{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"组合接口与选读边界","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20.html","markdown":"https://codingai-lec03.pages.dev/course/notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20.md","pages":[{"id":"reading-opening","title":"组合接口与选读边界","unit":"组合接口与选读边界","unitId":"notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20","chapter":"course","number":1}],"aliases":[{"oldId":"b-c63443de68904d79446816c1","targetId":"b-c63443de68904d79446816c1","page":"reading-opening","operation":"move","beforeSHA256":"ca7e12f2ca2547e526317fa1c76ed9ef674c5d7a7eee748c10ae0de4d06ceb2a","afterSHA256":"ca7e12f2ca2547e526317fa1c76ed9ef674c5d7a7eee748c10ae0de4d06ceb2a","reason":"Same source body in authored semantic page"},{"oldId":"b-355e72670d530668e3eb55da","targetId":"b-355e72670d530668e3eb55da","page":"reading-opening","operation":"move","beforeSHA256":"c9e6b5fad386f0e7140e82b4d48ef72f15471b17bacb6a032cd2ea392799c700","afterSHA256":"c9e6b5fad386f0e7140e82b4d48ef72f15471b17bacb6a032cd2ea392799c700","reason":"Same source body in authored semantic page"},{"oldId":"b-dafba67d2c686fc5106451c5","targetId":"b-dafba67d2c686fc5106451c5","page":"reading-opening","operation":"move","beforeSHA256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","afterSHA256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","reason":"Same source body in authored semantic page"},{"oldId":"b-b04c3107f0d004896f22feb7","targetId":"b-b04c3107f0d004896f22feb7","page":"reading-opening","operation":"move","beforeSHA256":"29c03b145408b9ee4ac1eb7996a1fb4e7b0f911ff21fb7cd9f6ac0b8b3b44714","afterSHA256":"29c03b145408b9ee4ac1eb7996a1fb4e7b0f911ff21fb7cd9f6ac0b8b3b44714","reason":"Same source body in authored semantic page"},{"oldId":"b-ad4e814034ea21197a4b4eb6","targetId":"b-ad4e814034ea21197a4b4eb6","page":"reading-opening","operation":"move","beforeSHA256":"ea081a32377fa4e7891b015a053f4c782391d173702b678a690a469e35517814","afterSHA256":"ea081a32377fa4e7891b015a053f4c782391d173702b678a690a469e35517814","reason":"Same source body in authored semantic page"},{"oldId":"b-5f25fe3ffb92f44b47a96779","targetId":"b-5f25fe3ffb92f44b47a96779","page":"reading-opening","operation":"move","beforeSHA256":"5fa92c124d5680ee446ca2019153d88daebd10b674b4494c5556722c5f7076de","afterSHA256":"5fa92c124d5680ee446ca2019153d88daebd10b674b4494c5556722c5f7076de","reason":"Same source body in authored semantic page"},{"oldId":"b-302df97051f1b7256dbc75d9","targetId":"b-302df97051f1b7256dbc75d9","page":"reading-opening","operation":"move","beforeSHA256":"4fb0f211cc5ea6b1f697b08c4e3a4ab0675c3aa6420552c28873e1c9b47164ed","afterSHA256":"4fb0f211cc5ea6b1f697b08c4e3a4ab0675c3aa6420552c28873e1c9b47164ed","reason":"Same source body in authored semantic page"},{"oldId":"b-aad1562cccc20fff3fe17861","targetId":"b-aad1562cccc20fff3fe17861","page":"reading-opening","operation":"move","beforeSHA256":"611142a0bb5d0a7d4cf516bc65f90ef3dc4787c6b456d246bb1b2683bfce1556","afterSHA256":"611142a0bb5d0a7d4cf516bc65f90ef3dc4787c6b456d246bb1b2683bfce1556","reason":"Same source body in authored semantic page"},{"oldId":"b-163c881802cd6b4c300d2d38","targetId":"b-163c881802cd6b4c300d2d38","page":"reading-opening","operation":"move","beforeSHA256":"6bd6829cbb9417b5e60425f8792e5d6b4dd1d1e3ca7aa225e6e61fe3b920ebfd","afterSHA256":"6bd6829cbb9417b5e60425f8792e5d6b4dd1d1e3ca7aa225e6e61fe3b920ebfd","reason":"Same source body in authored semantic page"},{"oldId":"block-30c185dacfcf4fb59e6a8b9c3ae7be20--组合接口与选读边界","targetId":"b-ad4e814034ea21197a4b4eb6","operation":"retain-source-identity"}],"blocks":[{"id":"b-c63443de68904d79446816c1","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20.html#b-c63443de68904d79446816c1","content":"[CLS 11 · 语言定位与视觉模型接口](/course/notebooks/95169c3e298c4c819373bb4534f3dff9.html)","sha256":"ca7e12f2ca2547e526317fa1c76ed9ef674c5d7a7eee748c10ae0de4d06ceb2a","chapter":{"id":"course","title":"组合接口与选读边界"},"unit":{"id":"notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20","title":"组合接口与选读边界"},"page":{"id":"reading-opening","title":"组合接口与选读边界","number":1},"sectionTitle":"组合接口与选读边界","currentQuestion":"组合接口与选读边界","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-355e72670d530668e3eb55da","prerequisites":[],"relatedBlocks":[]},{"id":"b-355e72670d530668e3eb55da","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20.html#b-355e72670d530668e3eb55da","content":"# 