{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"图像、前缀、目标，三个位置分开看","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html","markdown":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.md","pages":[{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","unit":"图像、前缀、目标，三个位置分开看","unitId":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","chapter":"course","number":1}],"aliases":[{"oldId":"b-357db54056cf2501bd95da51","targetId":"b-357db54056cf2501bd95da51","page":"reading-opening","operation":"move","beforeSHA256":"fa1d05e087449a3f39b93c1c4f3e6086732599f759cb40cfca565caa8afb8709","afterSHA256":"fa1d05e087449a3f39b93c1c4f3e6086732599f759cb40cfca565caa8afb8709","reason":"Same source body in authored semantic page"},{"oldId":"b-1a26fa22a11915c40c3c5922","targetId":"b-1a26fa22a11915c40c3c5922","page":"reading-opening","operation":"move","beforeSHA256":"77c37b608fd8aa55ce3223d141440f1b400d22c00bc453aa25f725d0f5e47610","afterSHA256":"77c37b608fd8aa55ce3223d141440f1b400d22c00bc453aa25f725d0f5e47610","reason":"Same source body in authored semantic page"},{"oldId":"b-e4fbe0dc1575bb96929ee494","targetId":"b-e4fbe0dc1575bb96929ee494","page":"reading-opening","operation":"move","beforeSHA256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","afterSHA256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","reason":"Same source body in authored semantic page"},{"oldId":"b-7b9b63e4812e8cbb0563586f","targetId":"b-7b9b63e4812e8cbb0563586f","page":"reading-opening","operation":"move","beforeSHA256":"a192bf58b03c510999643a276b8d2261f6c9389ea09aebd6ad3419fe70e2a7b9","afterSHA256":"a192bf58b03c510999643a276b8d2261f6c9389ea09aebd6ad3419fe70e2a7b9","reason":"Same source body in authored semantic page"},{"oldId":"b-803c5c50d09342ca589335c7","targetId":"b-803c5c50d09342ca589335c7","page":"reading-opening","operation":"move","beforeSHA256":"a57513e59eb07960710e5ece7951cc57af4a94c446630eb5fc6d70b460982042","afterSHA256":"a57513e59eb07960710e5ece7951cc57af4a94c446630eb5fc6d70b460982042","reason":"Same source body in authored semantic page"},{"oldId":"b-e9ff65a3dc6dcc06d33e0065","targetId":"b-e9ff65a3dc6dcc06d33e0065","page":"reading-opening","operation":"move","beforeSHA256":"6b3101dc74d613fd527b60ff21d1ac65db81cca6fc49d581090c9a90654d2057","afterSHA256":"6b3101dc74d613fd527b60ff21d1ac65db81cca6fc49d581090c9a90654d2057","reason":"Same source body in authored semantic page"},{"oldId":"b-aed5f706ce45f4ef3e5d9a4d","targetId":"b-aed5f706ce45f4ef3e5d9a4d","page":"reading-opening","operation":"move","beforeSHA256":"98588b7069ba7f92211a8e6cd5c8be5f8249ccf2425e964a426643a339f7a008","afterSHA256":"98588b7069ba7f92211a8e6cd5c8be5f8249ccf2425e964a426643a339f7a008","reason":"Same source body in authored semantic page"},{"oldId":"b-269f0352fbe73e55e448d907","targetId":"b-269f0352fbe73e55e448d907","page":"reading-opening","operation":"move","beforeSHA256":"46d053d1cc569605d484255e6af2c3466f7d40f4fe2b28cec955d4de4a9dd9d8","afterSHA256":"46d053d1cc569605d484255e6af2c3466f7d40f4fe2b28cec955d4de4a9dd9d8","reason":"Same source body in authored semantic page"},{"oldId":"b-425258f91a78c5039d475615","targetId":"b-425258f91a78c5039d475615","page":"reading-opening","operation":"move","beforeSHA256":"5d04e11d8c2e016f93a6fd003c5aafbcfa45431a430ae351461bb69f38e77a4d","afterSHA256":"5d04e11d8c2e016f93a6fd003c5aafbcfa45431a430ae351461bb69f38e77a4d","reason":"Same source body in authored semantic page"},{"oldId":"b-420d91de3ca262a0fe45bcec","targetId":"b-420d91de3ca262a0fe45bcec","page":"reading-opening","operation":"move","beforeSHA256":"621909530fe228f941698b8aa1acd5e19c591377be6a124d71ff93cc33cf1036","afterSHA256":"621909530fe228f941698b8aa1acd5e19c591377be6a124d71ff93cc33cf1036","reason":"Same source body in authored semantic