{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"可复用表示，还要经过任务训练","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html","markdown":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.md","pages":[{"id":"prompt-mask-task-training","title":"分开预训练起点、任务学习与使用时复用","unit":"提示怎样指定这一次要分的区域？","unitId":"prompted-spatial-output","chapter":"II","number":1},{"id":"prompt-lab16-rule-baseline","title":"选读：用透明规则检查点、框与容差","unit":"提示怎样指定这一次要分的区域？","unitId":"prompted-spatial-output","chapter":"II","number":2}],"aliases":[{"oldId":"explanation","targetId":"prompt-mask-task-training","operation":"split-explanation-range"},{"oldId":"b-5ba52526e177649fd1bda2a8","targetId":"b-5ba52526e177649fd1bda2a8","page":"prompt-mask-task-training","operation":"move","beforeSHA256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","afterSHA256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","reason":"Same source body in authored semantic page"},{"oldId":"b-293ceae45cfd16ba7c66bf90","targetId":"b-293ceae45cfd16ba7c66bf90","page":"prompt-mask-task-training","operation":"move","beforeSHA256":"38d0460b328901c3d6eb9c5d28f0bcb083f94cc3767995ef1b119eb21cbc8b49","afterSHA256":"38d0460b328901c3d6eb9c5d28f0bcb083f94cc3767995ef1b119eb21cbc8b49","reason":"Same source body in authored semantic page"},{"oldId":"b-017767a789871e84026449fa","targetId":"b-017767a789871e84026449fa","page":"prompt-mask-task-training","operation":"move","beforeSHA256":"3f74312c58902b7fc5b6c50187eec80b4436bbe4c0d78080270d152c18752588","afterSHA256":"3f74312c58902b7fc5b6c50187eec80b4436bbe4c0d78080270d152c18752588","reason":"Same source body in authored semantic page"},{"oldId":"b-54aa80700028ac00c1ff62d7","targetId":"b-54aa80700028ac00c1ff62d7","page":"prompt-mask-task-training","operation":"rewrite","beforeSHA256":"46e4a5fe50580d78d633cdbd3856a2cf4627d1ad56e7e7de12ce48f1a6810268","afterSHA256":"76104aa12bad4fa9ed7370dfe70d86ac7c412019596294708ba1d43b79fca250","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-1700bb16e7eabbd8d4dd7337","targetId":"b-1700bb16e7eabbd8d4dd7337","page":"prompt-mask-task-training","operation":"move","beforeSHA256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","afterSHA256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","reason":"Same source body in authored semantic page"},{"oldId":"b-871ce76f355b81b42c609f95","targetId":"b-871ce76f355b81b42c609f95","page":"prompt-mask-task-training","operation":"move","beforeSHA256":"0d463ab06d5ab902bf84dd0d24dece48c59a5a5a40db2894f1c3bad205106cb1","afterSHA256":"0d463ab06d5ab902bf84dd0d24dece48c59a5a5a40db2894f1c3bad205106cb1","reason":"Same source body in authored semantic page"},{"oldId":"b-fec693011d5f7d4e854de0da","targetId":"b-fec693011d5f7d4e854de0da","page":"prompt-mask-task-training","operation":"move","beforeSHA256":"a08e6e6999d2dedb21fd0147c6f8a5b3a99649156559b5ebe27d202606e6c687","afterSHA256":"a08e6e6999d2dedb21fd0147c6f8a5b3a99649156559b5ebe27d202606e6c687","reason":"Same source body in authored semantic page"},{"oldId":"b-25c75e9b7b2d3523f26b3a71","targetId":"b-25c75e9b7b2d3523f26b3a71","page":"prompt-mask-task-training","operation":"move","beforeSHA256":"d81be8895ad7921895d1f461f04b87ae6a1812cacfdfd8bae29397181d035b00","afterSHA256":"d81be8895ad7921895d1f461f04b87ae6a1812cacfdfd8bae29397181d035b00","reason":"Same source body in authored semantic page"},{"oldId":"b-748a33ca6b978d39bd1b3da5","targetId":"b-748a33ca6b978d39bd1b3da5","page":"prompt-mask-task-training","operation":"move","beforeSHA256":"67b3d3c3aad52c0d80e2601a93b15e852b3304b341f8f30e0eaa71ce59012241","afterSHA256":"67b3d3c3aad52c0d80e2601a93b15e852b3304b341f8f30e0eaa71ce59012241","reason":"Same source body in authored semantic page"},{"oldId":"b-bcad603c83c8d0bda2d06103","targetId":"b-bcad603c83c8d0bda2d06103","page":"prompt-mask-task-training","operation":"move","beforeSHA256":"288b13fb11a7b4bf50ae1b07b0d59fe594cb1b283e580e44cb03d1c60f4dcdf6","afterSHA256":"288b13fb11a7b4bf50ae1b07b0d59fe594cb1b283e580e44cb03d1c60f4dcdf6","reason":"Same source body in authored semantic