{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"一句话里的词，指向图中的哪一部分？","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html","markdown":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.md","pages":[{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","unit":"语言怎样选择当前需要的视觉信息？","unitId":"language-local-and-question","chapter":"III","number":1}],"aliases":[{"oldId":"explanation","targetId":"language-entities-and-relations","operation":"split-explanation-range"},{"oldId":"b-a57a8a9f78e732e961acf3e2","targetId":"b-a57a8a9f78e732e961acf3e2","page":"language-entities-and-relations","operation":"move","beforeSHA256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","afterSHA256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","reason":"Same source body in authored semantic page"},{"oldId":"b-d5f7db3875f1a5b15a031e50","targetId":"b-d5f7db3875f1a5b15a031e50","page":"language-entities-and-relations","operation":"move","beforeSHA256":"d063b7fae22fa09e2f2166dbba61f56478baf96c199d7ddd887c2a967e5e3317","afterSHA256":"d063b7fae22fa09e2f2166dbba61f56478baf96c199d7ddd887c2a967e5e3317","reason":"Same source body in authored semantic page"},{"oldId":"b-068b9ac5a4507a2fee695227","targetId":"b-068b9ac5a4507a2fee695227","page":"language-entities-and-relations","operation":"move","beforeSHA256":"dd1eb9eed20bff23b21561095fe860aa6cd309cba0f9dcc4dd6d1a61491e77d6","afterSHA256":"dd1eb9eed20bff23b21561095fe860aa6cd309cba0f9dcc4dd6d1a61491e77d6","reason":"Same source body in authored semantic page"},{"oldId":"b-0c1110fdfb44d1fa3c057bfa","targetId":"b-0c1110fdfb44d1fa3c057bfa","page":"language-entities-and-relations","operation":"rewrite","beforeSHA256":"40451a5e98b62eac32ed6997492fb6d05ed38da72ab885403cddfe1521720918","afterSHA256":"43692b0dff341e764f120f9d6ebb03b1ed2f7282ba28af30022081beebb44963","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-df8472fe433a10f7b2d1c4b4","targetId":"b-df8472fe433a10f7b2d1c4b4","page":"language-entities-and-relations","operation":"move","beforeSHA256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","afterSHA256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","reason":"Same source body in authored semantic page"},{"oldId":"b-39f70bfa616ad4b775f9c94e","targetId":"b-39f70bfa616ad4b775f9c94e","page":"language-entities-and-relations","operation":"move","beforeSHA256":"6a5f00e4b3a993b20cf62b2234af53dd5042f93fdafc06e9ee9c3ff2a2c69e32","afterSHA256":"6a5f00e4b3a993b20cf62b2234af53dd5042f93fdafc06e9ee9c3ff2a2c69e32","reason":"Same source body in authored semantic page"},{"oldId":"b-164e8d561c1902dd29e6674b","targetId":"b-164e8d561c1902dd29e6674b","page":"language-entities-and-relations","operation":"move","beforeSHA256":"dae7a3701df3517822b9d0e4fc6c8acf478e2b42ae0a43265f5cc5118a44e73f","afterSHA256":"dae7a3701df3517822b9d0e4fc6c8acf478e2b42ae0a43265f5cc5118a44e73f","reason":"Same source body in authored semantic page"},{"oldId":"b-c6a0d428d56cbfe0823a6403","targetId":"b-c6a0d428d56cbfe0823a6403","page":"language-entities-and-relations","operation":"move","beforeSHA256":"fcb2d4df06ee0911a7c0b6fa3a64259c4e275e40ed10912934f233ab1bd44e4d","afterSHA256":"fcb2d4df06ee0911a7c0b6fa3a64259c4e275e40ed10912934f233ab1bd44e4d","reason":"Same source body in authored