{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"多一个通道，和多几个像素不是同一回事","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html","markdown":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.md","pages":[{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","unit":"瓶颈是读错位置，还是差别根本没有保留？","unitId":"spatial-information-preservation","chapter":"II","number":1},{"id":"representation-reading-diagnosis","title":"把对齐收益与信息限制放回各自条件","unit":"瓶颈是读错位置，还是差别根本没有保留？","unitId":"spatial-information-preservation","chapter":"II","number":2}],"aliases":[{"oldId":"explanation","targetId":"representation-difference-channel","operation":"split-explanation-range"},{"oldId":"b-c93cca690377613178228c17","targetId":"b-c93cca690377613178228c17","page":"representation-difference-channel","operation":"move","beforeSHA256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","afterSHA256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","reason":"Same source body in authored semantic page"},{"oldId":"b-c56020ed72409c855ba502c5","targetId":"b-c56020ed72409c855ba502c5","page":"representation-difference-channel","operation":"move","beforeSHA256":"2ae5cf70f38d6df6a56dea197583a4b5f59e78c5a989c75bea318338baae5d8c","afterSHA256":"2ae5cf70f38d6df6a56dea197583a4b5f59e78c5a989c75bea318338baae5d8c","reason":"Same source body in authored semantic page"},{"oldId":"b-99e804bdacdcf177bdcb452b","targetId":"b-99e804bdacdcf177bdcb452b","page":"representation-difference-channel","operation":"move","beforeSHA256":"f1e865fc3cde53d4932e94c57834c0ee9649132af52dcd68311de82aeb089a09","afterSHA256":"f1e865fc3cde53d4932e94c57834c0ee9649132af52dcd68311de82aeb089a09","reason":"Same source body in authored semantic page"},{"oldId":"b-123e83a6b9c4a5e711e4720b","targetId":"b-123e83a6b9c4a5e711e4720b","page":"representation-difference-channel","operation":"rewrite","beforeSHA256":"21657267cb2693bdbfef79181c03c08e07fc56a6c269dec275a80e6a32a6befe","afterSHA256":"25f931f7c4d0dc51cf7212218361aef77c450f0bd3e6a58d0df50441940ca3b6","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-0d748845714882b8e6225060","targetId":"b-0d748845714882b8e6225060","page":"representation-difference-channel","operation":"move","beforeSHA256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","afterSHA256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","reason":"Same source body in authored semantic page"},{"oldId":"b-c507bc278255b36ea69fd959","targetId":"b-c507bc278255b36ea69fd959","page":"representation-difference-channel","operation":"move","beforeSHA256":"8a9e0053359dfd0cf616f0c55ef76dda855745043943041d1e8e1864c9128538","afterSHA256":"8a9e0053359dfd0cf616f0c55ef76dda855745043943041d1e8e1864c9128538","reason":"Same source body in authored semantic page"},{"oldId":"b-610a34b8ae3151c202857ec9","targetId":"b-610a34b8ae3151c202857ec9","page":"representation-difference-channel","operation":"move","beforeSHA256":"cd4d462a0b758d4eddf006b946e7941dcb81fc581db2267dbddf29371e12d4c6","afterSHA256":"cd4d462a0b758d4eddf006b946e7941dcb81fc581db2267dbddf29371e12d4c6","reason":"Same source body in authored semantic page"},{"oldId":"b-83cb0432dfcc71f2806f4226","targetId":"b-83cb0432dfcc71f2806f4226","page":"representation-difference-channel","operation":"move","beforeSHA256":"69e14b9329dfa6394b38b1e589c3af6717778d49081fe48652ccb98426c8e579","afterSHA256":"69e14b9329dfa6394b38b1e589c3af6717778d49081fe48652ccb98426c8e579","reason":"Same source body in authored semantic page"},{"oldId":"b-4fff9e0bb05abb4a0d0b7be1","targetId":"b-4fff9e0bb05abb4a0d0b7be1","page":"representation-difference-channel","operation":"move","beforeSHA256":"208acdd0b5622abec6a44e3e267f352029d54fdbc282bc24dfd0b20706eff900","afterSHA256":"208acdd0b5622abec6a44e3e267f352029d54fdbc282bc24dfd0b20706eff900","reason":"Same source body in authored semantic page"},{"oldId":"b-e5ade749daa8ef8730668d06","targetId":"b-e5ade749daa8ef8730668d06","page":"representation-difference-channel","operation":"move","beforeSHA256":"267c385c2b0ac60a03aa5a7733cc847d7db8e3c88469555a7902411117665aff","afterSHA256":"267c385c2b0ac60a03aa5a7733cc847d7db8e3c88469555a7902411117665aff","reason":"Same source body in authored semantic page"},{"oldId":"b-b5c95ad2be9a4b5765256f62","targetId":"b-b5c95ad2be9a4b5765256f62","page":"representation-difference-channel","operation":"move","beforeSHA256":"07085cb7de68e0b720715bd8a027b943df3cb0cb134e6b70b8a0357fb4714d98","afterSHA256":"07085cb7de68e0b720715bd8a027b943df3cb0cb134e6b70b8a0357fb4714d98","reason":"Same source body in authored semantic page"},{"oldId":"b-626c4aad2002483734e48e06","targetId":"b-626c4aad2002483734e48e06","page":"representation-difference-channel","operation":"move","beforeSHA256":"b91353a3c6d8558dcd2f8526820e8dca95e37aa79aa8c44616357e5416123406","afterSHA256":"b91353a3c6d8558dcd2f8526820e8dca95e37aa79aa8c44616357e5416123406","reason":"Same source body in authored semantic page"},{"oldId":"b-f629d62ed57f7b969d33294c","targetId":"b-f629d62ed57f7b969d33294c","page":"representation-difference-channel","operation":"move","beforeSHA256":"8d96e556e3d94a9886b363ef05b3491a391d6a4f902e1f9bb9875e253a3c2f4d","afterSHA256":"8d96e556e3d94a9886b363ef05b3491a391d6a4f902e1f9bb9875e253a3c2f4d","reason":"Same source body in authored semantic