page"},{"oldId":"b-a5de70cc7372ca4aa2cd24c9","targetId":"b-a5de70cc7372ca4aa2cd24c9","page":"prompt-mask-task-training","operation":"move","beforeSHA256":"d09830d2ab723a8ef4a85da99e5bb96bc3f95fcd6e26eeaad7e1b354c766da11","afterSHA256":"d09830d2ab723a8ef4a85da99e5bb96bc3f95fcd6e26eeaad7e1b354c766da11","reason":"Same source body in authored semantic page"},{"oldId":"b-7542201363a9e807d12d43a4","targetId":"b-7542201363a9e807d12d43a4","page":"prompt-mask-task-training","operation":"move","beforeSHA256":"9a9c2f0352f230e255d47fffb4d6880f574fbbcd31f6c0478c5e0a431550b3e1","afterSHA256":"9a9c2f0352f230e255d47fffb4d6880f574fbbcd31f6c0478c5e0a431550b3e1","reason":"Same source body in authored semantic page"},{"oldId":"b-d3f5f6d659f19a4150e4087f","targetId":"b-d3f5f6d659f19a4150e4087f","page":"prompt-lab16-rule-baseline","operation":"move","beforeSHA256":"8030ffbf81056c0f73dab79c4c6a5655a83eca33f4f75a7544143b163724db03","afterSHA256":"8030ffbf81056c0f73dab79c4c6a5655a83eca33f4f75a7544143b163724db03","reason":"Same source body in authored semantic page"},{"oldId":"b-d8b67a04a0521c16aa200954","targetId":"b-d8b67a04a0521c16aa200954","page":"prompt-lab16-rule-baseline","operation":"move","beforeSHA256":"0ea93c8189c66e77304ac1f4f9dde95a1d382a8f7d7323980890a7e8397e51e6","afterSHA256":"0ea93c8189c66e77304ac1f4f9dde95a1d382a8f7d7323980890a7e8397e51e6","reason":"Same source body in authored semantic page"},{"oldId":"b-4e62b8c8ae56234b982d4aef","targetId":"b-4e62b8c8ae56234b982d4aef","page":"prompt-lab16-rule-baseline","operation":"move","beforeSHA256":"4c557624835b36a583194eddac5a7e899b4c02b7570cac64ccd7d7bb29598e5a","afterSHA256":"4c557624835b36a583194eddac5a7e899b4c02b7570cac64ccd7d7bb29598e5a","reason":"Same source body in authored semantic page"},{"oldId":"b-ebc85b584097405547bdac4b","targetId":"b-ebc85b584097405547bdac4b","page":"prompt-lab16-rule-baseline","operation":"move","beforeSHA256":"4da8ba93b7a9d464be4d626338644c5c4d5729eb6ee6e69359bb744d7642035c","afterSHA256":"4da8ba93b7a9d464be4d626338644c5c4d5729eb6ee6e69359bb744d7642035c","reason":"Same source body in authored semantic page"},{"oldId":"b-f643f4322a6f65aaa02e8823","targetId":"b-f643f4322a6f65aaa02e8823","page":"prompt-lab16-rule-baseline","operation":"move","beforeSHA256":"014d3e686af1d67cf2f7363ad92439dd1a233edd6fea947dbe71b9cdac105596","afterSHA256":"014d3e686af1d67cf2f7363ad92439dd1a233edd6fea947dbe71b9cdac105596","reason":"Same source body in authored semantic page"},{"oldId":"b-56280caf8614727dac037970","targetId":"b-56280caf8614727dac037970","page":"prompt-lab16-rule-baseline","operation":"move","beforeSHA256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","afterSHA256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","reason":"Same source body in authored semantic page"},{"oldId":"b-241d42abf60523c1855576fe","targetId":"b-241d42abf60523c1855576fe","page":"prompt-lab16-rule-baseline","operation":"move","beforeSHA256":"b145cd37b1934012763c01e4a841065fbe4aaa9585a9205f1cc1134ec29a3c1b","afterSHA256":"b145cd37b1934012763c01e4a841065fbe4aaa9585a9205f1cc1134ec29a3c1b","reason":"Same source body in authored semantic page"},{"oldId":"prompt-mask-task-training-title","targetId":"b-dcde5db8c19ee0f9a75e0268","operation":"retain-source-identity"},{"oldId":"prompt-lab16-rule-baseline-title","targetId":"b-9fea87143a80163a78b4bf74","operation":"retain-source-identity"},{"oldId":"prompt-lab16-rule-baseline-bridge","targetId":"b-741a0f205da2ada8804d85dc","operation":"retain-source-identity"}],"blocks":[{"id":"b-dcde5db8c19ee0f9a75e0268","type":"h2","scope":"prompt-mask-task-training","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-dcde5db8c19ee0f9a75e0268","content":"## 分开预训练起点、任务学习与使用时复用","sha256":"c602bb3f6d7fe843c2b8e9de24eaf73548f1efb0d4ebba8c202d3dfb8dd45917","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-mask-task-training","title":"分开预训练起点、任务学习与使用时复用","number":1},"sectionTitle":"分开预训练起点、任务学习与使用时复用","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p47"],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-5ba52526e177649fd1bda2a8","prerequisites":[{"id":"b-017767a789871e84026449fa","url":"#b-017767a789871e84026449fa","role":"problem-context"}],"relatedBlocks":[{"id":"b-54aa80700028ac00c1ff62d7","url":"#b-54aa80700028ac00c1ff62d7","relation":"figure-and-explanation"}]},{"id":"b-5ba52526e177649fd1bda2a8","type":"h2","scope":"explanation","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-5ba52526e177649fd1bda2a8","content":"## 