{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/blocks/vqa-pixel-evidence.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"已保存的像素 VQA：为什么 44.44% 是结构化基线","url":"https://codingai-lec03.pages.dev/course/blocks/vqa-pixel-evidence.html","markdown":"https://codingai-lec03.pages.dev/course/blocks/vqa-pixel-evidence.md","pages":[{"id":"reading-opening","title":"已保存的像素 VQA：为什么 44.44% 是结构化基线","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":"b-1bd2bc411096013a4a01bf3a","targetId":"b-1bd2bc411096013a4a01bf3a","page":"reading-opening","operation":"move","beforeSHA256":"890d4f05be977f917889c9032a937c40d9b5a6056cf0508b70ae990187a5766e","afterSHA256":"890d4f05be977f917889c9032a937c40d9b5a6056cf0508b70ae990187a5766e","reason":"Same source body in authored semantic page"},{"oldId":"b-a413a156bb7d32a987a2f482","targetId":"b-a413a156bb7d32a987a2f482","page":"reading-opening","operation":"move","beforeSHA256":"e694c4320e7af6f266b80de8b9d0453b3a6861ef55dfd0b426416c93d755292d","afterSHA256":"e694c4320e7af6f266b80de8b9d0453b3a6861ef55dfd0b426416c93d755292d","reason":"Same source body in authored semantic page"},{"oldId":"b-b8fc45f5a4580bf818bce348","targetId":"b-b8fc45f5a4580bf818bce348","page":"reading-opening","operation":"move","beforeSHA256":"fcd7485dc2fc875315c5cfc8e58d8e69bd92e8ee4a2697ee9e296fe461019d35","afterSHA256":"fcd7485dc2fc875315c5cfc8e58d8e69bd92e8ee4a2697ee9e296fe461019d35","reason":"Same source body in authored semantic page"},{"oldId":"b-e7d5cbc3531b7f17766d5a44","targetId":"b-e7d5cbc3531b7f17766d5a44","page":"reading-opening","operation":"move","beforeSHA256":"8f89da27040af7a6e6079498b094994e2ce39ffaa7097395a9aaded0497702ee","afterSHA256":"8f89da27040af7a6e6079498b094994e2ce39ffaa7097395a9aaded0497702ee","reason":"Same source body in authored semantic page"},{"oldId":"b-d27878ef91bbc44059054934","targetId":"b-d27878ef91bbc44059054934","page":"reading-opening","operation":"rewrite","beforeSHA256":"18d38de4f1210841fff977f33e9a1628d7b29c97243b63c8ab8da5b881e7ded1","afterSHA256":"bc3dba74f3009c6b434eeacf463c0a84cdb73c41e93fce898cc9e87feae52e24","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-5db68e6bda7c8816c4026dac","targetId":"b-5db68e6bda7c8816c4026dac","page":"reading-opening","operation":"move","beforeSHA256":"5dca656f2c377109930a783b3082cbf5d2ad432af0a58c058e98f570fe78f094","afterSHA256":"5dca656f2c377109930a783b3082cbf5d2ad432af0a58c058e98f570fe78f094","reason":"Same source body in authored semantic page"},{"oldId":"b-81d9f34e5d2856fa724a65ad","targetId":"b-81d9f34e5d2856fa724a65ad","page":"reading-opening","operation":"move","beforeSHA256":"dd655aba933eeca3efa133773efe24b93715ba055621e0986f10caa85cd381e9","afterSHA256":"08d0aa022e98bb75079211b073bb26e8245e159a9c4dcb17e3839b2d504f1d23","reason":"Same source body in authored semantic page"},{"oldId":"b-ac92a3ed1ddb896d004303bd","targetId":"b-ac92a3ed1ddb896d004303bd","page":"reading-opening","operation":"move","beforeSHA256":"7d2f78f1122d4e8c9846b7131965a7250236f51bc8d561d375ce41bac04fe7f6","afterSHA256":"7d2f78f1122d4e8c9846b7131965a7250236f51bc8d561d375ce41bac04fe7f6","reason":"Same