page"},{"oldId":"b-dc4db5fb67e99fb23621bb7c","targetId":"b-dc4db5fb67e99fb23621bb7c","page":"reading-opening","operation":"move","beforeSHA256":"d3f6a0e9d3a04e431b7a9c6faca644a0357974e570ae3eea66067da85832db98","afterSHA256":"d3f6a0e9d3a04e431b7a9c6faca644a0357974e570ae3eea66067da85832db98","reason":"Same source body in authored semantic page"},{"oldId":"b-a454522b892395292d1d1585","targetId":"b-a454522b892395292d1d1585","page":"reading-opening","operation":"move","beforeSHA256":"132ec47bbd37bd8b6573626070e54b3184a5c5a21294a10bd9ab4c419181acc7","afterSHA256":"132ec47bbd37bd8b6573626070e54b3184a5c5a21294a10bd9ab4c419181acc7","reason":"Same source body in authored semantic page"},{"oldId":"b-58789586d61178dfcfebbe36","targetId":"b-58789586d61178dfcfebbe36","page":"reading-opening","operation":"move","beforeSHA256":"9e28174c8843e3c00738239016bd81d74ee8e3b5857fa1ff02121cc770476f86","afterSHA256":"9e28174c8843e3c00738239016bd81d74ee8e3b5857fa1ff02121cc770476f86","reason":"Same source body in authored semantic page"},{"oldId":"b-c1645412cfb593aa7f61e29f","targetId":"b-c1645412cfb593aa7f61e29f","page":"reading-opening","operation":"move","beforeSHA256":"18fc193179c81a0c886864470e3271551a6d8a3537b9fc9d935e5490e732aadb","afterSHA256":"18fc193179c81a0c886864470e3271551a6d8a3537b9fc9d935e5490e732aadb","reason":"Same source body in authored semantic page"},{"oldId":"b-709137a6f9d87f1c3ff701c2","targetId":"b-709137a6f9d87f1c3ff701c2","page":"reading-opening","operation":"move","beforeSHA256":"70da78e6d74f17d276ba01f6377f5f483608d6e486db66f7ffab4af473c8c68d","afterSHA256":"70da78e6d74f17d276ba01f6377f5f483608d6e486db66f7ffab4af473c8c68d","reason":"Same source body in authored semantic page"},{"oldId":"b-297c5a9500763b144618f1fa","targetId":"b-297c5a9500763b144618f1fa","page":"reading-opening","operation":"move","beforeSHA256":"6bca41597071d63a25f16e927be8d8382b5da7c125bd9f583e5399e0520cc939","afterSHA256":"6bca41597071d63a25f16e927be8d8382b5da7c125bd9f583e5399e0520cc939","reason":"Same source body in authored semantic page"},{"oldId":"b-0af341b5341eedefc785227d","targetId":"b-0af341b5341eedefc785227d","page":"reading-opening","operation":"rewrite","beforeSHA256":"a77ee9b5843b0305eb36f4a9cad8be08c2b7708ae228d53535a9946b02f8b488","afterSHA256":"76596d43d047ab4ae6858c71f16377d45d925bd1b3cb8a695d4f1606a5d9cbc1","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-8e89985e66ebcf261d135e59","targetId":"b-8e89985e66ebcf261d135e59","page":"reading-opening","operation":"move","beforeSHA256":"e85602c65ca5ff78574d827dfabe48bbd83b4b0913f2f7b07f928129a1cd5593","afterSHA256":"5fad46273be0df138b5d8cd83825e637a26e4d2054efd42ac82536606c85cd42","reason":"Same source body in authored semantic page"},{"oldId":"b-932569776390aee7fce67c5a","targetId":"b-932569776390aee7fce67c5a","page":"reading-opening","operation":"move","beforeSHA256":"54a440266f441a2ca24062791492c56d30a26070e099032b39ed5da18c6879ea","afterSHA256":"54a440266f441a2ca24062791492c56d30a26070e099032b39ed5da18c6879ea","reason":"Same source body in authored semantic page"},{"oldId":"block-dd1046d546e24b26aecee5957553a6e2--图像前缀目标三个位置分开看","targetId":"b-425258f91a78c5039d475615","operation":"retain-source-identity"},{"oldId":"block-dd1046d546e24b26aecee5957553a6e2--训练与推理的分界发生在前缀","targetId":"b-a454522b892395292d1d1585","operation":"retain-source-identity"},{"oldId":"block-dd1046d546e24b26aecee5957553a6e2--已有实验确实完成了图像条件生成","targetId":"b-709137a6f9d87f1c3ff701c2","operation":"retain-source-identity"}],"blocks":[{"id":"b-357db54056cf2501bd95da51","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-357db54056cf2501bd95da51","content":"[CLS 09 · 