page"},{"oldId":"b-40050de3235e03a790431546","targetId":"b-40050de3235e03a790431546","page":"representation-difference-channel","operation":"move","beforeSHA256":"3d4fd78e7f2a56f190846f789d165edb0885e330720b70f2b8332f1e6a32c61b","afterSHA256":"3d4fd78e7f2a56f190846f789d165edb0885e330720b70f2b8332f1e6a32c61b","reason":"Same source body in authored semantic page"},{"oldId":"b-fba4695d2bf9e31c95dd44a6","targetId":"b-fba4695d2bf9e31c95dd44a6","page":"representation-difference-channel","operation":"move","beforeSHA256":"e1c12b83e218dbc8f0971dfadc1a1ba8fefd45273dbf855f231ac7f44d1b9c24","afterSHA256":"e1c12b83e218dbc8f0971dfadc1a1ba8fefd45273dbf855f231ac7f44d1b9c24","reason":"Same source body in authored semantic page"},{"oldId":"b-27569fb134f383b3bdf95c17","targetId":"b-27569fb134f383b3bdf95c17","page":"representation-reading-diagnosis","operation":"move","beforeSHA256":"5f302b801240bbed114e210e77627b73f44a32151f366ad74ddfdcae7da68f5e","afterSHA256":"5f302b801240bbed114e210e77627b73f44a32151f366ad74ddfdcae7da68f5e","reason":"Same source body in authored semantic page"},{"oldId":"b-fc1b5b517496f8781056aced","targetId":"b-fc1b5b517496f8781056aced","page":"representation-reading-diagnosis","operation":"move","beforeSHA256":"34391c7353cd0dc7b5aabaaf1cf7152abb3c8a4a23569a9796d889c7b4be3b1f","afterSHA256":"34391c7353cd0dc7b5aabaaf1cf7152abb3c8a4a23569a9796d889c7b4be3b1f","reason":"Same source body in authored semantic page"},{"oldId":"b-79bba217f96d883cf9a2db27","targetId":"b-79bba217f96d883cf9a2db27","page":"representation-reading-diagnosis","operation":"move","beforeSHA256":"91da615abd9f11d79355bc8fe5654d1246f1d2ef6db22daa46bc247bda588c02","afterSHA256":"91da615abd9f11d79355bc8fe5654d1246f1d2ef6db22daa46bc247bda588c02","reason":"Same source body in authored semantic page"},{"oldId":"b-7231fd3be61e2d1ffbb52826","targetId":"b-7231fd3be61e2d1ffbb52826","page":"representation-reading-diagnosis","operation":"move","beforeSHA256":"b27d134d90acacfbb8892413111edb846c4f494c80826d510b8f1d49ce8ef265","afterSHA256":"b27d134d90acacfbb8892413111edb846c4f494c80826d510b8f1d49ce8ef265","reason":"Same source body in authored semantic page"},{"oldId":"b-89b0ad8821b91d8cb5b6810d","targetId":"b-89b0ad8821b91d8cb5b6810d","page":"representation-reading-diagnosis","operation":"move","beforeSHA256":"1b4cb1d1397deda2ab0267237a6a26278baa4ad4eb3b946fac42318a852403d5","afterSHA256":"1b4cb1d1397deda2ab0267237a6a26278baa4ad4eb3b946fac42318a852403d5","reason":"Same source body in authored semantic page"},{"oldId":"b-69e663b61c4f79cad9147730","targetId":"b-69e663b61c4f79cad9147730","page":"representation-reading-diagnosis","operation":"move","beforeSHA256":"22e51595d9a7cf03b3165a3d14e280446eba8bcb2d5f98f6235bb0f8017a3033","afterSHA256":"22e51595d9a7cf03b3165a3d14e280446eba8bcb2d5f98f6235bb0f8017a3033","reason":"Same source body in authored semantic page"},{"oldId":"b-e08f0c6de0b6eb7f20390724","targetId":"b-e08f0c6de0b6eb7f20390724","page":"representation-reading-diagnosis","operation":"move","beforeSHA256":"9c6214cc5b5845cbab52bb501fd71ad4ef2eee82703ed0a35e7c6f6d184ed5e1","afterSHA256":"9c6214cc5b5845cbab52bb501fd71ad4ef2eee82703ed0a35e7c6f6d184ed5e1","reason":"Same source body in authored semantic page"},{"oldId":"b-d5843ff28a595529c4a8c3fc","targetId":"b-d5843ff28a595529c4a8c3fc","page":"representation-reading-diagnosis","operation":"move","beforeSHA256":"889643e455f0b36432752ae388e97264df6a2d7c156a17c9b26333c62a8c2257","afterSHA256":"889643e455f0b36432752ae388e97264df6a2d7c156a17c9b26333c62a8c2257","reason":"Same source body in authored semantic page"},{"oldId":"b-b5ac25da73704b2cfbc21670","targetId":"b-b5ac25da73704b2cfbc21670","page":"representation-reading-diagnosis","operation":"move","beforeSHA256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","afterSHA256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","reason":"Same source body in authored semantic page"},{"oldId":"b-454ceb638ec7e2817b3a4023","targetId":"b-454ceb638ec7e2817b3a4023","page":"representation-reading-diagnosis","operation":"move","beforeSHA256":"b8b69250624fd8c15e6f5db8690409848aaec027fc3b89222c6665c3044c9586","afterSHA256":"b8b69250624fd8c15e6f5db8690409848aaec027fc3b89222c6665c3044c9586","reason":"Same source body in authored semantic page"},{"oldId":"representation-difference-channel-title","targetId":"b-260c63ffad88de75145740a8","operation":"retain-source-identity"},{"oldId":"representation-reading-diagnosis-title","targetId":"b-15013b92163eeefe8cb208c5","operation":"retain-source-identity"},{"oldId":"representation-reading-diagnosis-bridge","targetId":"b-38d1937293601b516a78be69","operation":"retain-source-identity"}],"blocks":[{"id":"b-260c63ffad88de75145740a8","type":"h2","scope":"representation-difference-channel","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-260c63ffad88de75145740a8","content":"## 加一项左右差统计，检查什么因此可分","sha256":"6b845753ce6a2a9116f21c3f19b8c50920cb834f0677a1b217a5b237a0bf7a30","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"加一项左右差统计，检查什么因此可分","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-c93cca690377613178228c17","prerequisites":[{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"}]},{"id":"b-c93cca690377613178228c17","type":"h2","scope":"explanation","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-c93cca690377613178228c17","content":"## 