semantic page"},{"oldId":"b-db00c20779932d50a8432ba5","targetId":"b-db00c20779932d50a8432ba5","page":"language-entities-and-relations","operation":"move","beforeSHA256":"753167e916d99733d10317a382bb23d2e9d7c82ad02c291fbefb3baefc81b303","afterSHA256":"753167e916d99733d10317a382bb23d2e9d7c82ad02c291fbefb3baefc81b303","reason":"Same source body in authored semantic page"},{"oldId":"b-7cb11c08176f746cb5f7dc50","targetId":"b-7cb11c08176f746cb5f7dc50","page":"language-entities-and-relations","operation":"move","beforeSHA256":"a789d8b2a0148c8d95268149cbeb62a417b9f42ebef95300851b7c7da1ea9651","afterSHA256":"a789d8b2a0148c8d95268149cbeb62a417b9f42ebef95300851b7c7da1ea9651","reason":"Same source body in authored semantic page"},{"oldId":"b-6eca937dd719a7be6bd6b0ca","targetId":"b-6eca937dd719a7be6bd6b0ca","page":"language-entities-and-relations","operation":"move","beforeSHA256":"0d25f6fba136681c0086f981761ef0e25643e37172a7e5968ed5d26704874802","afterSHA256":"0d25f6fba136681c0086f981761ef0e25643e37172a7e5968ed5d26704874802","reason":"Same source body in authored semantic page"},{"oldId":"b-ad029a49826688d2d0739434","targetId":"b-ad029a49826688d2d0739434","page":"language-entities-and-relations","operation":"move","beforeSHA256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","afterSHA256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","reason":"Same source body in authored semantic page"},{"oldId":"b-6ebc4ea32a191e3cddbd2de2","targetId":"b-6ebc4ea32a191e3cddbd2de2","page":"language-entities-and-relations","operation":"move","beforeSHA256":"94899f07ac7071d84b101a549079ae213c7354262533ed20e52674bf0044c3c1","afterSHA256":"94899f07ac7071d84b101a549079ae213c7354262533ed20e52674bf0044c3c1","reason":"Same source body in authored semantic page"},{"oldId":"b-1fc5cf61146c6fd3d0280392","targetId":"b-1fc5cf61146c6fd3d0280392","page":"language-entities-and-relations","operation":"move","beforeSHA256":"bf36d4a6a774824ae5c301a31f52d2d2304602edab6dd1b2688dff0fd378291e","afterSHA256":"bf36d4a6a774824ae5c301a31f52d2d2304602edab6dd1b2688dff0fd378291e","reason":"Same source body in authored semantic page"},{"oldId":"b-b7dbf794266d426887258120","targetId":"b-b7dbf794266d426887258120","page":"language-entities-and-relations","operation":"move","beforeSHA256":"105525ffb2e3006f8620f22c58fcd49d74cdca2f83f447203ae96fa9e4110903","afterSHA256":"105525ffb2e3006f8620f22c58fcd49d74cdca2f83f447203ae96fa9e4110903","reason":"Same source body in authored semantic page"},{"oldId":"b-a7235a19a5fea55dc613dc27","targetId":"b-a7235a19a5fea55dc613dc27","page":"language-entities-and-relations","operation":"move","beforeSHA256":"5a30d16283b0499e78618eb1fafa72b05263ad815a28c4682440ada3f2fee587","afterSHA256":"5a30d16283b0499e78618eb1fafa72b05263ad815a28c4682440ada3f2fee587","reason":"Same source body in authored semantic page"},{"oldId":"b-8c60149494a7abf4d7692487","targetId":"b-8c60149494a7abf4d7692487","page":"language-entities-and-relations","operation":"move","beforeSHA256":"47960a8da5145be34c6baec0798b54d953b14d42b6d504efd81723785e452a14","afterSHA256":"47960a8da5145be34c6baec0798b54d953b14d42b6d504efd81723785e452a14","reason":"Same source body in authored semantic