图像条件输出：描述与问答](/course/notebooks/c9f328a00ef44ecc90804e47276a7811.html)","sha256":"fa1d05e087449a3f39b93c1c4f3e6086732599f759cb40cfca565caa8afb8709","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-1a26fa22a11915c40c3c5922","prerequisites":[],"relatedBlocks":[]},{"id":"b-1a26fa22a11915c40c3c5922","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-1a26fa22a11915c40c3c5922","content":"# 图像、前缀、目标，三个位置分开看","sha256":"77c37b608fd8aa55ce3223d141440f1b400d22c00bc453aa25f725d0f5e47610","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-357db54056cf2501bd95da51","nextBlock":"b-e4fbe0dc1575bb96929ee494","prerequisites":[{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-e4fbe0dc1575bb96929ee494","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-e4fbe0dc1575bb96929ee494","content":"按问题继续","sha256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-1a26fa22a11915c40c3c5922","nextBlock":"b-7b9b63e4812e8cbb0563586f","prerequisites":[{"id":"b-1a26fa22a11915c40c3c5922","url":"#b-1a26fa22a11915c40c3c5922","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-7b9b63e4812e8cbb0563586f","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-7b9b63e4812e8cbb0563586f","content":"图像条件、teacher forcing、真实生成、greedy/beam与VQA控制","sha256":"a192bf58b03c510999643a276b8d2261f6c9389ea09aebd6ad3419fe70e2a7b9","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-e4fbe0dc1575bb96929ee494","nextBlock":"b-803c5c50d09342ca589335c7","prerequisites":[{"id":"b-1a26fa22a11915c40c3c5922","url":"#b-1a26fa22a11915c40c3c5922","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-803c5c50d09342ca589335c7","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-803c5c50d09342ca589335c7","content":"**学习路径：** [图像怎样约束句子与回答](/course/study/caption-vqa.html)","sha256":"a57513e59eb07960710e5ece7951cc57af4a94c446630eb5fc6d70b460982042","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-7b9b63e4812e8cbb0563586f","nextBlock":"b-e9ff65a3dc6dcc06d33e0065","prerequisites":[{"id":"b-1a26fa22a11915c40c3c5922","url":"#b-1a26fa22a11915c40c3c5922","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-e9ff65a3dc6dcc06d33e0065","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-e9ff65a3dc6dcc06d33e0065","content":"从[P51 看图描述](/course/pages/l3-r2-p51.html)与[P54 视觉问答](/course/pages/l3-r2-p54.html)进入：同一视觉编码器怎样服务“生成一句话”和“针对问题回答”两种输出？","sha256":"6b3101dc74d613fd527b60ff21d1ac65db81cca6fc49d581090c9a90654d2057","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-803c5c50d09342ca589335c7","nextBlock":"b-aed5f706ce45f4ef3e5d9a4d","prerequisites":[{"id":"b-1a26fa22a11915c40c3c5922","url":"#b-1a26fa22a11915c40c3c5922","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-aed5f706ce45f4ef3e5d9a4d","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-aed5f706ce45f4ef3e5d9a4d","content":"先读图像、文字前缀和下一词目标，再看VQA的同图换问/同问换图。按兴趣选择[LAB 13](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/3da47f3ee3554ee0a36e0bb6a2fdc61f)或[LAB 14](https://deepnote.com/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/11d9f74871544a77b42024a9fe06b81a)；已有结果能先读，运行时各册独立从头执行。","sha256":"98588b7069ba7f92211a8e6cd5c8be5f8249ccf2425e964a426643a339f7a008","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-e9ff65a3dc6dcc06d33e0065","nextBlock":"b-269f0352fbe73e55e448d907","prerequisites":[{"id":"b-1a26fa22a11915c40c3c5922","url":"#b-1a26fa22a11915c40c3c5922","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-269f0352fbe73e55e448d907","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-269f0352fbe73e55e448d907","content":"读完应能解释：为什么训练损失不能代替自由生成评价，为什么问题先验不能代替视觉证据。