解释、算例与证据","sha256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"加一项左右差统计，检查什么因此可分","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-260c63ffad88de75145740a8","nextBlock":"b-c56020ed72409c855ba502c5","prerequisites":[{"id":"b-260c63ffad88de75145740a8","url":"#b-260c63ffad88de75145740a8","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"}]},{"id":"b-c56020ed72409c855ba502c5","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-c56020ed72409c855ba502c5","content":"# P45 · 多一个通道，和多几个像素不是同一回事","sha256":"2ae5cf70f38d6df6a56dea197583a4b5f59e78c5a989c75bea318338baae5d8c","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"加一项左右差统计，检查什么因此可分","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c93cca690377613178228c17","nextBlock":"b-99e804bdacdcf177bdcb452b","prerequisites":[{"id":"b-c93cca690377613178228c17","url":"#b-c93cca690377613178228c17","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"}]},{"id":"b-99e804bdacdcf177bdcb452b","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-99e804bdacdcf177bdcb452b","content":"保持 1×1 网格，增加左右差通道即可让甲乙从同为 .5 变成 [.5,+1] 与 [.5,−1]。这说明信息不只由空间尺寸决定，也说明新通道只保证当前任务所需区别，并没有恢复任意原图。","sha256":"f1e865fc3cde53d4932e94c57834c0ee9649132af52dcd68311de82aeb089a09","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"加一项左右差统计，检查什么因此可分","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c56020ed72409c855ba502c5","nextBlock":"b-123e83a6b9c4a5e711e4720b","prerequisites":[{"id":"b-c56020ed72409c855ba502c5","url":"#b-c56020ed72409c855ba502c5","role":"section-title"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"}]},{"id":"b-123e83a6b9c4a5e711e4720b","type":"figure","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-123e83a6b9c4a5e711e4720b","content":"图内标签与数值：Lecture 03 课堂图解 P45；增加输出格点；0.5；0.5；0.5；0.5；0.5；相同数值被重复；保存一个新通道：左均值 − 右均值；1；0；1；0；0；1；0；1；A = [0.5, +1]；方向：1 − 0 = +1；B = [0.5, −1]；方向：0 − 1 = −1；更密 ≠ 更多区别；分辨率与信息量不是同一件事；构造说明“可以编码这项区别”，不是保证真实网络会自动学到它。\n\n左侧 0.5 重复成四格；右侧新增左列均值减右列均值通道，A=[.5,+1]、B=[.5,−1]。\n\n- 增加输出格点\n- 0.5\n- 0.5\n- 0.5\n- 0.5\n- 0.5\n- 相同数值被重复\n- 保存一个新通道：左均值 − 右均值\n- 1\n- 0\n- 1\n- 0\n- 0\n- 1\n- 0\n- 1\n- A = [0.5, +1]\n- 方向：1 − 0 = +1\n- B = [0.5, −1]\n- 方向：0 − 1 = −1\n- 更密 ≠ 更多区别\n- 分辨率与信息量不是同一件事\n- 构造说明“可以编码这项区别”，不是保证真实网络会自动学到它。左侧 0.5 重复成四格；右侧新增左列均值减右列均值通道，A=[.5,+1]、B=[.5,−1]。 [查看对应图解状态](/#/l3-r2-p45/0)[查看完整图解](/course/figures/l3-r2-p45-0.html)","sha256":"25f931f7c4d0dc51cf7212218361aef77c450f0bd3e6a58d0df50441940ca3b6","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"加一项左右差统计，检查什么因此可分","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"figure":{"caption":"左侧 0.5 重复成四格；右侧新增左列均值减右列均值通道，A=[.5,+1]、B=[.5,−1]。 查看对应图解状态","labels":["增加输出格点","0.5","0.5","0.5","0.5","0.5","相同数值被重复","保存一个新通道：左均值 − 右均值","1","0","1","0","0","1","0","1","A = [0.5, +1]","方向：1 − 0 = +1","B = [0.5, −1]","方向：0 − 1 = −1","更密 ≠ 更多区别","分辨率与信息量不是同一件事","构造说明“可以编码这项区别”，不是保证真实网络会自动学到它。"],"description":"左侧 0.5 重复成四格；右侧新增左列均值减右列均值通道，A=[.5,+1]、B=[.5,−1]。增加输出格点0.50.50.50.50.5相同数值被重复保存一个新通道：左均值 − 右均值10100101A = [0.5, +1]方向：1 − 0 = +1B = [0.5, −1]方向：0 − 1 = −1更密 ≠ 更多区别分辨率与信息量不是同一件事构造说明“可以编码这项区别”，不是保证真实网络会自动学到它。","evidenceIdentity":"左侧 0.5 重复成四格；右侧新增左列均值减右列均值通道，A=[.5,+1]、B=[.5,−1]。 查看对应图解状态","images":[]},"previousBlock":"b-99e804bdacdcf177bdcb452b","nextBlock":"b-0d748845714882b8e6225060","prerequisites":[{"id":"b-c56020ed72409c855ba502c5","url":"#b-c56020ed72409c855ba502c5","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-0d748845714882b8e6225060","type":"h2","scope":"narrative","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-0d748845714882b8e6225060","content":"## 课堂讲述","sha256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"加一项左右差统计，检查什么因此可分","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-123e83a6b9c4a5e711e4720b","nextBlock":"b-c507bc278255b36ea69fd959","prerequisites":[{"id":"b-c56020ed72409c855ba502c5","url":"#b-c56020ed72409c855ba502c5","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"}]},{"id":"b-c507bc278255b36ea69fd959","type":"p","scope":"narrative","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-c507bc278255b36ea69fd959","content":"要恢复刚才的区别，不一定只能提高空间分辨率。也可以让表示多保存一种统计：左列均值减右列均值。于是一个输入得到正一，另一个得到负一，下游就能够区分。\n这个例子说明，网格大小和信息量不能简单画等号。更粗的网格加上合适通道，可能保存某种细节；更密的输出格子，也可能只是重复同一个数。\n回到真实网络，要靠训练、结构和监督决定它保留什么。我们已经有了一个更有用的诊断顺序：先检查输入差异是否进入表示，再检查区域和采样是否对齐，最后看预测头怎样使用它。三件事都明确，分割错误才不再只是一个笼统的“模型不够强”。","sha256":"8a9e0053359dfd0cf616f0c55ef76dda855745043943041d1e8e1864c9128538","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"加一项左右差统计，检查什么因此可分","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-0d748845714882b8e6225060","nextBlock":"b-610a34b8ae3151c202857ec9","prerequisites":[{"id":"b-0d748845714882b8e6225060","url":"#b-0d748845714882b8e6225060","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"}]},{"id":"b-610a34b8ae3151c202857ec9","type":"h3","scope":"channel-retains-left-right","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-610a34b8ae3151c202857ec9","content":"### 一个空间格子也能保存左右差异","sha256":"cd4d462a0b758d4eddf006b946e7941dcb81fc581db2267dbddf29371e12d4c6","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"一个空间格子也能保存左右差异","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":"channel-retains-left-right","formulas":[],"previousBlock":"b-c507bc278255b36ea69fd959","nextBlock":"b-83cb0432dfcc71f2806f4226","prerequisites":[{"id":"b-0d748845714882b8e6225060","url":"#b-0d748845714882b8e6225060","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"},{"id":"b-83cb0432dfcc71f2806f4226","url":"#b-83cb0432dfcc71f2806f4226","relation":"same-reasoning"},{"id":"b-4fff9e0bb05abb4a0d0b7be1","url":"#b-4fff9e0bb05abb4a0d0b7be1","relation":"same-reasoning"},{"id":"b-e5ade749daa8ef8730668d06","url":"#b-e5ade749daa8ef8730668d06","relation":"same-reasoning"},{"id":"b-b5c95ad2be9a4b5765256f62","url":"#b-b5c95ad2be9a4b5765256f62","relation":"same-reasoning"}]},{"id":"b-83cb0432dfcc71f2806f4226","type":"p","scope":"channel-retains-left-right","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-83cb0432dfcc71f2806f4226","content":"继续用甲、乙两张 2×2 图。