page"},{"oldId":"language-entities-and-relations-title","targetId":"b-9229668fe58984681b77a72b","operation":"retain-source-identity"}],"blocks":[{"id":"b-9229668fe58984681b77a72b","type":"h2","scope":"language-entities-and-relations","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-9229668fe58984681b77a72b","content":"## 把名词对应与关系绑定分开","sha256":"e235400f0532528f26c60273e37a89571f47352ec2c69deb6d010cf5d8fbd8cc","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"把名词对应与关系绑定分开","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-a57a8a9f78e732e961acf3e2","prerequisites":[{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"}]},{"id":"b-a57a8a9f78e732e961acf3e2","type":"h2","scope":"explanation","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-a57a8a9f78e732e961acf3e2","content":"## 解释、算例与证据","sha256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"把名词对应与关系绑定分开","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-9229668fe58984681b77a72b","nextBlock":"b-d5f7db3875f1a5b15a031e50","prerequisites":[{"id":"b-9229668fe58984681b77a72b","url":"#b-9229668fe58984681b77a72b","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"}]},{"id":"b-d5f7db3875f1a5b15a031e50","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-d5f7db3875f1a5b15a031e50","content":"# P53 · 一句话里的词，指向图中的哪一部分？","sha256":"d063b7fae22fa09e2f2166dbba61f56478baf96c199d7ddd887c2a967e5e3317","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"把名词对应与关系绑定分开","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-a57a8a9f78e732e961acf3e2","nextBlock":"b-068b9ac5a4507a2fee695227","prerequisites":[{"id":"b-a57a8a9f78e732e961acf3e2","url":"#b-a57a8a9f78e732e961acf3e2","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"}]},{"id":"b-068b9ac5a4507a2fee695227","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-068b9ac5a4507a2fee695227","content":"在同一图上分开读实体和关系。三个名词可以各自对应框，“拿着”却必须绑定正确人物与球。局部区域使语言有可以指向的对象，但关系不由找到两个对象自动获得。图上的人工标注说明任务，已保存的像素规则实验则给出可检查的输出链。","sha256":"dd1eb9eed20bff23b21561095fe860aa6cd309cba0f9dcc4dd6d1a61491e77d6","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"把名词对应与关系绑定分开","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-d5f7db3875f1a5b15a031e50","nextBlock":"b-0c1110fdfb44d1fa3c057bfa","prerequisites":[{"id":"b-d5f7db3875f1a5b15a031e50","url":"#b-d5f7db3875f1a5b15a031e50","role":"section-title"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"}]},{"id":"b-0c1110fdfb44d1fa3c057bfa","type":"figure","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-0c1110fdfb44d1fa3c057bfa","content":"图内标签与数值：红衣人；拿着；球；实体：对应局部内容；关系：连接人物与球；区域表示 + 语言片段 → 学习更细的对应；教学框与关系弧线；区域对齐方法并不因此自动获得全部动作关系。\n\n红衣人、蓝衣人与球各有局部框，句中的实体连接区域，拿着连接人物与球的关系。\n\n- 红衣人\n- 拿着\n- 球\n- 实体：对应局部内容\n- 关系：连接人物与球\n- 区域表示 + 语言片段 → 学习更细的对应\n- 教学框与关系弧线；区域对齐方法并不因此自动获得全部动作关系。红衣人、蓝衣人与球各有局部框，句中的实体连接区域，拿着连接人物与球的关系。 [查看对应图解状态](/#/l3-r2-p53/0)[查看完整图解](/course/figures/l3-r2-p53-0.html)","sha256":"43692b0dff341e764f120f9d6ebb03b1ed2f7282ba28af30022081beebb44963","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"把名词对应与关系绑定分开","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":null,"formulas":[],"figure":{"caption":"红衣人、蓝衣人与球各有局部框，句中的实体连接区域，拿着连接人物与球的关系。 查看对应图解状态","labels":["红衣人","拿着","球","实体：对应局部内容","关系：连接人物与球","区域表示 + 语言片段 → 学习更细的对应","教学框与关系弧线；区域对齐方法并不因此自动获得全部动作关系。"],"description":"红衣人、蓝衣人与球各有局部框，句中的实体连接区域，拿着连接人物与球的关系。