解释、算例与证据","sha256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-mask-task-training","title":"分开预训练起点、任务学习与使用时复用","number":1},"sectionTitle":"分开预训练起点、任务学习与使用时复用","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p47"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-dcde5db8c19ee0f9a75e0268","nextBlock":"b-293ceae45cfd16ba7c66bf90","prerequisites":[{"id":"b-dcde5db8c19ee0f9a75e0268","url":"#b-dcde5db8c19ee0f9a75e0268","role":"section-title"},{"id":"b-017767a789871e84026449fa","url":"#b-017767a789871e84026449fa","role":"problem-context"}],"relatedBlocks":[{"id":"b-54aa80700028ac00c1ff62d7","url":"#b-54aa80700028ac00c1ff62d7","relation":"figure-and-explanation"}]},{"id":"b-293ceae45cfd16ba7c66bf90","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-293ceae45cfd16ba7c66bf90","content":"# P47 · 可复用表示，还要经过任务训练","sha256":"38d0460b328901c3d6eb9c5d28f0bcb083f94cc3767995ef1b119eb21cbc8b49","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-mask-task-training","title":"分开预训练起点、任务学习与使用时复用","number":1},"sectionTitle":"分开预训练起点、任务学习与使用时复用","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p47"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-5ba52526e177649fd1bda2a8","nextBlock":"b-017767a789871e84026449fa","prerequisites":[{"id":"b-5ba52526e177649fd1bda2a8","url":"#b-5ba52526e177649fd1bda2a8","role":"section-title"},{"id":"b-017767a789871e84026449fa","url":"#b-017767a789871e84026449fa","role":"problem-context"}],"relatedBlocks":[{"id":"b-54aa80700028ac00c1ff62d7","url":"#b-54aa80700028ac00c1ff62d7","relation":"figure-and-explanation"}]},{"id":"b-017767a789871e84026449fa","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-017767a789871e84026449fa","content":"把三个阶段分开：MAE 初始化视觉编码器，提示式 mask 训练建立任务能力，使用时缓存同图表示。具体训练目标和数据组织使提示能够起作用；这座桥把第一部分的表示来源与第二部分的空间接口连接起来。","sha256":"3f74312c58902b7fc5b6c50187eec80b4436bbe4c0d78080270d152c18752588","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-mask-task-training","title":"分开预训练起点、任务学习与使用时复用","number":1},"sectionTitle":"分开预训练起点、任务学习与使用时复用","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p47"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-293ceae45cfd16ba7c66bf90","nextBlock":"b-54aa80700028ac00c1ff62d7","prerequisites":[{"id":"b-293ceae45cfd16ba7c66bf90","url":"#b-293ceae45cfd16ba7c66bf90","role":"section-title"}],"relatedBlocks":[{"id":"b-54aa80700028ac00c1ff62d7","url":"#b-54aa80700028ac00c1ff62d7","relation":"figure-and-explanation"}]},{"id":"b-54aa80700028ac00c1ff62d7","type":"figure","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-54aa80700028ac00c1ff62d7","content":"图内标签与数值：Lecture 03 课堂图解 P47；训练：初始化 ≠ 完整分割能力；MAE 预训练；提供图像编码器起点；分割任务训练；SA-1B 掩码 + 提示监督；SAM；图像 / 提示 / 掩码模块；使用：图像编码可复用；图像表示；新点提示 → 一个掩码；新框提示 → 另一个掩码\n\n上层 MAE 初始化到 SA-1B 掩码和提示任务训练；下层一次图像表示服务多个提示。\n\n- 训练：初始化 ≠ 完整分割能力\n- MAE 预训练\n- 提供图像编码器起点\n- 分割任务训练\n- SA-1B 掩码 + 提示监督\n- SAM\n- 图像 / 提示 / 掩码模块\n- 使用：图像编码可复用\n- 图像表示\n- 新点提示 → 一个掩码\n- 新框提示 → 另一个掩码上层 MAE 初始化到 SA-1B 掩码和提示任务训练；下层一次图像表示服务多个提示。 [查看对应图解状态](/#/l3-r2-p47/0)[查看完整图解](/course/figures/l3-r2-p47-0.html)","sha256":"76104aa12bad4fa9ed7370dfe70d86ac7c412019596294708ba1d43b79fca250","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-mask-task-training","title":"分开预训练起点、任务学习与使用时复用","number":1},"sectionTitle":"分开预训练起点、任务学习与使用时复用","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p47"],"knowledgeBlock":null,"formulas":[],"figure":{"caption":"上层 MAE 初始化到 SA-1B 掩码和提示任务训练；下层一次图像表示服务多个提示。 查看对应图解状态","labels":["训练：初始化 ≠ 完整分割能力","MAE 预训练","提供图像编码器起点","分割任务训练","SA-1B 掩码 + 提示监督","SAM","图像 / 提示 / 掩码模块","使用：图像编码可复用","图像表示","新点提示 → 一个掩码","新框提示 → 另一个掩码"],"description":"上层 MAE 初始化到 SA-1B 掩码和提示任务训练；下层一次图像表示服务多个提示。训练：初始化 ≠ 完整分割能力MAE 预训练提供图像编码器起点分割任务训练SA-1B 掩码 + 提示监督SAM图像 / 提示 / 掩码模块使用：图像编码可复用图像表示新点提示 → 一个掩码新框提示 → 另一个掩码","evidenceIdentity":"上层 MAE 初始化到 SA-1B 掩码和提示任务训练；下层一次图像表示服务多个提示。 