第一通道取均值 .5，第二通道取左列均值减右列均值。甲得到 $1-0=+1$，乙得到 $0-1=-1$，因此表示分别为 $[.5,+1]$ 与 $[.5,-1]$。空间大小都是 1×1，却已经有了可区分输入。","sha256":"69e14b9329dfa6394b38b1e589c3af6717778d49081fe48652ccb98426c8e579","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"一个空间格子也能保存左右差异","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":"channel-retains-left-right","formulas":[{"latex":"1-0=+1","definitionContext":"继续用甲、乙两张 2×2 图。第一通道取均值 .5，第二通道取左列均值减右列均值。甲得到 $1-0=+1$，乙得到 $0-1=-1$，因此表示分别为 $[.5,+1]$ 与 $[.5,-1]$。空间大小都是 1×1，却已经有了可区分输入。\n一个简单读出只需看第二通道符号，就可判断亮列在左还是右。与 P44 相比，成功不来自输出图画得更密，而来自编码器在压缩时额外保存了一项与任务有关的统计。\n也不要因此推断两个通道足以恢复任意 2×2 图案。四个像素有更多自由度，均值与左右差只保存两种统计，仍可能有别的图案碰撞。表示的价值应围绕任务判断：如果只要左右，这项统计够用；若还要区分上下，就需要保留额外差别。\n真实网络的通道是训练学到的，通常不是这样手写。因此下游 probe、迁移任务与可控干预仍然重要。张量有多少格、多少通道，提供的是容量线索，不是任务信息已被利用的证明。"},{"latex":"0-1=-1","definitionContext":"继续用甲、乙两张 2×2 图。第一通道取均值 .5，第二通道取左列均值减右列均值。甲得到 $1-0=+1$，乙得到 $0-1=-1$，因此表示分别为 $[.5,+1]$ 与 $[.5,-1]$。空间大小都是 1×1，却已经有了可区分输入。\n一个简单读出只需看第二通道符号，就可判断亮列在左还是右。与 P44 相比，成功不来自输出图画得更密，而来自编码器在压缩时额外保存了一项与任务有关的统计。\n也不要因此推断两个通道足以恢复任意 2×2 图案。四个像素有更多自由度，均值与左右差只保存两种统计，仍可能有别的图案碰撞。表示的价值应围绕任务判断：如果只要左右，这项统计够用；若还要区分上下，就需要保留额外差别。\n真实网络的通道是训练学到的，通常不是这样手写。因此下游 probe、迁移任务与可控干预仍然重要。张量有多少格、多少通道，提供的是容量线索，不是任务信息已被利用的证明。"},{"latex":"[.5,+1]","definitionContext":"继续用甲、乙两张 2×2 图。第一通道取均值 .5，第二通道取左列均值减右列均值。甲得到 $1-0=+1$，乙得到 $0-1=-1$，因此表示分别为 $[.5,+1]$ 与 $[.5,-1]$。空间大小都是 1×1，却已经有了可区分输入。\n一个简单读出只需看第二通道符号，就可判断亮列在左还是右。与 P44 相比，成功不来自输出图画得更密，而来自编码器在压缩时额外保存了一项与任务有关的统计。\n也不要因此推断两个通道足以恢复任意 2×2 图案。四个像素有更多自由度，均值与左右差只保存两种统计，仍可能有别的图案碰撞。表示的价值应围绕任务判断：如果只要左右，这项统计够用；若还要区分上下，就需要保留额外差别。\n真实网络的通道是训练学到的，通常不是这样手写。因此下游 probe、迁移任务与可控干预仍然重要。张量有多少格、多少通道，提供的是容量线索，不是任务信息已被利用的证明。"},{"latex":"[.5,-1]","definitionContext":"继续用甲、乙两张 2×2 图。第一通道取均值 .5，第二通道取左列均值减右列均值。甲得到 $1-0=+1$，乙得到 $0-1=-1$，因此表示分别为 $[.5,+1]$ 与 $[.5,-1]$。空间大小都是 1×1，却已经有了可区分输入。\n一个简单读出只需看第二通道符号，就可判断亮列在左还是右。与 P44 相比，成功不来自输出图画得更密，而来自编码器在压缩时额外保存了一项与任务有关的统计。\n也不要因此推断两个通道足以恢复任意 2×2 图案。四个像素有更多自由度，均值与左右差只保存两种统计，仍可能有别的图案碰撞。表示的价值应围绕任务判断：如果只要左右，这项统计够用；若还要区分上下，就需要保留额外差别。\n真实网络的通道是训练学到的，通常不是这样手写。因此下游 probe、迁移任务与可控干预仍然重要。张量有多少格、多少通道，提供的是容量线索，不是任务信息已被利用的证明。"}],"previousBlock":"b-610a34b8ae3151c202857ec9","nextBlock":"b-4fff9e0bb05abb4a0d0b7be1","prerequisites":[{"id":"b-610a34b8ae3151c202857ec9","url":"#b-610a34b8ae3151c202857ec9","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"},{"id":"b-610a34b8ae3151c202857ec9","url":"#b-610a34b8ae3151c202857ec9","relation":"same-reasoning"},{"id":"b-4fff9e0bb05abb4a0d0b7be1","url":"#b-4fff9e0bb05abb4a0d0b7be1","relation":"same-reasoning"},{"id":"b-e5ade749daa8ef8730668d06","url":"#b-e5ade749daa8ef8730668d06","relation":"same-reasoning"},{"id":"b-b5c95ad2be9a4b5765256f62","url":"#b-b5c95ad2be9a4b5765256f62","relation":"same-reasoning"}]},{"id":"b-4fff9e0bb05abb4a0d0b7be1","type":"p","scope":"channel-retains-left-right","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-4fff9e0bb05abb4a0d0b7be1","content":"一个简单读出只需看第二通道符号，就可判断亮列在左还是右。与 P44 相比，成功不来自输出图画得更密，而来自编码器在压缩时额外保存了一项与任务有关的统计。","sha256":"208acdd0b5622abec6a44e3e267f352029d54fdbc282bc24dfd0b20706eff900","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"一个空间格子也能保存左右差异","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":"channel-retains-left-right","formulas":[],"previousBlock":"b-83cb0432dfcc71f2806f4226","nextBlock":"b-e5ade749daa8ef8730668d06","prerequisites":[{"id":"b-610a34b8ae3151c202857ec9","url":"#b-610a34b8ae3151c202857ec9","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"},{"id":"b-610a34b8ae3151c202857ec9","url":"#b-610a34b8ae3151c202857ec9","relation":"same-reasoning"},{"id":"b-83cb0432dfcc71f2806f4226","url":"#b-83cb0432dfcc71f2806f4226","relation":"same-reasoning"},{"id":"b-e5ade749daa8ef8730668d06","url":"#b-e5ade749daa8ef8730668d06","relation":"same-reasoning"},{"id":"b-b5c95ad2be9a4b5765256f62","url":"#b-b5c95ad2be9a4b5765256f62","relation":"same-reasoning"}]},{"id":"b-e5ade749daa8ef8730668d06","type":"p","scope":"channel-retains-left-right","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-e5ade749daa8ef8730668d06","content":"也不要因此推断两个通道足以恢复任意 2×2 图案。四个像素有更多自由度，均值与左右差只保存两种统计，仍可能有别的图案碰撞。表示的价值应围绕任务判断：如果只要左右，这项统计够用；若还要区分上下，就需要保留额外差别。","sha256":"267c385c2b0ac60a03aa5a7733cc847d7db8e3c88469555a7902411117665aff","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"一个空间格子也能保存左右差异","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":"channel-retains-left-right","formulas":[],"previousBlock":"b-4fff9e0bb05abb4a0d0b7be1","nextBlock":"b-b5c95ad2be9a4b5765256f62","prerequisites":[{"id":"b-610a34b8ae3151c202857ec9","url":"#b-610a34b8ae3151c202857ec9","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"},{"id":"b-610a34b8ae3151c202857ec9","url":"#b-610a34b8ae3151c202857ec9","relation":"same-reasoning"},{"id":"b-83cb0432dfcc71f2806f4226","url":"#b-83cb0432dfcc71f2806f4226","relation":"same-reasoning"},{"id":"b-4fff9e0bb05abb4a0d0b7be1","url":"#b-4fff9e0bb05abb4a0d0b7be1","relation":"same-reasoning"},{"id":"b-b5c95ad2be9a4b5765256f62","url":"#b-b5c95ad2be9a4b5765256f62","relation":"same-reasoning"}]},{"id":"b-b5c95ad2be9a4b5765256f62","type":"p","scope":"channel-retains-left-right","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-b5c95ad2be9a4b5765256f62","content":"真实网络的通道是训练学到的，通常不是这样手写。