红衣人拿着球实体：对应局部内容关系：连接人物与球区域表示 + 语言片段 → 学习更细的对应教学框与关系弧线；区域对齐方法并不因此自动获得全部动作关系。","evidenceIdentity":"红衣人、蓝衣人与球各有局部框，句中的实体连接区域，拿着连接人物与球的关系。 查看对应图解状态","images":[{"src":"/assets/display/512610a019c560ff0384e86a.webp","alt":""}]},"previousBlock":"b-068b9ac5a4507a2fee695227","nextBlock":"b-df8472fe433a10f7b2d1c4b4","prerequisites":[{"id":"b-d5f7db3875f1a5b15a031e50","url":"#b-d5f7db3875f1a5b15a031e50","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-df8472fe433a10f7b2d1c4b4","type":"h2","scope":"narrative","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-df8472fe433a10f7b2d1c4b4","content":"## 课堂讲述","sha256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"把名词对应与关系绑定分开","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-0c1110fdfb44d1fa3c057bfa","nextBlock":"b-39f70bfa616ad4b775f9c94e","prerequisites":[{"id":"b-d5f7db3875f1a5b15a031e50","url":"#b-d5f7db3875f1a5b15a031e50","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"}]},{"id":"b-39f70bfa616ad4b775f9c94e","type":"p","scope":"narrative","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-39f70bfa616ad4b775f9c94e","content":"整图和句子匹配，可以告诉我们它们总体相关，但句子里每个片段对应什么，还需要更细的组织。区域—语言对齐把注意力转向这些局部关系。\n“红衣人”不是一整张图，“球”也不是；“拿着”更不是单独一个物体标签，它描述对象之间的关系。一个系统即使找到了人物和球，也可能把动作关系说错。\n在这个位置，第二部分的区域表示就有用了：已经能够提取局部视觉内容，语言端又提供了可对齐的片段。新的训练任务要把它们接起来，而不是把整图相似度当成所有局部关系都已解决的证据。","sha256":"6a5f00e4b3a993b20cf62b2234af53dd5042f93fdafc06e9ee9c3ff2a2c69e32","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"把名词对应与关系绑定分开","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-df8472fe433a10f7b2d1c4b4","nextBlock":"b-164e8d561c1902dd29e6674b","prerequisites":[{"id":"b-df8472fe433a10f7b2d1c4b4","url":"#b-df8472fe433a10f7b2d1c4b4","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"}]},{"id":"b-164e8d561c1902dd29e6674b","type":"h3","scope":"region-word-alignment","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-164e8d561c1902dd29e6674b","content":"### 从整图相关到局部对应：名词与关系不是同一层信息","sha256":"dae7a3701df3517822b9d0e4fc6c8acf478e2b42ae0a43265f5cc5118a44e73f","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"从整图相关到局部对应：名词与关系不是同一层信息","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":"region-word-alignment","formulas":[],"previousBlock":"b-39f70bfa616ad4b775f9c94e","nextBlock":"b-c6a0d428d56cbfe0823a6403","prerequisites":[{"id":"b-df8472fe433a10f7b2d1c4b4","url":"#b-df8472fe433a10f7b2d1c4b4","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"},{"id":"b-c6a0d428d56cbfe0823a6403","url":"#b-c6a0d428d56cbfe0823a6403","relation":"same-reasoning"},{"id":"b-db00c20779932d50a8432ba5","url":"#b-db00c20779932d50a8432ba5","relation":"same-reasoning"},{"id":"b-7cb11c08176f746cb5f7dc50","url":"#b-7cb11c08176f746cb5f7dc50","relation":"same-reasoning"},{"id":"b-6eca937dd719a7be6bd6b0ca","url":"#b-6eca937dd719a7be6bd6b0ca","relation":"same-reasoning"}]},{"id":"b-c6a0d428d56cbfe0823a6403","type":"p","scope":"region-word-alignment","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-c6a0d428d56cbfe0823a6403","content":"整图相似度可以说“这幅画面与两人持球有关”，却没有说明“红衣人”对应哪块像素、“球”是哪一个区域。