source body in authored semantic page"},{"oldId":"b-4a918e35210636b359d23e79","targetId":"b-4a918e35210636b359d23e79","page":"reading-opening","operation":"move","beforeSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","afterSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","reason":"Same source body in authored semantic page"}],"blocks":[{"id":"b-1bd2bc411096013a4a01bf3a","type":"h1","scope":"vqa-pixel-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/vqa-pixel-evidence.html#b-1bd2bc411096013a4a01bf3a","content":"# 已保存的像素 VQA：为什么 44.44% 是结构化基线","sha256":"890d4f05be977f917889c9032a937c40d9b5a6056cf0508b70ae990187a5766e","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":"reading-opening","title":"已保存的像素 VQA：为什么 44.44% 是结构化基线","number":1},"sectionTitle":"已保存的像素 VQA：为什么 44.44% 是结构化基线","currentQuestion":"已保存的像素 VQA：为什么 44.44% 是结构化基线","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-a413a156bb7d32a987a2f482","prerequisites":[{"id":"b-a413a156bb7d32a987a2f482","url":"#b-a413a156bb7d32a987a2f482","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-a413a156bb7d32a987a2f482","type":"p","scope":"vqa-pixel-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/vqa-pixel-evidence.html#b-a413a156bb7d32a987a2f482","content":"问题在这里选择要读的信息，同一图可以问数量、属性和关系。本页说明沿两个方向组织检查：同图换问、同问换图。LAB14 当前已经有读取像素的可学习分支和单独训练的 question-only，对象属性字典保留为 oracle 热身；三种身份应分别阅读。","sha256":"e694c4320e7af6f266b80de8b9d0453b3a6861ef55dfd0b426416c93d755292d","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":"reading-opening","title":"已保存的像素 VQA：为什么 44.44% 是结构化基线","number":1},"sectionTitle":"已保存的像素 VQA：为什么 44.44% 是结构化基线","currentQuestion":"已保存的像素 VQA：为什么 44.44% 是结构化基线","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-1bd2bc411096013a4a01bf3a","nextBlock":"b-b8fc45f5a4580bf818bce348","prerequisites":[{"id":"b-1bd2bc411096013a4a01bf3a","url":"#b-1bd2bc411096013a4a01bf3a","role":"section-title"}],"relatedBlocks":[]},{"id":"b-b8fc45f5a4580bf818bce348","type":"p","scope":"vqa-pixel-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/vqa-pixel-evidence.html#b-b8fc45f5a4580bf818bce348","content":"LAB14 在 2026-09-22 保存了本地 CPU 学习结果，本次整理没有重跑。720 张训练图和 180 张保留图先按图像生成和划分，再对每图提出三种固定模板问题，共 2160/540 个图问对。模型实际读取 RGB 像素和问题 ID，不读取生成器属性字典。","sha256":"fcd7485dc2fc875315c5cfc8e58d8e69bd92e8ee4a2697ee9e296fe461019d35","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":"reading-opening","title":"已保存的像素 VQA：为什么 44.44% 是结构化基线","number":1},"sectionTitle":"已保存的像素 VQA：为什么 44.44% 是结构化基线","currentQuestion":"已保存的像素 VQA：为什么 44.44% 是结构化基线","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-a413a156bb7d32a987a2f482","nextBlock":"b-e7d5cbc3531b7f17766d5a44","prerequisites":[{"id":"b-1bd2bc411096013a4a01bf3a","url":"#b-1bd2bc411096013a4a01bf3a","role":"section-title"},{"id":"b-a413a156bb7d32a987a2f482","url":"#b-a413a156bb7d32a987a2f482","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-e7d5cbc3531b7f17766d5a44","type":"p","scope":"vqa-pixel-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/vqa-pixel-evidence.html#b-e7d5cbc3531b7f17766d5a44","content":"保留任务分别是数量二选一、左侧颜色三选一、右侧形状二选一。