因此下游 probe、迁移任务与可控干预仍然重要。张量有多少格、多少通道，提供的是容量线索，不是任务信息已被利用的证明。","sha256":"07085cb7de68e0b720715bd8a027b943df3cb0cb134e6b70b8a0357fb4714d98","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"一个空间格子也能保存左右差异","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":"channel-retains-left-right","formulas":[],"previousBlock":"b-e5ade749daa8ef8730668d06","nextBlock":"b-b10885dce1113ebc993877f6","prerequisites":[{"id":"b-610a34b8ae3151c202857ec9","url":"#b-610a34b8ae3151c202857ec9","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"},{"id":"b-610a34b8ae3151c202857ec9","url":"#b-610a34b8ae3151c202857ec9","relation":"same-reasoning"},{"id":"b-83cb0432dfcc71f2806f4226","url":"#b-83cb0432dfcc71f2806f4226","relation":"same-reasoning"},{"id":"b-4fff9e0bb05abb4a0d0b7be1","url":"#b-4fff9e0bb05abb4a0d0b7be1","relation":"same-reasoning"},{"id":"b-e5ade749daa8ef8730668d06","url":"#b-e5ade749daa8ef8730668d06","relation":"same-reasoning"}]},{"id":"b-b10885dce1113ebc993877f6","type":"h4","scope":"representation-difference-channel","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-b10885dce1113ebc993877f6","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":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"加一项左右差统计，检查什么因此可分","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-b5c95ad2be9a4b5765256f62","nextBlock":"b-626c4aad2002483734e48e06","prerequisites":[{"id":"b-610a34b8ae3151c202857ec9","url":"#b-610a34b8ae3151c202857ec9","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"}]},{"id":"b-626c4aad2002483734e48e06","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-626c4aad2002483734e48e06","content":"新增通道为左右列均值之差，A:+1，B:−1。它说明存在一种保存方向的编码，不保证真实网络或增加通道数量就会自动学到。","sha256":"b91353a3c6d8558dcd2f8526820e8dca95e37aa79aa8c44616357e5416123406","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"加一项左右差统计，检查什么因此可分","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-b10885dce1113ebc993877f6","nextBlock":"b-86278c2ceb319a3cb44a181e","prerequisites":[{"id":"b-b10885dce1113ebc993877f6","url":"#b-b10885dce1113ebc993877f6","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"}]},{"id":"b-86278c2ceb319a3cb44a181e","type":"h4","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-86278c2ceb319a3cb44a181e","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":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"加一项左右差统计，检查什么因此可分","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-626c4aad2002483734e48e06","nextBlock":"b-f629d62ed57f7b969d33294c","prerequisites":[{"id":"b-b10885dce1113ebc993877f6","url":"#b-b10885dce1113ebc993877f6","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"}]},{"id":"b-f629d62ed57f7b969d33294c","type":"ul","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-f629d62ed57f7b969d33294c","content":"- [三条改动，哪一条重新提供了差别？](/supplement.html?return=l3-r2-p45/0#/l3-three-paths/0) 本页说明\n- [网格不变，多留一个局部通道](/supplement.html?return=l3-r2-p45/0#/l3-channels/0) 本页说明","sha256":"8d96e556e3d94a9886b363ef05b3491a391d6a4f902e1f9bb9875e253a3c2f4d","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"加一项左右差统计，检查什么因此可分","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-86278c2ceb319a3cb44a181e","nextBlock":"b-40050de3235e03a790431546","prerequisites":[{"id":"b-86278c2ceb319a3cb44a181e","url":"#b-86278c2ceb319a3cb44a181e","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"}]},{"id":"b-40050de3235e03a790431546","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-40050de3235e03a790431546","content":"- [三条改动，哪一条重新提供了差别？](/supplement.html?return=l3-r2-p45/0#/l3-three-paths/0) 本页说明","sha256":"3d4fd78e7f2a56f190846f789d165edb0885e330720b70f2b8332f1e6a32c61b","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"加一项左右差统计，检查什么因此可分","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-f629d62ed57f7b969d33294c","nextBlock":"b-fba4695d2bf9e31c95dd44a6","prerequisites":[{"id":"b-86278c2ceb319a3cb44a181e","url":"#b-86278c2ceb319a3cb44a181e","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"}]},{"id":"b-fba4695d2bf9e31c95dd44a6","type":"li","scope":"retained-reading","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-fba4695d2bf9e31c95dd44a6","content":"- [网格不变，多留一个局部通道](/supplement.html?return=l3-r2-p45/0#/l3-channels/0) 本页说明","sha256":"e1c12b83e218dbc8f0971dfadc1a1ba8fefd45273dbf855f231ac7f44d1b9c24","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-difference-channel","title":"加一项左右差统计，检查什么因此可分","number":1},"sectionTitle":"加一项左右差统计，检查什么因此可分","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-40050de3235e03a790431546","nextBlock":null,"prerequisites":[{"id":"b-86278c2ceb319a3cb44a181e","url":"#b-86278c2ceb319a3cb44a181e","role":"section-title"},{"id":"b-99e804bdacdcf177bdcb452b","url":"#b-99e804bdacdcf177bdcb452b","role":"problem-context"}],"relatedBlocks":[{"id":"b-123e83a6b9c4a5e711e4720b","url":"#b-123e83a6b9c4a5e711e4720b","relation":"figure-and-explanation"}]},{"id":"b-15013b92163eeefe8cb208c5","type":"h2","scope":"representation-reading-diagnosis","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-15013b92163eeefe8cb208c5","content":"## 