{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/blocks/caption-visual-controls.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"已保存的条件生成实验：视觉输入确实改变了句子","url":"https://codingai-lec03.pages.dev/course/blocks/caption-visual-controls.html","markdown":"https://codingai-lec03.pages.dev/course/blocks/caption-visual-controls.md","pages":[{"id":"reading-opening","title":"已保存的条件生成实验：视觉输入确实改变了句子","unit":"图像怎样约束一句新生成的描述？","unitId":"caption-visual-conditioning","chapter":"III","number":1}],"aliases":[{"oldId":"b-5e7522fb237280023565d55a","targetId":"b-5e7522fb237280023565d55a","page":"reading-opening","operation":"move","beforeSHA256":"2bcf597662b8fc1bdc8606738c1cdc6fbb0fdaf190b3f79c0eb2533c26e2dd23","afterSHA256":"2bcf597662b8fc1bdc8606738c1cdc6fbb0fdaf190b3f79c0eb2533c26e2dd23","reason":"Same source body in authored semantic page"},{"oldId":"b-2c3b1b4a073a38d228501c8d","targetId":"b-2c3b1b4a073a38d228501c8d","page":"reading-opening","operation":"move","beforeSHA256":"aeaad29331bc2c94ac240ece646024336e7455de08d754b00ed438164663d9be","afterSHA256":"aeaad29331bc2c94ac240ece646024336e7455de08d754b00ed438164663d9be","reason":"Same source body in authored semantic page"},{"oldId":"b-8f6087b7058cd0db87b96a0a","targetId":"b-8f6087b7058cd0db87b96a0a","page":"reading-opening","operation":"move","beforeSHA256":"c1e2bdbd6a16066dce6e56451311b7266e3a63bc810800f91e36da038f829364","afterSHA256":"c1e2bdbd6a16066dce6e56451311b7266e3a63bc810800f91e36da038f829364","reason":"Same source body in authored semantic page"},{"oldId":"b-be1c06eb90a60a5b52bf8dba","targetId":"b-be1c06eb90a60a5b52bf8dba","page":"reading-opening","operation":"move","beforeSHA256":"d6e2fdeea4c473a4ad3b0b4a0a761eeafdcdf897215012754c0cfc5db9ebb042","afterSHA256":"d6e2fdeea4c473a4ad3b0b4a0a761eeafdcdf897215012754c0cfc5db9ebb042","reason":"Same source body in authored semantic page"},{"oldId":"b-9de81c5d3cd0ed1e93368423","targetId":"b-9de81c5d3cd0ed1e93368423","page":"reading-opening","operation":"rewrite","beforeSHA256":"8c976ba19945b3c06c115239582117ede7196427a44028e02bdff62b158b6800","afterSHA256":"6198a4971317db245f284fd359e91844808af55181f829edcbe77f7009c63809","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-d929ce6cc357e7da761f0b24","targetId":"b-d929ce6cc357e7da761f0b24","page":"reading-opening","operation":"move","beforeSHA256":"1f615531076a140376b37f33392e5543bedc80c8ddb1b9b6483dd6d7e4d5f06b","afterSHA256":"1f615531076a140376b37f33392e5543bedc80c8ddb1b9b6483dd6d7e4d5f06b","reason":"Same source body in authored semantic page"},{"oldId":"b-fb7d2245e1c57371369fda74","targetId":"b-fb7d2245e1c57371369fda74","page":"reading-opening","operation":"move","beforeSHA256":"890fbc23abc501b6a296dd283f5d91269fcdc5fe37222a92fa880db818085b66","afterSHA256":"f6aedf109a28f615cb59b69833b6929e30ff48023b1eabc37990766d7cfd0565","reason":"Same source body in authored semantic page"},{"oldId":"b-3bc536205ce348208c3c5636","targetId":"b-3bc536205ce348208c3c5636","page":"reading-opening","operation":"move","beforeSHA256":"38fc170e9722124fda4bcec90baa5dd1ee7e03550b153f899912a90c1fbb4ce0","afterSHA256":"38fc170e9722124fda4bcec90baa5dd1ee7e03550b153f899912a90c1fbb4ce0","reason":"Same