GRU/RNN 内部与 beam 搜索已移入[第五讲](https://deepnote.com/project/841ec8e7-e352-4840-b628-d3742b03d6d0/notebook/b1c8e2fd764441e8820b808fc40ec3cb?utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_content=b1c8e2fd764441e8820b808fc40ec3cb&utm_term=get_notebook)，不是理解视觉条件的前置要求。","sha256":"46d053d1cc569605d484255e6af2c3466f7d40f4fe2b28cec955d4de4a9dd9d8","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-aed5f706ce45f4ef3e5d9a4d","nextBlock":"b-425258f91a78c5039d475615","prerequisites":[{"id":"b-1a26fa22a11915c40c3c5922","url":"#b-1a26fa22a11915c40c3c5922","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-425258f91a78c5039d475615","type":"h2","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-425258f91a78c5039d475615","content":"## 图像、前缀、目标，三个位置分开看","sha256":"5d04e11d8c2e016f93a6fd003c5aafbcfa45431a430ae351461bb69f38e77a4d","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-269f0352fbe73e55e448d907","nextBlock":"b-420d91de3ca262a0fe45bcec","prerequisites":[{"id":"b-1a26fa22a11915c40c3c5922","url":"#b-1a26fa22a11915c40c3c5922","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-420d91de3ca262a0fe45bcec","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-420d91de3ca262a0fe45bcec","content":"令图像为 $I$、句子为 $y_1,\\ldots,y_T$，模型学习\n$p(y_1,\\ldots,y_T\\mid I)=\\prod_t p(y_t\\mid I,y_{<t}).$\n视觉编码器把像素变成特征；文字预测器接收视觉特征和允许看到的文字；目标是下一词。课堂从“红衣人”后预测“拿着”，再从“红衣人拿着”后预测“球”。图上的目标词是人工参考，不是实测生成。","sha256":"621909530fe228f941698b8aa1acd5e19c591377be6a124d71ff93cc33cf1036","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[{"latex":"I","definitionContext":"图像、前缀、目标，三个位置分开看"},{"latex":"y_1,\\ldots,y_T","definitionContext":"图像、前缀、目标，三个位置分开看"},{"latex":"p(y_1,\\ldots,y_T\\mid I)=\\prod_t p(y_t\\mid I,y_{<t}).","definitionContext":"图像、前缀、目标，三个位置分开看"}],"previousBlock":"b-425258f91a78c5039d475615","nextBlock":"b-dc4db5fb67e99fb23621bb7c","prerequisites":[{"id":"b-425258f91a78c5039d475615","url":"#b-425258f91a78c5039d475615","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-dc4db5fb67e99fb23621bb7c","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-dc4db5fb67e99fb23621bb7c","content":"一图可以有“两人在球场”与“红衣人拿球”等不同真实描述。模型学习条件下的表达分布，任务再决定必须包含哪些内容。[完整信息流与多参考解释](/course/blocks/caption-conditioned-token.html)。","sha256":"d3f6a0e9d3a04e431b7a9c6faca644a0357974e570ae3eea66067da85832db98","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-420d91de3ca262a0fe45bcec","nextBlock":"b-a454522b892395292d1d1585","prerequisites":[{"id":"b-425258f91a78c5039d475615","url":"#b-425258f91a78c5039d475615","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-a454522b892395292d1d1585","type":"h2","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-a454522b892395292d1d1585","content":"## 训练与推理的分界发生在前缀","sha256":"132ec47bbd37bd8b6573626070e54b3184a5c5a21294a10bd9ab4c419181acc7","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-dc4db5fb67e99fb23621bb7c","nextBlock":"b-58789586d61178dfcfebbe36","prerequisites":[{"id":"b-425258f91a78c5039d475615","url":"#b-425258f91a78c5039d475615","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-58789586d61178dfcfebbe36","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-58789586d61178dfcfebbe36","content":"参考 `[BOS,a,red,square,EOS]` 的输入是 `[BOS,a,red,square]`，目标是 `[a,red,square,EOS]`。PAD 仅为批量补齐，不计损失；EOS 是有效停止目标。损失按有效词求平均，不能把补齐格数也算入分母。","sha256":"9e28174c8843e3c00738239016bd81d74ee8e3b5857fa1ff02121cc770476f86","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-a454522b892395292d1d1585","nextBlock":"b-c1645412cfb593aa7f61e29f","prerequisites":[{"id":"b-a454522b892395292d1d1585","url":"#b-a454522b892395292d1d1585","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-c1645412cfb593aa7f61e29f","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-c1645412cfb593aa7f61e29f","content":"训练时看到真实前缀叫 teacher forcing；推理必须接续自己的输出。先误说 blue 后，就要在 blue 条件下继续，不能改回参考 red。保留集 teacher-forced 交叉熵因此不能代替自由生成评价。