把对齐收益与信息限制放回各自条件","sha256":"eb532b3c5d4f75a628a2097bd4c044eb9b61f0a717ab4c3fae9f4a2659ea9cb8","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-reading-diagnosis","title":"把对齐收益与信息限制放回各自条件","number":2},"sectionTitle":"把对齐收益与信息限制放回各自条件","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p43","l3-r2-p44","l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-38d1937293601b516a78be69","prerequisites":[{"id":"b-38d1937293601b516a78be69","url":"#b-38d1937293601b516a78be69","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-38d1937293601b516a78be69","type":"p","scope":"representation-reading-diagnosis","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-38d1937293601b516a78be69","content":"这段模拟追问帮助核对两类实验的固定条件，不作为已发生课堂对话。","sha256":"b7f189442e3d137d26b346d7ec813cc0669a6c6938636af2ce91b91164568df1","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-reading-diagnosis","title":"把对齐收益与信息限制放回各自条件","number":2},"sectionTitle":"把对齐收益与信息限制放回各自条件","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p43","l3-r2-p44","l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-15013b92163eeefe8cb208c5","nextBlock":"b-27569fb134f383b3bdf95c17","prerequisites":[{"id":"b-15013b92163eeefe8cb208c5","url":"#b-15013b92163eeefe8cb208c5","role":"section-title"}],"relatedBlocks":[]},{"id":"b-27569fb134f383b3bdf95c17","type":"h3","scope":"roi-information-dialogue","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-27569fb134f383b3bdf95c17","content":"### 备课推演：对齐有用，为什么问卷里又不能恢复？","sha256":"5f302b801240bbed114e210e77627b73f44a32151f366ad74ddfdcae7da68f5e","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-reading-diagnosis","title":"把对齐收益与信息限制放回各自条件","number":2},"sectionTitle":"备课推演：对齐有用，为什么问卷里又不能恢复？","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p43","l3-r2-p44","l3-r2-p45"],"knowledgeBlock":"roi-information-dialogue","formulas":[],"previousBlock":"b-38d1937293601b516a78be69","nextBlock":"b-fc1b5b517496f8781056aced","prerequisites":[{"id":"b-15013b92163eeefe8cb208c5","url":"#b-15013b92163eeefe8cb208c5","role":"section-title"},{"id":"b-38d1937293601b516a78be69","url":"#b-38d1937293601b516a78be69","role":"problem-context"}],"relatedBlocks":[{"id":"b-fc1b5b517496f8781056aced","url":"#b-fc1b5b517496f8781056aced","relation":"same-reasoning"},{"id":"b-79bba217f96d883cf9a2db27","url":"#b-79bba217f96d883cf9a2db27","relation":"same-reasoning"},{"id":"b-7231fd3be61e2d1ffbb52826","url":"#b-7231fd3be61e2d1ffbb52826","relation":"same-reasoning"},{"id":"b-89b0ad8821b91d8cb5b6810d","url":"#b-89b0ad8821b91d8cb5b6810d","relation":"same-reasoning"},{"id":"b-69e663b61c4f79cad9147730","url":"#b-69e663b61c4f79cad9147730","relation":"same-reasoning"},{"id":"b-e08f0c6de0b6eb7f20390724","url":"#b-e08f0c6de0b6eb7f20390724","relation":"same-reasoning"},{"id":"b-d5843ff28a595529c4a8c3fc","url":"#b-d5843ff28a595529c4a8c3fc","relation":"same-reasoning"}]},{"id":"b-fc1b5b517496f8781056aced","type":"p","scope":"roi-information-dialogue","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-fc1b5b517496f8781056aced","content":"以下是备课用的模拟追问，不是已发生的课堂对话或学习效果记录。","sha256":"34391c7353cd0dc7b5aabaaf1cf7152abb3c8a4a23569a9796d889c7b4be3b1f","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-reading-diagnosis","title":"把对齐收益与信息限制放回各自条件","number":2},"sectionTitle":"备课推演：对齐有用，为什么问卷里又不能恢复？","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p43","l3-r2-p44","l3-r2-p45"],"knowledgeBlock":"roi-information-dialogue","formulas":[],"previousBlock":"b-27569fb134f383b3bdf95c17","nextBlock":"b-79bba217f96d883cf9a2db27","prerequisites":[{"id":"b-27569fb134f383b3bdf95c17","url":"#b-27569fb134f383b3bdf95c17","role":"section-title"},{"id":"b-38d1937293601b516a78be69","url":"#b-38d1937293601b516a78be69","role":"problem-context"}],"relatedBlocks":[{"id":"b-27569fb134f383b3bdf95c17","url":"#b-27569fb134f383b3bdf95c17","relation":"same-reasoning"},{"id":"b-79bba217f96d883cf9a2db27","url":"#b-79bba217f96d883cf9a2db27","relation":"same-reasoning"},{"id":"b-7231fd3be61e2d1ffbb52826","url":"#b-7231fd3be61e2d1ffbb52826","relation":"same-reasoning"},{"id":"b-89b0ad8821b91d8cb5b6810d","url":"#b-89b0ad8821b91d8cb5b6810d","relation":"same-reasoning"},{"id":"b-69e663b61c4f79cad9147730","url":"#b-69e663b61c4f79cad9147730","relation":"same-reasoning"},{"id":"b-e08f0c6de0b6eb7f20390724","url":"#b-e08f0c6de0b6eb7f20390724","relation":"same-reasoning"},{"id":"b-d5843ff28a595529c4a8c3fc","url":"#b-d5843ff28a595529c4a8c3fc","relation":"same-reasoning"}]},{"id":"b-79bba217f96d883cf9a2db27","type":"p","scope":"roi-information-dialogue","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-79bba217f96d883cf9a2db27","content":"**学生：** 论文说 RoIAlign 