source body in authored semantic page"},{"oldId":"b-85b72d32a30ec24c377f63b0","targetId":"b-85b72d32a30ec24c377f63b0","page":"reading-opening","operation":"move","beforeSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","afterSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","reason":"Same source body in authored semantic page"}],"blocks":[{"id":"b-5e7522fb237280023565d55a","type":"h1","scope":"caption-visual-controls","url":"https://codingai-lec03.pages.dev/course/blocks/caption-visual-controls.html#b-5e7522fb237280023565d55a","content":"# 已保存的条件生成实验：视觉输入确实改变了句子","sha256":"2bcf597662b8fc1bdc8606738c1cdc6fbb0fdaf190b3f79c0eb2533c26e2dd23","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"caption-visual-conditioning","title":"图像怎样约束一句新生成的描述？"},"page":{"id":"reading-opening","title":"已保存的条件生成实验：视觉输入确实改变了句子","number":1},"sectionTitle":"已保存的条件生成实验：视觉输入确实改变了句子","currentQuestion":"已保存的条件生成实验：视觉输入确实改变了句子","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-2c3b1b4a073a38d228501c8d","prerequisites":[{"id":"b-2c3b1b4a073a38d228501c8d","url":"#b-2c3b1b4a073a38d228501c8d","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-2c3b1b4a073a38d228501c8d","type":"p","scope":"caption-visual-controls","url":"https://codingai-lec03.pages.dev/course/blocks/caption-visual-controls.html#b-2c3b1b4a073a38d228501c8d","content":"先沿画面指出视觉特征进入哪里，再指出已有文字与下一词目标。连续说明展开 shifted target、EOS、PAD 和实际生成，LAB13 提供已保存的实验结果。理解本页可先把文字预测器作为输入输出模块；整数移位只演示目标构造，完整图像描述还需要图像条件与实际生成。","sha256":"aeaad29331bc2c94ac240ece646024336e7455de08d754b00ed438164663d9be","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"caption-visual-conditioning","title":"图像怎样约束一句新生成的描述？"},"page":{"id":"reading-opening","title":"已保存的条件生成实验：视觉输入确实改变了句子","number":1},"sectionTitle":"已保存的条件生成实验：视觉输入确实改变了句子","currentQuestion":"已保存的条件生成实验：视觉输入确实改变了句子","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-5e7522fb237280023565d55a","nextBlock":"b-8f6087b7058cd0db87b96a0a","prerequisites":[{"id":"b-5e7522fb237280023565d55a","url":"#b-5e7522fb237280023565d55a","role":"section-title"}],"relatedBlocks":[]},{"id":"b-8f6087b7058cd0db87b96a0a","type":"p","scope":"caption-visual-controls","url":"https://codingai-lec03.pages.dev/course/blocks/caption-visual-controls.html#b-8f6087b7058cd0db87b96a0a","content":"LAB13 当前保存的是 2026-09-22 本地 CPU 基准，当前整理没有重新训练。输入为真实渲染的 $32\\times32$ RGB 图像；训练 720 张，保留 180 张。颜色、形状、位置共有 18 种组合，两侧都包含这些组合，但保留图由独立随机渲染产生、像素哈希不重复。这检验新渲染泛化，不是未见组合泛化。","sha256":"c1e2bdbd6a16066dce6e56451311b7266e3a63bc810800f91e36da038f829364","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"caption-visual-conditioning","title":"图像怎样约束一句新生成的描述？"},"page":{"id":"reading-opening","title":"已保存的条件生成实验：视觉输入确实改变了句子","number":1},"sectionTitle":"已保存的条件生成实验：视觉输入确实改变了句子","currentQuestion":"已保存的条件生成实验：视觉输入确实改变了句子","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"32\\times32","definitionContext":"已保存的条件生成实验：视觉输入确实改变了句子"}],"previousBlock":"b-2c3b1b4a073a38d228501c8d","nextBlock":"b-be1c06eb90a60a5b52bf8dba","prerequisites":[{"id":"b-5e7522fb237280023565d55a","url":"#b-5e7522fb237280023565d55a","role":"section-title"},{"id":"b-2c3b1b4a073a38d228501c8d","url":"#b-2c3b1b4a073a38d228501c8d","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-be1c06eb90a60a5b52bf8dba","type":"p","scope":"caption-visual-controls","url":"https://codingai-lec03.pages.dev/course/blocks/caption-visual-controls.html#b-be1c06eb90a60a5b52bf8dba","content":"CNN 输出 48 维视觉向量；文字单位有 24 维嵌入；实验采用一个小型 GRU 文字模块。