只知道问题时，各任务的最佳常量正确率为 $1/2,1/3,1/2$，平均 $(1/2+1/3+1/2)/3=4/9$。把所有七种答案混在一起写“机会率 1/7”会忽略问题已给出的答案类型。","sha256":"8f89da27040af7a6e6079498b094994e2ce39ffaa7097395a9aaded0497702ee","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":"reading-opening","title":"已保存的像素 VQA：为什么 44.44% 是结构化基线","number":1},"sectionTitle":"已保存的像素 VQA：为什么 44.44% 是结构化基线","currentQuestion":"已保存的像素 VQA：为什么 44.44% 是结构化基线","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"1/2,1/3,1/2","definitionContext":"已保存的像素 VQA：为什么 44.44% 是结构化基线"},{"latex":"(1/2+1/3+1/2)/3=4/9","definitionContext":"已保存的像素 VQA：为什么 44.44% 是结构化基线"}],"previousBlock":"b-b8fc45f5a4580bf818bce348","nextBlock":"b-d27878ef91bbc44059054934","prerequisites":[{"id":"b-1bd2bc411096013a4a01bf3a","url":"#b-1bd2bc411096013a4a01bf3a","role":"section-title"},{"id":"b-a413a156bb7d32a987a2f482","url":"#b-a413a156bb7d32a987a2f482","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-d27878ef91bbc44059054934","type":"table","scope":"vqa-pixel-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/vqa-pixel-evidence.html#b-d27878ef91bbc44059054934","content":"| 保留条件 | 数量 | 左色 | 右形 | 全部 |\n| --- | --- | --- | --- | --- |\n| 图像＋问题 | 100% | 100% | 100% | 100% |\n| 单独训练 question-only | 50% | 33.33% | 50% | 44.44% |\n| 同一视觉模型，像素置零 | 50% | 33.33% | 50% | 44.44% |\n| 同一视觉模型，循环错配图像 | 83.33% | 50% | 0% | 44.44% |","sha256":"bc3dba74f3009c6b434eeacf463c0a84cdb73c41e93fce898cc9e87feae52e24","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":"reading-opening","title":"已保存的像素 VQA：为什么 44.44% 是结构化基线","number":1},"sectionTitle":"已保存的像素 VQA：为什么 44.44% 是结构化基线","currentQuestion":"已保存的像素 VQA：为什么 44.44% 是结构化基线","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"table":{"headers":["保留条件","数量","左色","右形","全部"],"rows":[["保留条件","数量","左色","右形","全部"],["图像＋问题","100%","100%","100%","100%"],["单独训练 question-only","50%","33.33%","50%","44.44%"],["同一视觉模型，像素置零","50%","33.33%","50%","44.44%"],["同一视觉模型，循环错配图像","83.33%","50%","0%","44.44%"]]},"previousBlock":"b-e7d5cbc3531b7f17766d5a44","nextBlock":"b-5db68e6bda7c8816c4026dac","prerequisites":[{"id":"b-1bd2bc411096013a4a01bf3a","url":"#b-1bd2bc411096013a4a01bf3a","role":"section-title"},{"id":"b-a413a156bb7d32a987a2f482","url":"#b-a413a156bb7d32a987a2f482","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-5db68e6bda7c8816c4026dac","type":"p","scope":"vqa-pixel-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/vqa-pixel-evidence.html#b-5db68e6bda7c8816c4026dac","content":"最后一行不是随机机会率：数据按组合排列，循环移位常保留数量却恰好改变形状，因此总分相同仍有不同错误来源。正确图像下的模型通过了三个有限任务；像素置零和错配的下降支持它使用正确视觉输入。