更准确，这里却说换采样没用，是不是自相矛盾？","sha256":"91da615abd9f11d79355bc8fe5654d1246f1d2ef6db22daa46bc247bda588c02","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-reading-diagnosis","title":"把对齐收益与信息限制放回各自条件","number":2},"sectionTitle":"备课推演：对齐有用，为什么问卷里又不能恢复？","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p43","l3-r2-p44","l3-r2-p45"],"knowledgeBlock":"roi-information-dialogue","formulas":[],"previousBlock":"b-fc1b5b517496f8781056aced","nextBlock":"b-7231fd3be61e2d1ffbb52826","prerequisites":[{"id":"b-27569fb134f383b3bdf95c17","url":"#b-27569fb134f383b3bdf95c17","role":"section-title"},{"id":"b-38d1937293601b516a78be69","url":"#b-38d1937293601b516a78be69","role":"problem-context"}],"relatedBlocks":[{"id":"b-27569fb134f383b3bdf95c17","url":"#b-27569fb134f383b3bdf95c17","relation":"same-reasoning"},{"id":"b-fc1b5b517496f8781056aced","url":"#b-fc1b5b517496f8781056aced","relation":"same-reasoning"},{"id":"b-7231fd3be61e2d1ffbb52826","url":"#b-7231fd3be61e2d1ffbb52826","relation":"same-reasoning"},{"id":"b-89b0ad8821b91d8cb5b6810d","url":"#b-89b0ad8821b91d8cb5b6810d","relation":"same-reasoning"},{"id":"b-69e663b61c4f79cad9147730","url":"#b-69e663b61c4f79cad9147730","relation":"same-reasoning"},{"id":"b-e08f0c6de0b6eb7f20390724","url":"#b-e08f0c6de0b6eb7f20390724","relation":"same-reasoning"},{"id":"b-d5843ff28a595529c4a8c3fc","url":"#b-d5843ff28a595529c4a8c3fc","relation":"same-reasoning"}]},{"id":"b-7231fd3be61e2d1ffbb52826","type":"p","scope":"roi-information-dialogue","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-7231fd3be61e2d1ffbb52826","content":"**教师：** 先把两个输入条件指出来。论文比较不同区域读取方式训练的模型，特征里有可以利用的空间线索；正确坐标让它读得更好。现在图上甲乙都被我们规定为同一个 .5，框也相同，读取器没有得到任何左右差别。它不是读取错了地方，而是没有不同内容可读。","sha256":"b27d134d90acacfbb8892413111edb846c4f494c80826d510b8f1d49ce8ef265","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-reading-diagnosis","title":"把对齐收益与信息限制放回各自条件","number":2},"sectionTitle":"备课推演：对齐有用，为什么问卷里又不能恢复？","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p43","l3-r2-p44","l3-r2-p45"],"knowledgeBlock":"roi-information-dialogue","formulas":[],"previousBlock":"b-79bba217f96d883cf9a2db27","nextBlock":"b-89b0ad8821b91d8cb5b6810d","prerequisites":[{"id":"b-27569fb134f383b3bdf95c17","url":"#b-27569fb134f383b3bdf95c17","role":"section-title"},{"id":"b-38d1937293601b516a78be69","url":"#b-38d1937293601b516a78be69","role":"problem-context"}],"relatedBlocks":[{"id":"b-27569fb134f383b3bdf95c17","url":"#b-27569fb134f383b3bdf95c17","relation":"same-reasoning"},{"id":"b-fc1b5b517496f8781056aced","url":"#b-fc1b5b517496f8781056aced","relation":"same-reasoning"},{"id":"b-79bba217f96d883cf9a2db27","url":"#b-79bba217f96d883cf9a2db27","relation":"same-reasoning"},{"id":"b-89b0ad8821b91d8cb5b6810d","url":"#b-89b0ad8821b91d8cb5b6810d","relation":"same-reasoning"},{"id":"b-69e663b61c4f79cad9147730","url":"#b-69e663b61c4f79cad9147730","relation":"same-reasoning"},{"id":"b-e08f0c6de0b6eb7f20390724","url":"#b-e08f0c6de0b6eb7f20390724","relation":"same-reasoning"},{"id":"b-d5843ff28a595529c4a8c3fc","url":"#b-d5843ff28a595529c4a8c3fc","relation":"same-reasoning"}]},{"id":"b-89b0ad8821b91d8cb5b6810d","type":"p","scope":"roi-information-dialogue","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-89b0ad8821b91d8cb5b6810d","content":"**学生：** 那我先把 .5 放大成很多像素呢？","sha256":"1b4cb1d1397deda2ab0267237a6a26278baa4ad4eb3b946fac42318a852403d5","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-reading-diagnosis","title":"把对齐收益与信息限制放回各自条件","number":2},"sectionTitle":"备课推演：对齐有用，为什么问卷里又不能恢复？","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p43","l3-r2-p44","l3-r2-p45"],"knowledgeBlock":"roi-information-dialogue","formulas":[],"previousBlock":"b-7231fd3be61e2d1ffbb52826","nextBlock":"b-69e663b61c4f79cad9147730","prerequisites":[{"id":"b-27569fb134f383b3bdf95c17","url":"#b-27569fb134f383b3bdf95c17","role":"section-title"},{"id":"b-38d1937293601b516a78be69","url":"#b-38d1937293601b516a78be69","role":"problem-context"}],"relatedBlocks":[{"id":"b-27569fb134f383b3bdf95c17","url":"#b-27569fb134f383b3bdf95c17","relation":"same-reasoning"},{"id":"b-fc1b5b517496f8781056aced","url":"#b-fc1b5b517496f8781056aced","relation":"same-reasoning"},{"id":"b-79bba217f96d883cf9a2db27","url":"#b-79bba217f96d883cf9a2db27","relation":"same-reasoning"},{"id":"b-7231fd3be61e2d1ffbb52826","url":"#b-7231fd3be61e2d1ffbb52826","relation":"same-reasoning"},{"id":"b-69e663b61c4f79cad9147730","url":"#b-69e663b61c4f79cad9147730","relation":"same-reasoning"},{"id":"b-e08f0c6de0b6eb7f20390724","url":"#b-e08f0c6de0b6eb7f20390724","relation":"same-reasoning"},{"id":"b-d5843ff28a595529c4a8c3fc","url":"#b-d5843ff28a595529c4a8c3fc","relation":"same-reasoning"}]},{"id":"b-69e663b61c4f79cad9147730","type":"p","scope":"roi-information-dialogue","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-69e663b61c4f79cad9147730","content":"**教师：** 指着甲乙放大的第一行看，它们是不是还一样？增加输出位置没有改变函数的输入。再看下一页，我们增加左右差通道，甲是 +1、乙是 −1。这次确实提供了新信息，所以同一个读出可以给出不同答案。","sha256":"22e51595d9a7cf03b3165a3d14e280446eba8bcb2d5f98f6235bb0f8017a3033","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-reading-diagnosis","title":"把对齐收益与信息限制放回各自条件","number":2},"sectionTitle":"备课推演：对齐有用，为什么问卷里又不能恢复？","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p43","l3-r2-p44","l3-r2-p45"],"knowledgeBlock":"roi-information-dialogue","formulas":[],"previousBlock":"b-89b0ad8821b91d8cb5b6810d","nextBlock":"b-e08f0c6de0b6eb7f20390724","prerequisites":[{"id":"b-27569fb134f383b3bdf95c17","url":"#b-27569fb134f383b3bdf95c17","role":"section-title"},{"id":"b-38d1937293601b516a78be69","url":"#b-38d1937293601b516a78be69","role":"problem-context"}],"relatedBlocks":[{"id":"b-27569fb134f383b3bdf95c17","url":"#b-27569fb134f383b3bdf95c17","relation":"same-reasoning"},{"id":"b-fc1b5b517496f8781056aced","url":"#b-fc1b5b517496f8781056aced","relation":"same-reasoning"},{"id":"b-79bba217f96d883cf9a2db27","url":"#b-79bba217f96d883cf9a2db27","relation":"same-reasoning"},{"id":"b-7231fd3be61e2d1ffbb52826","url":"#b-7231fd3be61e2d1ffbb52826","relation":"same-reasoning"},{"id":"b-89b0ad8821b91d8cb5b6810d","url":"#b-89b0ad8821b91d8cb5b6810d","relation":"same-reasoning"},{"id":"b-e08f0c6de0b6eb7f20390724","url":"#b-e08f0c6de0b6eb7f20390724","relation":"same-reasoning"},{"id":"b-d5843ff28a595529c4a8c3fc","url":"#b-d5843ff28a595529c4a8c3fc","relation":"same-reasoning"}]},{"id":"b-e08f0c6de0b6eb7f20390724","type":"p","scope":"roi-information-dialogue","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-e08f0c6de0b6eb7f20390724","content":"**学生：** 