock":"b-5491e5f19f9a71a13b968fb1","prerequisites":[{"id":"b-3dcaf90e5f5082a9d8ce9b23","url":"#b-3dcaf90e5f5082a9d8ce9b23","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-5491e5f19f9a71a13b968fb1","type":"h2","scope":"narrative","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-5491e5f19f9a71a13b968fb1","content":"## 课堂讲述","sha256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-9d1f282ac12206dcc381fe1a","nextBlock":"b-ac15c054e0011d48adb093f7","prerequisites":[{"id":"b-3dcaf90e5f5082a9d8ce9b23","url":"#b-3dcaf90e5f5082a9d8ce9b23","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-ac15c054e0011d48adb093f7","type":"p","scope":"narrative","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-ac15c054e0011d48adb093f7","content":"Mask R-CNN 沿用刚才的区域检测思路：先对整张图计算共享特征，再按这个 RoI 读取一份小的二维特征图。请看中间的 F_RoI，它来自共享特征，不是把下方的原始杯子照片重新送进分类网络。这个区域既要判断类别和修正位置，也要在局部坐标中预测当前实例的掩码。\n这不是多写一个标签。掩码分支需要保留区域内部的二维布局，才能回答哪里是对象、哪里是背景。类别与局部掩码的预测方式也不同，不能把语义分割的所有像素类别竞争原样套过来。\n图中同一杯子对应一份区域表示，三项输出都能追踪到它。现在有个容易被忽略的问题：区域的边界和特征图的采样位置未必落在整数网格上。如果区域提取时把坐标对齐关系弄歪，后面就可能在错误位置上努力预测轮廓。","sha256":"ed61d718051d81e0362abee888390853ebbb272f2c3b9f475b8b4b09777b8659","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-5491e5f19f9a71a13b968fb1","nextBlock":"b-bb8710bf286dada57448d53e","prerequisites":[{"id":"b-5491e5f19f9a71a13b968fb1","url":"#b-5491e5f19f9a71a13b968fb1","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-bb8710bf286dada57448d53e","type":"h3","scope":"mask-branch-target","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-bb8710bf286dada57448d53e","content":"### 一个 RoI 的三项输出，为什么不是三个类别标签","sha256":"b2fa9c6eee329eb81acaad0b52114a3d6c31549bc70beb3f8229c7e1b79df5f4","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"一个 RoI 的三项输出，为什么不是三个类别标签","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":"mask-branch-target","formulas":[],"previousBlock":"b-ac15c054e0011d48adb093f7","nextBlock":"b-e2cad64d8b9c2afc84baf980","prerequisites":[{"id":"b-5491e5f19f9a71a13b968fb1","url":"#b-5491e5f19f9a71a13b968fb1","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"},{"id":"b-e2cad64d8b9c2afc84baf980","url":"#b-e2cad64d8b9c2afc84baf980","relation":"same-reasoning"},{"id":"b-eda6f32608e21f64766f92e5","url":"#b-eda6f32608e21f64766f92e5","relation":"same-reasoning"},{"id":"b-27a8a2ad7577d0e0cae04f2a","url":"#b-27a8a2ad7577d0e0cae04f2a","relation":"same-reasoning"},{"id":"b-f99a33e0ae5fe7926e7a1a9b","url":"#b-f99a33e0ae5fe7926e7a1a9b","relation":"same-reasoning"},{"id":"b-3b5243f1a38ccc1850b7e4b5","url":"#b-3b5243f1a38ccc1850b7e4b5","relation":"same-reasoning"}]},{"id":"b-e2cad64d8b9c2afc84baf980","type":"p","scope":"mask-branch-target","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-e2cad64d8b9c2afc84baf980","content":"同一 RoI 可以给出类别、四个框偏移和一个二维掩码。前两项可以汇总为短向量；mask 却需要保留“区域左上、右下分别是什么”的空间关系。把区域特征完全摊成不带布局的统计，会给逐像素对应增加负担。","sha256":"0b511a4330a11c16ba46f599835b5c88e40c9d998df27f4bf90d4c02cb2264ca","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"一个 RoI 的三项输出，为什么不是三个类别标签","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":"mask-branch-target","formulas":[],"previousBlock":"b-bb8710bf286dada57448d53e","nextBlock":"b-eda6f32608e21f64766f92e5","prerequisites":[{"id":"b-bb8710bf286dada57448d53e","url":"#b-bb8710bf286dada57448d53e","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"},{"id":"b-bb8710bf286dada57448d53e","url":"#b-bb8710bf286dada57448d53e","relation":"same-reasoning"},{"id":"b-eda6f32608e21f64766f92e5","url":"#b-eda6f32608e21f64766f92e5","relation":"same-reasoning"},{"id":"b-27a8a2ad7577d0e0cae04f2a","url":"#b-27a8a2ad7577d0e0cae04f2a","relation":"same-reasoning"},{"id":"b-f99a33e0ae5fe7926e7a1a9b","url":"#b-f99a33e0ae5fe7926e7a1a9b","relation":"same-reasoning"},{"id":"b-3b5243f1a38ccc1850b7e4b5","url":"#b-3b5243f1a38ccc1850b7e4b5","relation":"same-reasoning"}]},{"id":"b-eda6f32608e21f64766f92e5","type":"p","scope":"mask-branch-target","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-eda6f32608e21f64766f92e5","content":"画面中原图局部只帮助我们指认对象。