训练左侧红色比例为 73.61%，保留颜色均衡，进一步显示问题先验不能替代视觉读取。","sha256":"5dca656f2c377109930a783b3082cbf5d2ad432af0a58c058e98f570fe78f094","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":"reading-opening","title":"已保存的像素 VQA：为什么 44.44% 是结构化基线","number":1},"sectionTitle":"已保存的像素 VQA：为什么 44.44% 是结构化基线","currentQuestion":"已保存的像素 VQA：为什么 44.44% 是结构化基线","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-d27878ef91bbc44059054934","nextBlock":"b-81d9f34e5d2856fa724a65ad","prerequisites":[{"id":"b-1bd2bc411096013a4a01bf3a","url":"#b-1bd2bc411096013a4a01bf3a","role":"section-title"},{"id":"b-a413a156bb7d32a987a2f482","url":"#b-a413a156bb7d32a987a2f482","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-81d9f34e5d2856fa724a65ad","type":"p","scope":"vqa-pixel-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/vqa-pixel-evidence.html#b-81d9f34e5d2856fa724a65ad","content":"![同图换问、同问换图](/assets/display/2f7b2168f5b21ee2095709fa.webp)","sha256":"08d0aa022e98bb75079211b073bb26e8245e159a9c4dcb17e3839b2d504f1d23","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":"reading-opening","title":"已保存的像素 VQA：为什么 44.44% 是结构化基线","number":1},"sectionTitle":"已保存的像素 VQA：为什么 44.44% 是结构化基线","currentQuestion":"已保存的像素 VQA：为什么 44.44% 是结构化基线","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-5db68e6bda7c8816c4026dac","nextBlock":"b-ac92a3ed1ddb896d004303bd","prerequisites":[{"id":"b-1bd2bc411096013a4a01bf3a","url":"#b-1bd2bc411096013a4a01bf3a","role":"section-title"},{"id":"b-a413a156bb7d32a987a2f482","url":"#b-a413a156bb7d32a987a2f482","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-ac92a3ed1ddb896d004303bd","type":"p","scope":"vqa-pixel-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/vqa-pixel-evidence.html#b-ac92a3ed1ddb896d004303bd","content":"问题输入在这里是三个模板 ID，不是任意句子的语言理解。先承认这个小实验确实完成的像素问答，再把开放问题、未见对象与组合列为下一层检验。","sha256":"7d2f78f1122d4e8c9846b7131965a7250236f51bc8d561d375ce41bac04fe7f6","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":"reading-opening","title":"已保存的像素 VQA：为什么 44.44% 是结构化基线","number":1},"sectionTitle":"已保存的像素 VQA：为什么 44.44% 是结构化基线","currentQuestion":"已保存的像素 VQA：为什么 44.44% 是结构化基线","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-81d9f34e5d2856fa724a65ad","nextBlock":"b-4a918e35210636b359d23e79","prerequisites":[{"id":"b-1bd2bc411096013a4a01bf3a","url":"#b-1bd2bc411096013a4a01bf3a","role":"section-title"},{"id":"b-a413a156bb7d32a987a2f482","url":"#b-a413a156bb7d32a987a2f482","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-4a918e35210636b359d23e79","type":"h2","scope":"vqa-pixel-evidence","url":"https://codingai-lec03.pages.dev/course/blocks/vqa-pixel-evidence.html#b-4a918e35210636b359d23e79","content":"## 继续阅读与实验","sha256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","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":"reading-opening","title":"已保存的像素 VQA：为什么 44.44% 是结构化基线","number":1},"sectionTitle":"已保存的像素 VQA：为什么 44.44% 是结构化基线","currentQuestion":"已保存的像素 VQA：为什么 44.44% 是结构化基线","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-ac92a3ed1ddb896d004303bd","nextBlock":null,"prerequisites":[{"id":"b-1bd2bc411096013a4a01bf3a","url":"#b-1bd2bc411096013a4a01bf3a","role":"section-title"},{"id":"b-a413a156bb7d32a987a2f482","url":"#b-a413a156bb7d32a987a2f482","role":"problem-context"}],"relatedBlocks":[]}]}