实际分割不好，我怎么知道是哪种情况？","sha256":"9c6214cc5b5845cbab52bb501fd71ad4ef2eee82703ed0a35e7c6f6d184ed5e1","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-reading-diagnosis","title":"把对齐收益与信息限制放回各自条件","number":2},"sectionTitle":"备课推演：对齐有用，为什么问卷里又不能恢复？","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p43","l3-r2-p44","l3-r2-p45"],"knowledgeBlock":"roi-information-dialogue","formulas":[],"previousBlock":"b-69e663b61c4f79cad9147730","nextBlock":"b-d5843ff28a595529c4a8c3fc","prerequisites":[{"id":"b-27569fb134f383b3bdf95c17","url":"#b-27569fb134f383b3bdf95c17","role":"section-title"},{"id":"b-38d1937293601b516a78be69","url":"#b-38d1937293601b516a78be69","role":"problem-context"}],"relatedBlocks":[{"id":"b-27569fb134f383b3bdf95c17","url":"#b-27569fb134f383b3bdf95c17","relation":"same-reasoning"},{"id":"b-fc1b5b517496f8781056aced","url":"#b-fc1b5b517496f8781056aced","relation":"same-reasoning"},{"id":"b-79bba217f96d883cf9a2db27","url":"#b-79bba217f96d883cf9a2db27","relation":"same-reasoning"},{"id":"b-7231fd3be61e2d1ffbb52826","url":"#b-7231fd3be61e2d1ffbb52826","relation":"same-reasoning"},{"id":"b-89b0ad8821b91d8cb5b6810d","url":"#b-89b0ad8821b91d8cb5b6810d","relation":"same-reasoning"},{"id":"b-69e663b61c4f79cad9147730","url":"#b-69e663b61c4f79cad9147730","relation":"same-reasoning"},{"id":"b-d5843ff28a595529c4a8c3fc","url":"#b-d5843ff28a595529c4a8c3fc","relation":"same-reasoning"}]},{"id":"b-d5843ff28a595529c4a8c3fc","type":"p","scope":"roi-information-dialogue","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-d5843ff28a595529c4a8c3fc","content":"**教师：** 先检查标注与坐标能否对上，再固定模型做采样对照；若疑似特征不区分，可以尝试接入较细层，或设计有针对性的 probe。一次失败可能有多处原因，不能从一张模糊 mask 直接断言。","sha256":"889643e455f0b36432752ae388e97264df6a2d7c156a17c9b26333c62a8c2257","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-reading-diagnosis","title":"把对齐收益与信息限制放回各自条件","number":2},"sectionTitle":"备课推演：对齐有用，为什么问卷里又不能恢复？","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p43","l3-r2-p44","l3-r2-p45"],"knowledgeBlock":"roi-information-dialogue","formulas":[],"previousBlock":"b-e08f0c6de0b6eb7f20390724","nextBlock":"b-b5ac25da73704b2cfbc21670","prerequisites":[{"id":"b-27569fb134f383b3bdf95c17","url":"#b-27569fb134f383b3bdf95c17","role":"section-title"},{"id":"b-38d1937293601b516a78be69","url":"#b-38d1937293601b516a78be69","role":"problem-context"}],"relatedBlocks":[{"id":"b-27569fb134f383b3bdf95c17","url":"#b-27569fb134f383b3bdf95c17","relation":"same-reasoning"},{"id":"b-fc1b5b517496f8781056aced","url":"#b-fc1b5b517496f8781056aced","relation":"same-reasoning"},{"id":"b-79bba217f96d883cf9a2db27","url":"#b-79bba217f96d883cf9a2db27","relation":"same-reasoning"},{"id":"b-7231fd3be61e2d1ffbb52826","url":"#b-7231fd3be61e2d1ffbb52826","relation":"same-reasoning"},{"id":"b-89b0ad8821b91d8cb5b6810d","url":"#b-89b0ad8821b91d8cb5b6810d","relation":"same-reasoning"},{"id":"b-69e663b61c4f79cad9147730","url":"#b-69e663b61c4f79cad9147730","relation":"same-reasoning"},{"id":"b-e08f0c6de0b6eb7f20390724","url":"#b-e08f0c6de0b6eb7f20390724","relation":"same-reasoning"}]},{"id":"b-b5ac25da73704b2cfbc21670","type":"h2","scope":"practice","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-b5ac25da73704b2cfbc21670","content":"## 沿这个问题继续","sha256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-reading-diagnosis","title":"把对齐收益与信息限制放回各自条件","number":2},"sectionTitle":"把对齐收益与信息限制放回各自条件","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p43","l3-r2-p44","l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-d5843ff28a595529c4a8c3fc","nextBlock":"b-454ceb638ec7e2817b3a4023","prerequisites":[{"id":"b-27569fb134f383b3bdf95c17","url":"#b-27569fb134f383b3bdf95c17","role":"section-title"},{"id":"b-38d1937293601b516a78be69","url":"#b-38d1937293601b516a78be69","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-454ceb638ec7e2817b3a4023","type":"p","scope":"practice","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p45.html#b-454ceb638ec7e2817b3a4023","content":"先读本页解释与图，再到 LAB03 改一个坐标或对象。","sha256":"b8b69250624fd8c15e6f5db8690409848aaec027fc3b89222c6665c3044c9586","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"spatial-information-preservation","title":"瓶颈是读错位置，还是差别根本没有保留？"},"page":{"id":"representation-reading-diagnosis","title":"把对齐收益与信息限制放回各自条件","number":2},"sectionTitle":"把对齐收益与信息限制放回各自条件","currentQuestion":"怎样把信息碰撞的构造例与区域对齐的正面证据放在同一判断中？","sourceSlides":["l3-r2-p43","l3-r2-p44","l3-r2-p45"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-b5ac25da73704b2cfbc21670","nextBlock":null,"prerequisites":[{"id":"b-b5ac25da73704b2cfbc21670","url":"#b-b5ac25da73704b2cfbc21670","role":"section-title"},{"id":"b-38d1937293601b516a78be69","url":"#b-38d1937293601b516a78be69","role":"problem-context"}],"relatedBlocks":[]}]}