查看对应图解状态","images":[{"src":"/assets/display/8ced939a1fd55047fbf73ea5.webp","alt":""}]},"previousBlock":"b-017767a789871e84026449fa","nextBlock":"b-1700bb16e7eabbd8d4dd7337","prerequisites":[{"id":"b-293ceae45cfd16ba7c66bf90","url":"#b-293ceae45cfd16ba7c66bf90","role":"section-title"},{"id":"b-017767a789871e84026449fa","url":"#b-017767a789871e84026449fa","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-1700bb16e7eabbd8d4dd7337","type":"h2","scope":"narrative","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-1700bb16e7eabbd8d4dd7337","content":"## 课堂讲述","sha256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-mask-task-training","title":"分开预训练起点、任务学习与使用时复用","number":1},"sectionTitle":"分开预训练起点、任务学习与使用时复用","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p47"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-54aa80700028ac00c1ff62d7","nextBlock":"b-871ce76f355b81b42c609f95","prerequisites":[{"id":"b-293ceae45cfd16ba7c66bf90","url":"#b-293ceae45cfd16ba7c66bf90","role":"section-title"},{"id":"b-017767a789871e84026449fa","url":"#b-017767a789871e84026449fa","role":"problem-context"}],"relatedBlocks":[{"id":"b-54aa80700028ac00c1ff62d7","url":"#b-54aa80700028ac00c1ff62d7","relation":"figure-and-explanation"}]},{"id":"b-871ce76f355b81b42c609f95","type":"p","scope":"narrative","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-871ce76f355b81b42c609f95","content":"SAM 给我们一个很清楚的连接：原始方案的图像编码器使用 MAE 预训练初始化，但分割能力还依赖后续的大规模掩码数据与任务训练。\n所以不能说“MAE 会补图，于是自然就会 Segment Anything”。预训练提供起点，分割训练学习提示与区域之间的关系，使用时再复用图像编码来处理不同提示。\n至此，两条主线在同一个系统里接上了：第一部分回答表示怎样学，第二部分回答它怎样被适配成空间能力。下一步，提示能不能不只是点和框，而是“书左边的红杯”这样的自然语言？要做成这件事，我们又需要新的对应关系。","sha256":"0d463ab06d5ab902bf84dd0d24dece48c59a5a5a40db2894f1c3bad205106cb1","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-mask-task-training","title":"分开预训练起点、任务学习与使用时复用","number":1},"sectionTitle":"分开预训练起点、任务学习与使用时复用","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p47"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-1700bb16e7eabbd8d4dd7337","nextBlock":"b-fec693011d5f7d4e854de0da","prerequisites":[{"id":"b-1700bb16e7eabbd8d4dd7337","url":"#b-1700bb16e7eabbd8d4dd7337","role":"section-title"},{"id":"b-017767a789871e84026449fa","url":"#b-017767a789871e84026449fa","role":"problem-context"}],"relatedBlocks":[{"id":"b-54aa80700028ac00c1ff62d7","url":"#b-54aa80700028ac00c1ff62d7","relation":"figure-and-explanation"}]},{"id":"b-fec693011d5f7d4e854de0da","type":"h3","scope":"sam-task-training","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-fec693011d5f7d4e854de0da","content":"### MAE 提供起点，掩码任务训练学会怎样响应提示","sha256":"a08e6e6999d2dedb21fd0147c6f8a5b3a99649156559b5ebe27d202606e6c687","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-mask-task-training","title":"分开预训练起点、任务学习与使用时复用","number":1},"sectionTitle":"MAE 提供起点，掩码任务训练学会怎样响应提示","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p47"],"knowledgeBlock":"sam-task-training","formulas":[],"previousBlock":"b-871ce76f355b81b42c609f95","nextBlock":"b-25c75e9b7b2d3523f26b3a71","prerequisites":[{"id":"b-1700bb16e7eabbd8d4dd7337","url":"#b-1700bb16e7eabbd8d4dd7337","role":"section-title"},{"id":"b-017767a789871e84026449fa","url":"#b-017767a789871e84026449fa","role":"problem-context"}],"relatedBlocks":[{"id":"b-54aa80700028ac00c1ff62d7","url":"#b-54aa80700028ac00c1ff62d7","relation":"figure-and-explanation"},{"id":"b-25c75e9b7b2d3523f26b3a71","url":"#b-25c75e9b7b2d3523f26b3a71","relation":"same-reasoning"},{"id":"b-748a33ca6b978d39bd1b3da5","url":"#b-748a33ca6b978d39bd1b3da5","relation":"same-reasoning"},{"id":"b-bcad603c83c8d0bda2d06103","url":"#b-bcad603c83c8d0bda2d06103","relation":"same-reasoning"},{"id":"b-a5de70cc7372ca4aa2cd24c9","url":"#b-a5de70cc7372ca4aa2cd24c9","relation":"same-reasoning"}]},{"id":"b-25c75e9b7b2d3523f26b3a71","type":"p","scope":"sam-task-training","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-25c75e9b7b2d3523f26b3a71","content":"原始 SAM 的图像编码器使用 MAE 预训练初始化。这连接了第一部分的表示学习，但“遮挡重建”目标并没有直接提供“某个正点应对应哪个对象 mask”的全部监督。