今天把它作为已明确输入输出的实现，内部循环状态留第五讲。视觉向量确实参与每一步预测，并用于初始化文字模块；真实颜色或位置标签只用于构造监督，不作为推理输入。","sha256":"d6e2fdeea4c473a4ad3b0b4a0a761eeafdcdf897215012754c0cfc5db9ebb042","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"caption-visual-conditioning","title":"图像怎样约束一句新生成的描述？"},"page":{"id":"reading-opening","title":"已保存的条件生成实验：视觉输入确实改变了句子","number":1},"sectionTitle":"已保存的条件生成实验：视觉输入确实改变了句子","currentQuestion":"已保存的条件生成实验：视觉输入确实改变了句子","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-8f6087b7058cd0db87b96a0a","nextBlock":"b-9de81c5d3cd0ed1e93368423","prerequisites":[{"id":"b-5e7522fb237280023565d55a","url":"#b-5e7522fb237280023565d55a","role":"section-title"},{"id":"b-2c3b1b4a073a38d228501c8d","url":"#b-2c3b1b4a073a38d228501c8d","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-9de81c5d3cd0ed1e93368423","type":"table","scope":"caption-visual-controls","url":"https://codingai-lec03.pages.dev/course/blocks/caption-visual-controls.html#b-9de81c5d3cd0ed1e93368423","content":"| 同一组 180 张保留图 | 整句精确匹配 |\n| --- | --- |\n| 正确图像，greedy 生成 | 100% |\n| 正确图像，beam 宽度 3 | 100% |\n| 同一模型，像素置零 | 5.56% |\n| 同一模型，图像循环错配 | 0% |","sha256":"6198a4971317db245f284fd359e91844808af55181f829edcbe77f7009c63809","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"caption-visual-conditioning","title":"图像怎样约束一句新生成的描述？"},"page":{"id":"reading-opening","title":"已保存的条件生成实验：视觉输入确实改变了句子","number":1},"sectionTitle":"已保存的条件生成实验：视觉输入确实改变了句子","currentQuestion":"已保存的条件生成实验：视觉输入确实改变了句子","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"table":{"headers":["同一组 180 张保留图","整句精确匹配"],"rows":[["同一组 180 张保留图","整句精确匹配"],["正确图像，greedy 生成","100%"],["正确图像，beam 宽度 3","100%"],["同一模型，像素置零","5.56%"],["同一模型，图像循环错配","0%"]]},"previousBlock":"b-be1c06eb90a60a5b52bf8dba","nextBlock":"b-d929ce6cc357e7da761f0b24","prerequisites":[{"id":"b-5e7522fb237280023565d55a","url":"#b-5e7522fb237280023565d55a","role":"section-title"},{"id":"b-2c3b1b4a073a38d228501c8d","url":"#b-2c3b1b4a073a38d228501c8d","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-d929ce6cc357e7da761f0b24","type":"p","scope":"caption-visual-controls","url":"https://codingai-lec03.pages.dev/course/blocks/caption-visual-controls.html#b-d929ce6cc357e7da761f0b24","content":"固定前缀 `BOS a`，红方块左侧图生成 `a red square on left`，蓝圆右侧图生成 `a blue circle on right`。这里正面成立的判断是：这个有限任务中，训练后的输出使用了正确视觉条件。两种搜索同分，不支持 beam 比 greedy 更好；零图像是分布外条件，错配对照则进一步隔离了图像对应关系。自然图像有多种合理描述，不能把这里唯一规范句子的精确匹配直接移作通用 caption 评价。","sha256":"1f615531076a140376b37f33392e5543bedc80c8ddb1b9b6483dd6d7e4d5f06b","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"caption-visual-conditioning","title":"图像怎样约束一句新生成的描述？"