把第二部分学过的区域表示取出来，句中的片段就有了可以对齐的视觉对象。","sha256":"fcb2d4df06ee0911a7c0b6fa3a64259c4e275e40ed10912934f233ab1bd44e4d","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"从整图相关到局部对应：名词与关系不是同一层信息","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":"region-word-alignment","formulas":[],"previousBlock":"b-164e8d561c1902dd29e6674b","nextBlock":"b-db00c20779932d50a8432ba5","prerequisites":[{"id":"b-164e8d561c1902dd29e6674b","url":"#b-164e8d561c1902dd29e6674b","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"},{"id":"b-164e8d561c1902dd29e6674b","url":"#b-164e8d561c1902dd29e6674b","relation":"same-reasoning"},{"id":"b-db00c20779932d50a8432ba5","url":"#b-db00c20779932d50a8432ba5","relation":"same-reasoning"},{"id":"b-7cb11c08176f746cb5f7dc50","url":"#b-7cb11c08176f746cb5f7dc50","relation":"same-reasoning"},{"id":"b-6eca937dd719a7be6bd6b0ca","url":"#b-6eca937dd719a7be6bd6b0ca","relation":"same-reasoning"}]},{"id":"b-db00c20779932d50a8432ba5","type":"p","scope":"region-word-alignment","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-db00c20779932d50a8432ba5","content":"P53 中，红衣人、蓝衣人和球分别有候选框；“红衣人”和“球”可以各自指向区域。但“拿着”不是第四种独立物体，它描述两个实体之间的关系。即使候选框都正确，如果把关系连到另一位人物，整句依然错误。实体识别和关系绑定因而要分别检查。","sha256":"753167e916d99733d10317a382bb23d2e9d7c82ad02c291fbefb3baefc81b303","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"从整图相关到局部对应：名词与关系不是同一层信息","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":"region-word-alignment","formulas":[],"previousBlock":"b-c6a0d428d56cbfe0823a6403","nextBlock":"b-7cb11c08176f746cb5f7dc50","prerequisites":[{"id":"b-164e8d561c1902dd29e6674b","url":"#b-164e8d561c1902dd29e6674b","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"},{"id":"b-164e8d561c1902dd29e6674b","url":"#b-164e8d561c1902dd29e6674b","relation":"same-reasoning"},{"id":"b-c6a0d428d56cbfe0823a6403","url":"#b-c6a0d428d56cbfe0823a6403","relation":"same-reasoning"},{"id":"b-7cb11c08176f746cb5f7dc50","url":"#b-7cb11c08176f746cb5f7dc50","relation":"same-reasoning"},{"id":"b-6eca937dd719a7be6bd6b0ca","url":"#b-6eca937dd719a7be6bd6b0ca","relation":"same-reasoning"}]},{"id":"b-7cb11c08176f746cb5f7dc50","type":"p","scope":"region-word-alignment","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-7cb11c08176f746cb5f7dc50","content":"Karpathy 与 Li Fei-Fei 的区域—语言对齐工作从图像与句子资料学习局部跨模态对应，并进一步生成区域描述；论文在图文检索及区域描述上给出实证。其价值是让局部对应成为可学习、可复用的对象，而不是把整图压成一个类别之后才补语言。课堂图上的框和关系线是人工标注，不冒充该论文或本课程模型的预测。","sha256":"a789d8b2a0148c8d95268149cbeb62a417b9f42ebef95300851b7c7da1ea9651","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"从整图相关到局部对应：名词与关系不是同一层信息","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":"region-word-alignment","formulas":[],"previousBlock":"b-db00c20779932d50a8432ba5","nextBlock":"b-6eca937dd719a7be6bd6b0ca","prerequisites":[{"id":"b-164e8d561c1902dd29e6674b","url":"#b-164e8d561c1902dd29e6674b","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"},{"id":"b-164e8d561c1902dd29e6674b","url":"#b-164e8d561c1902dd29e6674b","relation":"same-reasoning"},{"id":"b-c6a0d428d56cbfe0823a6403","url":"#b-c6a0d428d56cbfe0823a6403","relation":"same-reasoning"},{"id":"b-db00c20779932d50a8432ba5","url":"#b-db00c20779932d50a8432ba5","relation":"same-reasoning"},{"id":"b-6eca937dd719a7be6bd6b0ca","url":"#b-6eca937dd719a7be6bd6b0ca","relation":"same-reasoning"}]},{"id":"b-6eca937dd719a7be6bd6b0ca","type":"p","scope":"region-word-alignment","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-6eca937dd719a7be6bd6b0ca","content":"做一个最小自查：把句子改为“蓝衣人在红衣人旁边”，两个实体框可能保持不变，关系应随语句变化；改为“拿着球的人”，输出目标变成受关系约束的人物。