后续还要进行提示式分割训练。","sha256":"d81be8895ad7921895d1f461f04b87ae6a1812cacfdfd8bae29397181d035b00","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-mask-task-training","title":"分开预训练起点、任务学习与使用时复用","number":1},"sectionTitle":"MAE 提供起点，掩码任务训练学会怎样响应提示","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p47"],"knowledgeBlock":"sam-task-training","formulas":[],"previousBlock":"b-fec693011d5f7d4e854de0da","nextBlock":"b-748a33ca6b978d39bd1b3da5","prerequisites":[{"id":"b-fec693011d5f7d4e854de0da","url":"#b-fec693011d5f7d4e854de0da","role":"section-title"},{"id":"b-017767a789871e84026449fa","url":"#b-017767a789871e84026449fa","role":"problem-context"}],"relatedBlocks":[{"id":"b-54aa80700028ac00c1ff62d7","url":"#b-54aa80700028ac00c1ff62d7","relation":"figure-and-explanation"},{"id":"b-fec693011d5f7d4e854de0da","url":"#b-fec693011d5f7d4e854de0da","relation":"same-reasoning"},{"id":"b-748a33ca6b978d39bd1b3da5","url":"#b-748a33ca6b978d39bd1b3da5","relation":"same-reasoning"},{"id":"b-bcad603c83c8d0bda2d06103","url":"#b-bcad603c83c8d0bda2d06103","relation":"same-reasoning"},{"id":"b-a5de70cc7372ca4aa2cd24c9","url":"#b-a5de70cc7372ca4aa2cd24c9","relation":"same-reasoning"}]},{"id":"b-748a33ca6b978d39bd1b3da5","type":"p","scope":"sam-task-training","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-748a33ca6b978d39bd1b3da5","content":"训练样本包含图像、目标 mask 与模拟提示。模型预测后，可以从错误区域追加正点或负点，再要求它修正；原论文训练模拟每个 mask 的多轮交互，掩码损失组合 focal 与 dice。这样，推理时的“改点、改框”对应训练中学过的任务，而不是凭空获得一个通用按钮。","sha256":"67b3d3c3aad52c0d80e2601a93b15e852b3304b341f8f30e0eaa71ce59012241","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-mask-task-training","title":"分开预训练起点、任务学习与使用时复用","number":1},"sectionTitle":"MAE 提供起点，掩码任务训练学会怎样响应提示","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p47"],"knowledgeBlock":"sam-task-training","formulas":[],"previousBlock":"b-25c75e9b7b2d3523f26b3a71","nextBlock":"b-bcad603c83c8d0bda2d06103","prerequisites":[{"id":"b-fec693011d5f7d4e854de0da","url":"#b-fec693011d5f7d4e854de0da","role":"section-title"},{"id":"b-017767a789871e84026449fa","url":"#b-017767a789871e84026449fa","role":"problem-context"}],"relatedBlocks":[{"id":"b-54aa80700028ac00c1ff62d7","url":"#b-54aa80700028ac00c1ff62d7","relation":"figure-and-explanation"},{"id":"b-fec693011d5f7d4e854de0da","url":"#b-fec693011d5f7d4e854de0da","relation":"same-reasoning"},{"id":"b-25c75e9b7b2d3523f26b3a71","url":"#b-25c75e9b7b2d3523f26b3a71","relation":"same-reasoning"},{"id":"b-bcad603c83c8d0bda2d06103","url":"#b-bcad603c83c8d0bda2d06103","relation":"same-reasoning"},{"id":"b-a5de70cc7372ca4aa2cd24c9","url":"#b-a5de70cc7372ca4aa2cd24c9","relation":"same-reasoning"}]},{"id":"b-bcad603c83c8d0bda2d06103","type":"p","scope":"sam-task-training","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-bcad603c83c8d0bda2d06103","content":"数据组织也属于能力来源。SA-1B 来自模型辅助标注、半自动补充、全自动生成的逐步演进，最终收集约 1100 万张图像与 11 亿个 mask。这里的数据单位是区域轮廓，并非现成的图文语句对。","sha256":"288b13fb11a7b4bf50ae1b07b0d59fe594cb1b283e580e44cb03d1c60f4dcdf6","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-mask-task-training","title":"分开预训练起点、任务学习与使用时复用","number":1},"sectionTitle":"MAE 提供起点，掩码任务训练学会怎样响应提示","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p47"],"knowledgeBlock":"sam-task-training","formulas":[],"previousBlock":"b-748a33ca6b978d39bd1b3da5","nextBlock":"b-a5de70cc7372ca4aa2cd24c9","prerequisites":[{"id":"b-fec693011d5f7d4e854de0da","url":"#b-fec693011d5f7d4e854de0da","role":"section-title"},{"id":"b-017767a789871e84026449fa","url":"#b-017767a789871e84026449fa","role":"problem-context"}],"relatedBlocks":[{"id":"b-54aa80700028ac00c1ff62d7","url":"#b-54aa80700028ac00c1ff62d7","relation":"figure-and-explanation"},{"id":"b-fec693011d5f7d4e854de0da","url":"#b-fec693011d5f7d4e854de0da","relation":"same-reasoning"},{"id":"b-25c75e9b7b2d3523f26b3a71","url":"#b-25c75e9b7b2d3523f26b3a71","relation":"same-reasoning"},{"id":"b-748a33ca6b978d39bd1b3da5","url":"#b-748a33ca6b978d39bd1b3da5","relation":"same-reasoning"},{"id":"b-a5de70cc7372ca4aa2cd24c9","url":"#b-a5de70cc7372ca4aa2cd24c9","relation":"same-reasoning"}]},{"id":"b-a5de70cc7372ca4aa2cd24c9","type":"p","scope":"sam-task-training","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-a5de70cc7372ca4aa2cd24c9","content":"预训练参数、任务监督和使用接口可以分别复用：先以自监督表示为起点，再学习图像/提示到掩码，使用时缓存同图表示。如果只说“MAE 会补图所以自然会分割”，就跳过了任务训练这一关键桥梁。