[逐位置表与损失公式](/course/blocks/caption-shift-and-stop.html)。","sha256":"18fc193179c81a0c886864470e3271551a6d8a3537b9fc9d935e5490e732aadb","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-58789586d61178dfcfebbe36","nextBlock":"b-709137a6f9d87f1c3ff701c2","prerequisites":[{"id":"b-a454522b892395292d1d1585","url":"#b-a454522b892395292d1d1585","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-709137a6f9d87f1c3ff701c2","type":"h2","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-709137a6f9d87f1c3ff701c2","content":"## 已有实验确实完成了图像条件生成","sha256":"70da78e6d74f17d276ba01f6377f5f483608d6e486db66f7ffab4af473c8c68d","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c1645412cfb593aa7f61e29f","nextBlock":"b-297c5a9500763b144618f1fa","prerequisites":[{"id":"b-a454522b892395292d1d1585","url":"#b-a454522b892395292d1d1585","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-297c5a9500763b144618f1fa","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-297c5a9500763b144618f1fa","content":"[LAB 13](https://deepnote.com/workspace/ise3309jingao-5ba66b0a-4283-4f5a-ae06-3f68d3201e99/project/38755e4e-23f6-4f5f-8c68-8a2e4114b1c1/notebook/3da47f3ee3554ee0a36e0bb6a2fdc61f?utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_content=3da47f3ee3554ee0a36e0bb6a2fdc61f&utm_term=get_notebook) 保存2026-09-22本地 CPU 基准：720张训练图、180张独立渲染保留图，18种语义组合两侧都出现。CNN产生48维视觉向量，文字模块采用小型GRU；本讲把它作为明确输入输出模块，内部留L05。推理不读取真实颜色/位置标签。","sha256":"6bca41597071d63a25f16e927be8d8382b5da7c125bd9f583e5399e0520cc939","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-709137a6f9d87f1c3ff701c2","nextBlock":"b-0af341b5341eedefc785227d","prerequisites":[{"id":"b-709137a6f9d87f1c3ff701c2","url":"#b-709137a6f9d87f1c3ff701c2","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-0af341b5341eedefc785227d","type":"table","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-0af341b5341eedefc785227d","content":"| 保留条件 | 整句精确匹配 |\n| --- | --- |\n| 正确图像，greedy / beam=3 | 100% / 100% |\n| 同一模型，像素置零 | 5.56% |\n| 同一模型，循环错配图像 | 0% |","sha256":"76596d43d047ab4ae6858c71f16377d45d925bd1b3cb8a695d4f1606a5d9cbc1","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"table":{"headers":["保留条件","整句精确匹配"],"rows":[["保留条件","整句精确匹配"],["正确图像，greedy / beam=3","100% / 100%"],["同一模型，像素置零","5.56%"],["同一模型，循环错配图像","0%"]]},"previousBlock":"b-297c5a9500763b144618f1fa","nextBlock":"b-8e89985e66ebcf261d135e59","prerequisites":[{"id":"b-709137a6f9d87f1c3ff701c2","url":"#b-709137a6f9d87f1c3ff701c2","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-8e89985e66ebcf261d135e59","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-8e89985e66ebcf261d135e59","content":"![同一前缀、不同图像的真实生成](/assets/display/9ebe76d49a449b9940b6470f.webp)","sha256":"5fad46273be0df138b5d8cd83825e637a26e4d2054efd42ac82536606c85cd42","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-0af341b5341eedefc785227d","nextBlock":"b-932569776390aee7fce67c5a","prerequisites":[{"id":"b-709137a6f9d87f1c3ff701c2","url":"#b-709137a6f9d87f1c3ff701c2","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-932569776390aee7fce67c5a","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/dd1046d546e24b26aecee5957553a6e2.html#b-932569776390aee7fce67c5a","content":"固定 `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`。这支持当前有限任务使用正确视觉条件；不是自然照片或未见组合泛化，也没有显示 beam 优于 greedy。本轮整理未重跑训练。[数据、协议与证据细节](/course/blocks/caption-visual-controls.html)。","sha256":"54a440266f441a2ca24062791492c56d30a26070e099032b39ed5da18c6879ea","chapter":{"id":"course","title":"图像、前缀、目标，三个位置分开看"},"unit":{"id":"notebook-blocks/dd1046d546e24b26aecee5957553a6e2","title":"图像、前缀、目标，三个位置分开看"},"page":{"id":"reading-opening","title":"图像、前缀、目标，三个位置分开看","number":1},"sectionTitle":"图像、前缀、目标，三个位置分开看","currentQuestion":"图像、前缀、目标，三个位置分开看","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-8e89985e66ebcf261d135e59","nextBlock":null,"prerequisites":[{"id":"b-709137a6f9d87f1c3ff701c2","url":"#b-709137a6f9d87f1c3ff701c2","role":"section-title"},{"id":"b-357db54056cf2501bd95da51","url":"#b-357db54056cf2501bd95da51","role":"problem-context"}],"relatedBlocks":[]}]}