真正进入区域任务头的是从共享特征 F 读取的区域特征；其空间格子保存相对位置，每格还可以有多个通道。后两页检验这一读取怎样保持连续坐标对应。","sha256":"157aeac42b5c38f069caaedc9a6700d329e9f40f7dc6c00ff43a490b1a1c1683","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"一个 RoI 的三项输出，为什么不是三个类别标签","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":"mask-branch-target","formulas":[],"previousBlock":"b-e2cad64d8b9c2afc84baf980","nextBlock":"b-27a8a2ad7577d0e0cae04f2a","prerequisites":[{"id":"b-bb8710bf286dada57448d53e","url":"#b-bb8710bf286dada57448d53e","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"},{"id":"b-bb8710bf286dada57448d53e","url":"#b-bb8710bf286dada57448d53e","relation":"same-reasoning"},{"id":"b-e2cad64d8b9c2afc84baf980","url":"#b-e2cad64d8b9c2afc84baf980","relation":"same-reasoning"},{"id":"b-27a8a2ad7577d0e0cae04f2a","url":"#b-27a8a2ad7577d0e0cae04f2a","relation":"same-reasoning"},{"id":"b-f99a33e0ae5fe7926e7a1a9b","url":"#b-f99a33e0ae5fe7926e7a1a9b","relation":"same-reasoning"},{"id":"b-3b5243f1a38ccc1850b7e4b5","url":"#b-3b5243f1a38ccc1850b7e4b5","relation":"same-reasoning"}]},{"id":"b-27a8a2ad7577d0e0cae04f2a","type":"p","scope":"mask-branch-target","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-27a8a2ad7577d0e0cae04f2a","content":"Mask R-CNN 的掩码分支在 RoI 内预测固定大小的二值图。原论文的默认实现对每个类别各预测一张 sigmoid mask，训练时只计算真值类别对应的掩码损失；区域分类分支负责决定类别。因此掩码像素回答的是“是否属于本实例”，不是让所有对象类别在每个 RoI 像素上再次竞争。","sha256":"04214673903fd768071a053ae57e5f39eda5e720d70baaf627515e96a7369fb7","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"一个 RoI 的三项输出，为什么不是三个类别标签","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":"mask-branch-target","formulas":[],"previousBlock":"b-eda6f32608e21f64766f92e5","nextBlock":"b-f99a33e0ae5fe7926e7a1a9b","prerequisites":[{"id":"b-bb8710bf286dada57448d53e","url":"#b-bb8710bf286dada57448d53e","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"},{"id":"b-bb8710bf286dada57448d53e","url":"#b-bb8710bf286dada57448d53e","relation":"same-reasoning"},{"id":"b-e2cad64d8b9c2afc84baf980","url":"#b-e2cad64d8b9c2afc84baf980","relation":"same-reasoning"},{"id":"b-eda6f32608e21f64766f92e5","url":"#b-eda6f32608e21f64766f92e5","relation":"same-reasoning"},{"id":"b-f99a33e0ae5fe7926e7a1a9b","url":"#b-f99a33e0ae5fe7926e7a1a9b","relation":"same-reasoning"},{"id":"b-3b5243f1a38ccc1850b7e4b5","url":"#b-3b5243f1a38ccc1850b7e4b5","relation":"same-reasoning"}]},{"id":"b-f99a33e0ae5fe7926e7a1a9b","type":"p","scope":"mask-branch-target","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-f99a33e0ae5fe7926e7a1a9b","content":"用杯子的例子理解：RoI 已对应杯一，框里即使还露出杯二的一部分，杯一的目标 mask 仍只标杯一。两个区域可以都被分类为杯子，分别交付各自的二值轮廓。这种分工让实例身份与像素预测接在一起。","sha256":"ff964f198c4004d329d895e4061f11e6b2f004b82f8f73aa6eb186ea34262bd3","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"一个 RoI 的三项输出，为什么不是三个类别标签","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":"mask-branch-target","formulas":[],"previousBlock":"b-27a8a2ad7577d0e0cae04f2a","nextBlock":"b-3b5243f1a38ccc1850b7e4b5","prerequisites":[{"id":"b-bb8710bf286dada57448d53e","url":"#b-bb8710bf286dada57448d53e","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"},{"id":"b-bb8710bf286dada57448d53e","url":"#b-bb8710bf286dada57448d53e","relation":"same-reasoning"},{"id":"b-e2cad64d8b9c2afc84baf980","url":"#b-e2cad64d8b9c2afc84baf980","relation":"same-reasoning"},{"id":"b-eda6f32608e21f64766f92e5","url":"#b-eda6f32608e21f64766f92e5","relation":"same-reasoning"},{"id":"b-27a8a2ad7577d0e0cae04f2a","url":"#b-27a8a2ad7577d0e0cae04f2a","relation":"same-reasoning"},{"id":"b-3b5243f1a38ccc1850b7e4b5","url":"#b-3b5243f1a38ccc1850b7e4b5","relation":"same-reasoning"}]},{"id":"b-3b5243f1a38ccc1850b7e4b5","type":"p","scope":"mask-branch-target","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-3b5243f1a38ccc1850b7e4b5","content":"本课程并没有在 LAB11 训练完整 Mask R-CNN；LAB11 是三类语义分割。