区域表示提供候选，语言条件决定怎样选择，二者缺一不可。","sha256":"0d25f6fba136681c0086f981761ef0e25643e37172a7e5968ed5d26704874802","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"从整图相关到局部对应：名词与关系不是同一层信息","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":"region-word-alignment","formulas":[],"previousBlock":"b-7cb11c08176f746cb5f7dc50","nextBlock":"b-ad029a49826688d2d0739434","prerequisites":[{"id":"b-164e8d561c1902dd29e6674b","url":"#b-164e8d561c1902dd29e6674b","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"},{"id":"b-164e8d561c1902dd29e6674b","url":"#b-164e8d561c1902dd29e6674b","relation":"same-reasoning"},{"id":"b-c6a0d428d56cbfe0823a6403","url":"#b-c6a0d428d56cbfe0823a6403","relation":"same-reasoning"},{"id":"b-db00c20779932d50a8432ba5","url":"#b-db00c20779932d50a8432ba5","relation":"same-reasoning"},{"id":"b-7cb11c08176f746cb5f7dc50","url":"#b-7cb11c08176f746cb5f7dc50","relation":"same-reasoning"}]},{"id":"b-ad029a49826688d2d0739434","type":"h2","scope":"practice","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-ad029a49826688d2d0739434","content":"## 沿这个问题继续","sha256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"把名词对应与关系绑定分开","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-6eca937dd719a7be6bd6b0ca","nextBlock":"b-6ebc4ea32a191e3cddbd2de2","prerequisites":[{"id":"b-164e8d561c1902dd29e6674b","url":"#b-164e8d561c1902dd29e6674b","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"}]},{"id":"b-6ebc4ea32a191e3cddbd2de2","type":"p","scope":"practice","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-6ebc4ea32a191e3cddbd2de2","content":"先回看本页画面并指出输入、条件和输出，再按“从整图相关到局部对应：名词与关系不是同一层信息”阅读。相关实验以其已保存结果为起点，需要动手时再顺序执行 Notebook；","sha256":"94899f07ac7071d84b101a549079ae213c7354262533ed20e52674bf0044c3c1","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"把名词对应与关系绑定分开","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-ad029a49826688d2d0739434","nextBlock":"b-0621394c3249ac7975901a64","prerequisites":[{"id":"b-ad029a49826688d2d0739434","url":"#b-ad029a49826688d2d0739434","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"}]},{"id":"b-0621394c3249ac7975901a64","type":"h4","scope":"language-entities-and-relations","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-0621394c3249ac7975901a64","content":"#### 图示条件与公式约定","sha256":"ef7f915c1ae63ed67bd23fecf050249273ba0efa8965d2c8d1c1c6b2d3178122","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"把名词对应与关系绑定分开","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-6ebc4ea32a191e3cddbd2de2","nextBlock":"b-1fc5cf61146c6fd3d0280392","prerequisites":[{"id":"b-ad029a49826688d2d0739434","url":"#b-ad029a49826688d2d0739434","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"}]},{"id":"b-1fc5cf61146c6fd3d0280392","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-1fc5cf61146c6fd3d0280392","content":"当前实际人物场景不含桌子，按本图的红衣人、蓝衣人、球组织局部框。