注意此处是原始 SAM；文字提示在论文中有另行探索，不等于公开几何提示 checkpoint 默认已具备任意语言理解。","sha256":"d09830d2ab723a8ef4a85da99e5bb96bc3f95fcd6e26eeaad7e1b354c766da11","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-mask-task-training","title":"分开预训练起点、任务学习与使用时复用","number":1},"sectionTitle":"MAE 提供起点，掩码任务训练学会怎样响应提示","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p47"],"knowledgeBlock":"sam-task-training","formulas":[],"previousBlock":"b-bcad603c83c8d0bda2d06103","nextBlock":"b-5283caf5ef348a0d93506f13","prerequisites":[{"id":"b-fec693011d5f7d4e854de0da","url":"#b-fec693011d5f7d4e854de0da","role":"section-title"},{"id":"b-017767a789871e84026449fa","url":"#b-017767a789871e84026449fa","role":"problem-context"}],"relatedBlocks":[{"id":"b-54aa80700028ac00c1ff62d7","url":"#b-54aa80700028ac00c1ff62d7","relation":"figure-and-explanation"},{"id":"b-fec693011d5f7d4e854de0da","url":"#b-fec693011d5f7d4e854de0da","relation":"same-reasoning"},{"id":"b-25c75e9b7b2d3523f26b3a71","url":"#b-25c75e9b7b2d3523f26b3a71","relation":"same-reasoning"},{"id":"b-748a33ca6b978d39bd1b3da5","url":"#b-748a33ca6b978d39bd1b3da5","relation":"same-reasoning"},{"id":"b-bcad603c83c8d0bda2d06103","url":"#b-bcad603c83c8d0bda2d06103","relation":"same-reasoning"}]},{"id":"b-5283caf5ef348a0d93506f13","type":"h4","scope":"prompt-mask-task-training","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-5283caf5ef348a0d93506f13","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":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-mask-task-training","title":"分开预训练起点、任务学习与使用时复用","number":1},"sectionTitle":"分开预训练起点、任务学习与使用时复用","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p47"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-a5de70cc7372ca4aa2cd24c9","nextBlock":"b-7542201363a9e807d12d43a4","prerequisites":[{"id":"b-fec693011d5f7d4e854de0da","url":"#b-fec693011d5f7d4e854de0da","role":"section-title"},{"id":"b-017767a789871e84026449fa","url":"#b-017767a789871e84026449fa","role":"problem-context"}],"relatedBlocks":[{"id":"b-54aa80700028ac00c1ff62d7","url":"#b-54aa80700028ac00c1ff62d7","relation":"figure-and-explanation"}]},{"id":"b-7542201363a9e807d12d43a4","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-7542201363a9e807d12d43a4","content":"原论文§3明确 MAE 预训练图像编码器，§8强调大规模分割任务训练对能力的重要性。SA-1B 属于分割掩码数据，不是 CLIP/LAION。使用阶段每图编码一次，多提示复用。","sha256":"9a9c2f0352f230e255d47fffb4d6880f574fbbcd31f6c0478c5e0a431550b3e1","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-mask-task-training","title":"分开预训练起点、任务学习与使用时复用","number":1},"sectionTitle":"分开预训练起点、任务学习与使用时复用","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p47"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-5283caf5ef348a0d93506f13","nextBlock":null,"prerequisites":[{"id":"b-5283caf5ef348a0d93506f13","url":"#b-5283caf5ef348a0d93506f13","role":"section-title"},{"id":"b-017767a789871e84026449fa","url":"#b-017767a789871e84026449fa","role":"problem-context"}],"relatedBlocks":[{"id":"b-54aa80700028ac00c1ff62d7","url":"#b-54aa80700028ac00c1ff62d7","relation":"figure-and-explanation"}]},{"id":"b-9fea87143a80163a78b4bf74","type":"h2","scope":"prompt-lab16-rule-baseline","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-9fea87143a80163a78b4bf74","content":"## 选读：用透明规则检查点、框与容差","sha256":"56c5c03d13a4d645a156e50d4a6c55013f7baa55d8fed21e1d31de7c8aadc5c5","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-lab16-rule-baseline","title":"选读：用透明规则检查点、框与容差","number":2},"sectionTitle":"选读：用透明规则检查点、框与容差","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p46","l3-r2-p47"],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-741a0f205da2ada8804d85dc","prerequisites":[{"id":"b-741a0f205da2ada8804d85dc","url":"#b-741a0f205da2ada8804d85dc","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-741a0f205da2ada8804d85dc","type":"p","scope":"prompt-lab16-rule-baseline","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-741a0f205da2ada8804d85dc","content":"规则对照让每项变化可解释，同时保留与 SAM 