Mask R-CNN 的训练证据来自原论文，坐标、局部目标和插值机制则可由 LAB03 的小算例独立检验。","sha256":"5cea2ec3cb0e60da6a03638a2746272cc387521d0c8b748c2e2c06412a9e1640","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"一个 RoI 的三项输出，为什么不是三个类别标签","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":"mask-branch-target","formulas":[],"previousBlock":"b-f99a33e0ae5fe7926e7a1a9b","nextBlock":"b-a834f80b3e3bff8a5f931aa9","prerequisites":[{"id":"b-bb8710bf286dada57448d53e","url":"#b-bb8710bf286dada57448d53e","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"},{"id":"b-bb8710bf286dada57448d53e","url":"#b-bb8710bf286dada57448d53e","relation":"same-reasoning"},{"id":"b-e2cad64d8b9c2afc84baf980","url":"#b-e2cad64d8b9c2afc84baf980","relation":"same-reasoning"},{"id":"b-eda6f32608e21f64766f92e5","url":"#b-eda6f32608e21f64766f92e5","relation":"same-reasoning"},{"id":"b-27a8a2ad7577d0e0cae04f2a","url":"#b-27a8a2ad7577d0e0cae04f2a","relation":"same-reasoning"},{"id":"b-f99a33e0ae5fe7926e7a1a9b","url":"#b-f99a33e0ae5fe7926e7a1a9b","relation":"same-reasoning"}]},{"id":"b-a834f80b3e3bff8a5f931aa9","type":"h3","scope":"dialogue","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-a834f80b3e3bff8a5f931aa9","content":"### 沿着图解追问","sha256":"f16be0424762a44344338696f06ef266c4571d15367ae2cbbd28fac98ac7e9cb","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-3b5243f1a38ccc1850b7e4b5","nextBlock":"b-95712553f4a1077f129ee97a","prerequisites":[{"id":"b-bb8710bf286dada57448d53e","url":"#b-bb8710bf286dada57448d53e","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-95712553f4a1077f129ee97a","type":"p","scope":"dialogue","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-95712553f4a1077f129ee97a","content":"以下是备课用的模拟师生对话，不是课堂实录。","sha256":"b25685f04cd1b885f3a5999e6a912460c7424021cdc5b2190584242086a82681","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-a834f80b3e3bff8a5f931aa9","nextBlock":"b-673d72ce2887a7323cb723ed","prerequisites":[{"id":"b-a834f80b3e3bff8a5f931aa9","url":"#b-a834f80b3e3bff8a5f931aa9","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-673d72ce2887a7323cb723ed","type":"p","scope":"dialogue","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-673d72ce2887a7323cb723ed","content":"**学生：**为什么前面刚说不必反复裁图，这里又是一张杯子的裁剪图？","sha256":"5b48775ae7fd8fa08f842c069963bd28d82e07c289193b465137eec0416689e8","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-95712553f4a1077f129ee97a","nextBlock":"b-95ec6c1178709a89096220f6","prerequisites":[{"id":"b-a834f80b3e3bff8a5f931aa9","url":"#b-a834f80b3e3bff8a5f931aa9","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-95ec6c1178709a89096220f6","type":"p","scope":"dialogue","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-95ec6c1178709a89096220f6","content":"**教师：**照片只让你看清对应的是哪只杯子。参与计算的是旁边的F_RoI特征格，它从已经算出的共享F中读取。区域身份相同，原始像素和区域特征是两种不同对象。","sha256":"44f17f66694339f3941506852e2533d2d9d284791343964e2122dff4993e6375","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-673d72ce2887a7323cb723ed","nextBlock":"b-b0a3da4b7db4bd5d1391b9ec","prerequisites":[{"id":"b-a834f80b3e3bff8a5f931aa9","url":"#b-a834f80b3e3bff8a5f931aa9","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-b0a3da4b7db4bd5d1391b9ec","type":"h2","scope":"practice","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-b0a3da4b7db4bd5d1391b9ec","content":"## 