},"page":{"id":"reading-opening","title":"已保存的条件生成实验：视觉输入确实改变了句子","number":1},"sectionTitle":"已保存的条件生成实验：视觉输入确实改变了句子","currentQuestion":"已保存的条件生成实验：视觉输入确实改变了句子","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-9de81c5d3cd0ed1e93368423","nextBlock":"b-fb7d2245e1c57371369fda74","prerequisites":[{"id":"b-5e7522fb237280023565d55a","url":"#b-5e7522fb237280023565d55a","role":"section-title"},{"id":"b-2c3b1b4a073a38d228501c8d","url":"#b-2c3b1b4a073a38d228501c8d","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-fb7d2245e1c57371369fda74","type":"p","scope":"caption-visual-controls","url":"https://codingai-lec03.pages.dev/course/blocks/caption-visual-controls.html#b-fb7d2245e1c57371369fda74","content":"![固定前缀下的真实生成](/assets/display/9ebe76d49a449b9940b6470f.webp)","sha256":"f6aedf109a28f615cb59b69833b6929e30ff48023b1eabc37990766d7cfd0565","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"caption-visual-conditioning","title":"图像怎样约束一句新生成的描述？"},"page":{"id":"reading-opening","title":"已保存的条件生成实验：视觉输入确实改变了句子","number":1},"sectionTitle":"已保存的条件生成实验：视觉输入确实改变了句子","currentQuestion":"已保存的条件生成实验：视觉输入确实改变了句子","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-d929ce6cc357e7da761f0b24","nextBlock":"b-3bc536205ce348208c3c5636","prerequisites":[{"id":"b-5e7522fb237280023565d55a","url":"#b-5e7522fb237280023565d55a","role":"section-title"},{"id":"b-2c3b1b4a073a38d228501c8d","url":"#b-2c3b1b4a073a38d228501c8d","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-3bc536205ce348208c3c5636","type":"p","scope":"caption-visual-controls","url":"https://codingai-lec03.pages.dev/course/blocks/caption-visual-controls.html#b-3bc536205ce348208c3c5636","content":"图片与数字均来自已保存基准，不是把数据生成器的目标文字重新印成“模型回答”。","sha256":"38fc170e9722124fda4bcec90baa5dd1ee7e03550b153f899912a90c1fbb4ce0","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"caption-visual-conditioning","title":"图像怎样约束一句新生成的描述？"},"page":{"id":"reading-opening","title":"已保存的条件生成实验：视觉输入确实改变了句子","number":1},"sectionTitle":"已保存的条件生成实验：视觉输入确实改变了句子","currentQuestion":"已保存的条件生成实验：视觉输入确实改变了句子","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-fb7d2245e1c57371369fda74","nextBlock":"b-85b72d32a30ec24c377f63b0","prerequisites":[{"id":"b-5e7522fb237280023565d55a","url":"#b-5e7522fb237280023565d55a","role":"section-title"},{"id":"b-2c3b1b4a073a38d228501c8d","url":"#b-2c3b1b4a073a38d228501c8d","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-85b72d32a30ec24c377f63b0","type":"h2","scope":"caption-visual-controls","url":"https://codingai-lec03.pages.dev/course/blocks/caption-visual-controls.html#b-85b72d32a30ec24c377f63b0","content":"## 继续阅读与实验","sha256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","chapter":{"id":"III","title":"视觉与语言"},"unit":{"id":"caption-visual-conditioning","title":"图像怎样约束一句新生成的描述？"},"page":{"id":"reading-opening","title":"已保存的条件生成实验：视觉输入确实改变了句子","number":1},"sectionTitle":"已保存的条件生成实验：视觉输入确实改变了句子","currentQuestion":"已保存的条件生成实验：视觉输入确实改变了句子","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-3bc536205ce348208c3c5636","nextBlock":null,"prerequisites":[{"id":"b-5e7522fb237280023565d55a","url":"#b-5e7522fb237280023565d55a","role":"section-title"},{"id":"b-2c3b1b4a073a38d228501c8d","url":"#b-2c3b1b4a073a38d228501c8d","role":"problem-context"}],"relatedBlocks":[]}]}