框由课程手工标注，关系弧线说明目标而不声称模型已预测。","sha256":"bf36d4a6a774824ae5c301a31f52d2d2304602edab6dd1b2688dff0fd378291e","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"把名词对应与关系绑定分开","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-0621394c3249ac7975901a64","nextBlock":"b-81e3c001b083de050c697f14","prerequisites":[{"id":"b-0621394c3249ac7975901a64","url":"#b-0621394c3249ac7975901a64","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"}]},{"id":"b-81e3c001b083de050c697f14","type":"h4","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-81e3c001b083de050c697f14","content":"#### 这条线索的展开与后续阅读","sha256":"48b8229f7559b36ed4d5628416e3991e5c73293fbce43374db67386c8736f3bc","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"把名词对应与关系绑定分开","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-1fc5cf61146c6fd3d0280392","nextBlock":"b-b7dbf794266d426887258120","prerequisites":[{"id":"b-0621394c3249ac7975901a64","url":"#b-0621394c3249ac7975901a64","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"}]},{"id":"b-b7dbf794266d426887258120","type":"ul","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-b7dbf794266d426887258120","content":"- [区域 token 让语言预训练接入物体｜问题从哪里出现](/supplement.html?return=l3-r2-p53/0#/l3-region-language-problem/0) L03补充阅读\n- [区域 token 让语言预训练接入物体｜一起算一个完整例子](/supplement.html?return=l3-r2-p53/0#/l3-region-language-example/0) L03补充阅读","sha256":"105525ffb2e3006f8620f22c58fcd49d74cdca2f83f447203ae96fa9e4110903","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"把名词对应与关系绑定分开","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-81e3c001b083de050c697f14","nextBlock":"b-a7235a19a5fea55dc613dc27","prerequisites":[{"id":"b-81e3c001b083de050c697f14","url":"#b-81e3c001b083de050c697f14","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"}]},{"id":"b-a7235a19a5fea55dc613dc27","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-a7235a19a5fea55dc613dc27","content":"- [区域 token 让语言预训练接入物体｜问题从哪里出现](/supplement.html?return=l3-r2-p53/0#/l3-region-language-problem/0) L03补充阅读","sha256":"5a30d16283b0499e78618eb1fafa72b05263ad815a28c4682440ada3f2fee587","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"把名词对应与关系绑定分开","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-b7dbf794266d426887258120","nextBlock":"b-8c60149494a7abf4d7692487","prerequisites":[{"id":"b-81e3c001b083de050c697f14","url":"#b-81e3c001b083de050c697f14","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"}]},{"id":"b-8c60149494a7abf4d7692487","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p53.html#b-8c60149494a7abf4d7692487","content":"- [区域 token 让语言预训练接入物体｜一起算一个完整例子](/supplement.html?return=l3-r2-p53/0#/l3-region-language-example/0) L03补充阅读","sha256":"47960a8da5145be34c6baec0798b54d953b14d42b6d504efd81723785e452a14","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"language-local-and-question","title":"语言怎样选择当前需要的视觉信息？"},"page":{"id":"language-entities-and-relations","title":"把名词对应与关系绑定分开","number":1},"sectionTitle":"把名词对应与关系绑定分开","currentQuestion":"局部对应、问题条件和联合预训练分别增加了哪条关系？","sourceSlides":["l3-r2-p53"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-a7235a19a5fea55dc613dc27","nextBlock":null,"prerequisites":[{"id":"b-81e3c001b083de050c697f14","url":"#b-81e3c001b083de050c697f14","role":"section-title"},{"id":"b-068b9ac5a4507a2fee695227","url":"#b-068b9ac5a4507a2fee695227","role":"problem-context"}],"relatedBlocks":[{"id":"b-0c1110fdfb44d1fa3c057bfa","url":"#b-0c1110fdfb44d1fa3c057bfa","relation":"figure-and-explanation"}]}]}