权重推理的身份区别。","sha256":"c97e2c454a9a2bdb21e76e4dbe02f44990a99fcadd409a26f09aa5aadece41c2","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-lab16-rule-baseline","title":"选读：用透明规则检查点、框与容差","number":2},"sectionTitle":"选读：用透明规则检查点、框与容差","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p46","l3-r2-p47"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-9fea87143a80163a78b4bf74","nextBlock":"b-d3f5f6d659f19a4150e4087f","prerequisites":[{"id":"b-9fea87143a80163a78b4bf74","url":"#b-9fea87143a80163a78b4bf74","role":"section-title"}],"relatedBlocks":[]},{"id":"b-d3f5f6d659f19a4150e4087f","type":"h3","scope":"sam-rule-baseline-explained","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-d3f5f6d659f19a4150e4087f","content":"### LAB16：颜色连通域让提示的作用变得透明","sha256":"8030ffbf81056c0f73dab79c4c6a5655a83eca33f4f75a7544143b163724db03","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-lab16-rule-baseline","title":"选读：用透明规则检查点、框与容差","number":2},"sectionTitle":"LAB16：颜色连通域让提示的作用变得透明","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p46","l3-r2-p47"],"knowledgeBlock":"sam-rule-baseline-explained","formulas":[],"previousBlock":"b-741a0f205da2ada8804d85dc","nextBlock":"b-d8b67a04a0521c16aa200954","prerequisites":[{"id":"b-9fea87143a80163a78b4bf74","url":"#b-9fea87143a80163a78b4bf74","role":"section-title"},{"id":"b-741a0f205da2ada8804d85dc","url":"#b-741a0f205da2ada8804d85dc","role":"problem-context"}],"relatedBlocks":[{"id":"b-d8b67a04a0521c16aa200954","url":"#b-d8b67a04a0521c16aa200954","relation":"same-reasoning"},{"id":"b-4e62b8c8ae56234b982d4aef","url":"#b-4e62b8c8ae56234b982d4aef","relation":"same-reasoning"},{"id":"b-ebc85b584097405547bdac4b","url":"#b-ebc85b584097405547bdac4b","relation":"same-reasoning"},{"id":"b-f643f4322a6f65aaa02e8823","url":"#b-f643f4322a6f65aaa02e8823","relation":"same-reasoning"}]},{"id":"b-d8b67a04a0521c16aa200954","type":"p","scope":"sam-rule-baseline-explained","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-d8b67a04a0521c16aa200954","content":"当前 LAB16 输入 48×64 RGB 图像，含红色外壳、深红内部零件与蓝物体。代码从正点像素取颜色，选择颜色距离在容差内的像素，再保留与正点四邻连通的部分。负点排除与负点颜色太相似的位置；框只限制允许区域。","sha256":"0ea93c8189c66e77304ac1f4f9dde95a1d382a8f7d7323980890a7e8397e51e6","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-lab16-rule-baseline","title":"选读：用透明规则检查点、框与容差","number":2},"sectionTitle":"LAB16：颜色连通域让提示的作用变得透明","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p46","l3-r2-p47"],"knowledgeBlock":"sam-rule-baseline-explained","formulas":[],"previousBlock":"b-d3f5f6d659f19a4150e4087f","nextBlock":"b-4e62b8c8ae56234b982d4aef","prerequisites":[{"id":"b-d3f5f6d659f19a4150e4087f","url":"#b-d3f5f6d659f19a4150e4087f","role":"section-title"},{"id":"b-741a0f205da2ada8804d85dc","url":"#b-741a0f205da2ada8804d85dc","role":"problem-context"}],"relatedBlocks":[{"id":"b-d3f5f6d659f19a4150e4087f","url":"#b-d3f5f6d659f19a4150e4087f","relation":"same-reasoning"},{"id":"b-4e62b8c8ae56234b982d4aef","url":"#b-4e62b8c8ae56234b982d4aef","relation":"same-reasoning"},{"id":"b-ebc85b584097405547bdac4b","url":"#b-ebc85b584097405547bdac4b","relation":"same-reasoning"},{"id":"b-f643f4322a6f65aaa02e8823","url":"#b-f643f4322a6f65aaa02e8823","relation":"same-reasoning"}]},{"id":"b-4e62b8c8ae56234b982d4aef","type":"p","scope":"sam-rule-baseline-explained","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-4e62b8c8ae56234b982d4aef","content":"已有本地输出的面积为：紧容差 110，宽容差 837，增加外壳负点后 110，加框后 156。紧容差选择内部零件，宽容差把整个外壳纳入；框的面积 156 也可能截断物体。这些变化能直接从规则解释。","sha256":"4c557624835b36a583194eddac5a7e899b4c02b7570cac64ccd7d7bb29598e5a","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-lab16-rule-baseline","title":"选读：用透明规则检查点、框与容差","number":2},"sectionTitle":"LAB16：颜色连通域让提示的作用变得透明","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p46","l3-r2-p47"],"knowledgeBlock":"sam-rule-baseline-explained","formulas":[],"previousBlock":"b-d8b67a04a0521c16aa200954","nextBlock":"b-ebc85b584097405547bdac4b","prerequisites":[{"id":"b-d3f5f6d659f19a4150e4087f","url":"#b-d3f5f6d659f19a4150e4087f","role":"section-title"},{"id":"b-741a0f205da2ada8804d85dc","url":"#b-741a0f205da2ada8804d85dc","role":"problem-context"}],"relatedBlocks":[{"id":"b-d3f5f6d659f19a4150e4087f","url":"#b-d3f5f6d659f19a4150e4087f","relation":"same-reasoning"},{"id":"b-d8b67a04a0521c16aa200954","url":"#b-d8b67a04a0521c16aa200954","relation":"same-reasoning"},{"id":"b-ebc85b584097405547bdac4b","url":"#b-ebc85b584097405547bdac4b","relation":"same-reasoning"},{"id":"b-f643f4322a6f65aaa02e8823","url":"#b-f643f4322a6f65aaa02e8823","relation":"same-reasoning"}]},{"id":"b-ebc85b584097405547bdac4b","type":"p","scope":"sam-rule-baseline-explained","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-ebc85b584097405547bdac4b","content":"因此它适合回答“提示和算法假设共同怎样改变结果”，不能回答“SAM 学会了什么”。