沿这个问题继续","sha256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-95ec6c1178709a89096220f6","nextBlock":"b-cb89a907e6708cf35bec0307","prerequisites":[{"id":"b-a834f80b3e3bff8a5f931aa9","url":"#b-a834f80b3e3bff8a5f931aa9","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-cb89a907e6708cf35bec0307","type":"p","scope":"practice","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-cb89a907e6708cf35bec0307","content":"沿 CLS08 的像素目标与指标阅读，再选 LAB11 查看已有预测和错误图。","sha256":"30703647a4a582d742b4724deb64e595de3208a7897216648bca33414cceabd0","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-b0a3da4b7db4bd5d1391b9ec","nextBlock":"b-00ab07e0a639e1034971aee3","prerequisites":[{"id":"b-b0a3da4b7db4bd5d1391b9ec","url":"#b-b0a3da4b7db4bd5d1391b9ec","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-00ab07e0a639e1034971aee3","type":"h4","scope":"instance-roi-mask-head","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-00ab07e0a639e1034971aee3","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":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-cb89a907e6708cf35bec0307","nextBlock":"b-099b7d66a65defd7dfc4a75a","prerequisites":[{"id":"b-b0a3da4b7db4bd5d1391b9ec","url":"#b-b0a3da4b7db4bd5d1391b9ec","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-099b7d66a65defd7dfc4a75a","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-099b7d66a65defd7dfc4a75a","content":"依据原始 Mask R-CNN：三个任务头读取从共享F提取的区域特征F_RoI，RGB照片只作对象对应参考。mask预测保持二维结构，原论文按类二值mask，训练只使用真值类别对应的mask loss。显示手工教学目标轮廓而非模型预测结果。","sha256":"e73d190332b7077e63d983c09848058cabb628eb81bd55af0097760b571651ab","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-00ab07e0a639e1034971aee3","nextBlock":"b-6d11d121311d62c7aac1176f","prerequisites":[{"id":"b-00ab07e0a639e1034971aee3","url":"#b-00ab07e0a639e1034971aee3","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-6d11d121311d62c7aac1176f","type":"h4","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-6d11d121311d62c7aac1176f","content":"#### 这条线索的展开与后续阅读","sha256":"48b8229f7559b36ed4d5628416e3991e5c73293fbce43374db67386c8736f3bc","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-099b7d66a65defd7dfc4a75a","nextBlock":"b-fb8c863b5ec02d943248a161","prerequisites":[{"id":"b-00ab07e0a639e1034971aee3","url":"#b-00ab07e0a639e1034971aee3","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-fb8c863b5ec02d943248a161","type":"ul","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-fb8c863b5ec02d943248a161","content":"- [先找到这个杯子，再描出它的轮廓](/supplement.html?return=l3-r2-p40/0#/l3-mask-rcnn/0) 本页说明","sha256":"abba6e46583bde30a6ed7cbd1ddf57f4ed3d3f3a382d99060f07097487d9eb36","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-6d11d121311d62c7aac1176f","nextBlock":"b-4594e117e668de110d3e37be","prerequisites":[{"id":"b-6d11d121311d62c7aac1176f","url":"#b-6d11d121311d62c7aac1176f","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]},{"id":"b-4594e117e668de110d3e37be","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p40.html#b-4594e117e668de110d3e37be","content":"- [先找到这个杯子，再描出它的轮廓](/supplement.html?return=l3-r2-p40/0#/l3-mask-rcnn/0) 本页说明","sha256":"abba6e46583bde30a6ed7cbd1ddf57f4ed3d3f3a382d99060f07097487d9eb36","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"instance-mask-identity","title":"怎样让每一份轮廓属于一个明确对象？"},"page":{"id":"instance-roi-mask-head","title":"用二维区域特征预测当前实例的掩码","number":1},"sectionTitle":"用二维区域特征预测当前实例的掩码","currentQuestion":"类别地图还缺哪种身份，区域 mask 如何补上它？","sourceSlides":["l3-r2-p40"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-fb8c863b5ec02d943248a161","nextBlock":null,"prerequisites":[{"id":"b-6d11d121311d62c7aac1176f","url":"#b-6d11d121311d62c7aac1176f","role":"section-title"},{"id":"b-845f534ac48e246cdc0eb27b","url":"#b-845f534ac48e246cdc0eb27b","role":"problem-context"}],"relatedBlocks":[{"id":"b-9d1f282ac12206dcc381fe1a","url":"#b-9d1f282ac12206dcc381fe1a","relation":"figure-and-explanation"}]}]}