它确实读入像素并输出 mask，但没有可学习参数，没有加载 SAM checkpoint，也没有语义泛化证据。保留这项透明练习，比给它错误模型身份更有教学价值。","sha256":"4da8ba93b7a9d464be4d626338644c5c4d5729eb6ee6e69359bb744d7642035c","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-lab16-rule-baseline","title":"选读：用透明规则检查点、框与容差","number":2},"sectionTitle":"LAB16：颜色连通域让提示的作用变得透明","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p46","l3-r2-p47"],"knowledgeBlock":"sam-rule-baseline-explained","formulas":[],"previousBlock":"b-4e62b8c8ae56234b982d4aef","nextBlock":"b-f643f4322a6f65aaa02e8823","prerequisites":[{"id":"b-d3f5f6d659f19a4150e4087f","url":"#b-d3f5f6d659f19a4150e4087f","role":"section-title"},{"id":"b-741a0f205da2ada8804d85dc","url":"#b-741a0f205da2ada8804d85dc","role":"problem-context"}],"relatedBlocks":[{"id":"b-d3f5f6d659f19a4150e4087f","url":"#b-d3f5f6d659f19a4150e4087f","relation":"same-reasoning"},{"id":"b-d8b67a04a0521c16aa200954","url":"#b-d8b67a04a0521c16aa200954","relation":"same-reasoning"},{"id":"b-4e62b8c8ae56234b982d4aef","url":"#b-4e62b8c8ae56234b982d4aef","relation":"same-reasoning"},{"id":"b-f643f4322a6f65aaa02e8823","url":"#b-f643f4322a6f65aaa02e8823","relation":"same-reasoning"}]},{"id":"b-f643f4322a6f65aaa02e8823","type":"p","scope":"sam-rule-baseline-explained","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-f643f4322a6f65aaa02e8823","content":"读完可以把同一个正点移到蓝物体，预测容差和负点的作用；或对照官方 SAM predictor 的输入输出接口，标明哪些步骤仍需真实权重推理。此处不把未来模型运行写成已经完成。","sha256":"014d3e686af1d67cf2f7363ad92439dd1a233edd6fea947dbe71b9cdac105596","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-lab16-rule-baseline","title":"选读：用透明规则检查点、框与容差","number":2},"sectionTitle":"LAB16：颜色连通域让提示的作用变得透明","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p46","l3-r2-p47"],"knowledgeBlock":"sam-rule-baseline-explained","formulas":[],"previousBlock":"b-ebc85b584097405547bdac4b","nextBlock":"b-56280caf8614727dac037970","prerequisites":[{"id":"b-d3f5f6d659f19a4150e4087f","url":"#b-d3f5f6d659f19a4150e4087f","role":"section-title"},{"id":"b-741a0f205da2ada8804d85dc","url":"#b-741a0f205da2ada8804d85dc","role":"problem-context"}],"relatedBlocks":[{"id":"b-d3f5f6d659f19a4150e4087f","url":"#b-d3f5f6d659f19a4150e4087f","relation":"same-reasoning"},{"id":"b-d8b67a04a0521c16aa200954","url":"#b-d8b67a04a0521c16aa200954","relation":"same-reasoning"},{"id":"b-4e62b8c8ae56234b982d4aef","url":"#b-4e62b8c8ae56234b982d4aef","relation":"same-reasoning"},{"id":"b-ebc85b584097405547bdac4b","url":"#b-ebc85b584097405547bdac4b","relation":"same-reasoning"}]},{"id":"b-56280caf8614727dac037970","type":"h2","scope":"practice","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-56280caf8614727dac037970","content":"## 沿这个问题继续","sha256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-lab16-rule-baseline","title":"选读：用透明规则检查点、框与容差","number":2},"sectionTitle":"选读：用透明规则检查点、框与容差","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p46","l3-r2-p47"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-f643f4322a6f65aaa02e8823","nextBlock":"b-241d42abf60523c1855576fe","prerequisites":[{"id":"b-d3f5f6d659f19a4150e4087f","url":"#b-d3f5f6d659f19a4150e4087f","role":"section-title"},{"id":"b-741a0f205da2ada8804d85dc","url":"#b-741a0f205da2ada8804d85dc","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-241d42abf60523c1855576fe","type":"p","scope":"practice","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p47.html#b-241d42abf60523c1855576fe","content":"先区分提示接口与训练来源；LAB16 作为规则对照选读，原始 SAM 机制按原论文核对。","sha256":"b145cd37b1934012763c01e4a841065fbe4aaa9585a9205f1cc1134ec29a3c1b","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"prompted-spatial-output","title":"提示怎样指定这一次要分的区域？"},"page":{"id":"prompt-lab16-rule-baseline","title":"选读：用透明规则检查点、框与容差","number":2},"sectionTitle":"选读：用透明规则检查点、框与容差","currentQuestion":"点、框和已有表示怎样结合，提示能力又从哪里学来？","sourceSlides":["l3-r2-p46","l3-r2-p47"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-56280caf8614727dac037970","nextBlock":null,"prerequisites":[{"id":"b-56280caf8614727dac037970","url":"#b-56280caf8614727dac037970","role":"section-title"},{"id":"b-741a0f205da2ada8804d85dc","url":"#b-741a0f205da2ada8804d85dc","role":"problem-context"}],"relatedBlocks":[]}]}