组合接口与选读边界","sha256":"c9e6b5fad386f0e7140e82b4d48ef72f15471b17bacb6a032cd2ea392799c700","chapter":{"id":"course","title":"组合接口与选读边界"},"unit":{"id":"notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20","title":"组合接口与选读边界"},"page":{"id":"reading-opening","title":"组合接口与选读边界","number":1},"sectionTitle":"组合接口与选读边界","currentQuestion":"组合接口与选读边界","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c63443de68904d79446816c1","nextBlock":"b-dafba67d2c686fc5106451c5","prerequisites":[{"id":"b-c63443de68904d79446816c1","url":"#b-c63443de68904d79446816c1","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-dafba67d2c686fc5106451c5","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20.html#b-dafba67d2c686fc5106451c5","content":"按问题继续","sha256":"e91f24e4e4ca40e2099a0d81d30db6568f4bf2074b499eeee6e3959617bd97ac","chapter":{"id":"course","title":"组合接口与选读边界"},"unit":{"id":"notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20","title":"组合接口与选读边界"},"page":{"id":"reading-opening","title":"组合接口与选读边界","number":1},"sectionTitle":"组合接口与选读边界","currentQuestion":"组合接口与选读边界","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-355e72670d530668e3eb55da","nextBlock":"b-b04c3107f0d004896f22feb7","prerequisites":[{"id":"b-355e72670d530668e3eb55da","url":"#b-355e72670d530668e3eb55da","role":"section-title"},{"id":"b-c63443de68904d79446816c1","url":"#b-c63443de68904d79446816c1","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-b04c3107f0d004896f22feb7","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20.html#b-b04c3107f0d004896f22feb7","content":"目标/参照/关系、受控图像干预、短语区域学习、分割提示及连接器","sha256":"29c03b145408b9ee4ac1eb7996a1fb4e7b0f911ff21fb7cd9f6ac0b8b3b44714","chapter":{"id":"course","title":"组合接口与选读边界"},"unit":{"id":"notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20","title":"组合接口与选读边界"},"page":{"id":"reading-opening","title":"组合接口与选读边界","number":1},"sectionTitle":"组合接口与选读边界","currentQuestion":"组合接口与选读边界","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-dafba67d2c686fc5106451c5","nextBlock":"b-ad4e814034ea21197a4b4eb6","prerequisites":[{"id":"b-355e72670d530668e3eb55da","url":"#b-355e72670d530668e3eb55da","role":"section-title"},{"id":"b-c63443de68904d79446816c1","url":"#b-c63443de68904d79446816c1","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-ad4e814034ea21197a4b4eb6","type":"h2","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20.html#b-ad4e814034ea21197a4b4eb6","content":"## 组合接口与选读边界","sha256":"ea081a32377fa4e7891b015a053f4c782391d173702b678a690a469e35517814","chapter":{"id":"course","title":"组合接口与选读边界"},"unit":{"id":"notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20","title":"组合接口与选读边界"},"page":{"id":"reading-opening","title":"组合接口与选读边界","number":1},"sectionTitle":"组合接口与选读边界","currentQuestion":"组合接口与选读边界","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-b04c3107f0d004896f22feb7","nextBlock":"b-5f25fe3ffb92f44b47a96779","prerequisites":[{"id":"b-355e72670d530668e3eb55da","url":"#b-355e72670d530668e3eb55da","role":"section-title"},{"id":"b-c63443de68904d79446816c1","url":"#b-c63443de68904d79446816c1","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-5f25fe3ffb92f44b47a96779","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20.html#b-5f25fe3ffb92f44b47a96779","content":"语言定位可以先产出框，再将框交给 SAM 一类提示式分割接口。若语言阶段把参照杯当成目标，后续 mask 可能边界很准但对象错误；若框身份正确而边界差，应检查分割与分辨率。由此把错误沿接口定位，见 [提示式分割](/course/pages/l3-r2-p47.html) 和 [语言定位到局部输出](/course/blocks/glip-language-objective.html)。","sha256":"5fa92c124d5680ee446ca2019153d88daebd10b674b4494c5556722c5f7076de","chapter":{"id":"course","title":"组合接口与选读边界"},"unit":{"id":"notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20","title":"组合接口与选读边界"},"page":{"id":"reading-opening","title":"组合接口与选读边界","number":1},"sectionTitle":"组合接口与选读边界","currentQuestion":"组合接口与选读边界","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-ad4e814034ea21197a4b4eb6","nextBlock":"b-302df97051f1b7256dbc75d9","prerequisites":[{"id":"b-ad4e814034ea21197a4b4eb6","url":"#b-ad4e814034ea21197a4b4eb6","role":"section-title"},{"id":"b-c63443de68904d79446816c1","url":"#b-c63443de68904d79446816c1","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-302df97051f1b7256dbc75d9","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20.html#b-302df97051f1b7256dbc75d9","content":"**第五讲预备阅读。** BLIP-2 的 Q-Former、Flamingo 的视觉重采样和跨模态交互，提供与原始 LLaVA 线性投影不同的信息选择方式。它们留作接口比较，不是第三讲必须提前完成的 attention 专题。","sha256":"4fb0f211cc5ea6b1f697b08c4e3a4ab0675c3aa6420552c28873e1c9b47164ed","chapter":{"id":"course","title":"组合接口与选读边界"},"unit":{"id":"notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20","title":"组合接口与选读边界"},"page":{"id":"reading-opening","title":"组合接口与选读边界","number":1},"sectionTitle":"组合接口与选读边界","currentQuestion":"组合接口与选读边界","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-5f25fe3ffb92f44b47a96779","nextBlock":"b-aad1562cccc20fff3fe17861","prerequisites":[{"id":"b-ad4e814034ea21197a4b4eb6","url":"#b-ad4e814034ea21197a4b4eb6","role":"section-title"},{"id":"b-c63443de68904d79446816c1","url":"#b-c63443de68904d79446816c1","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-aad1562cccc20fff3fe17861","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20.html#b-aad1562cccc20fff3fe17861","content":"压缩视觉向量可降低后续计算，但小字或小物体可能被丢失；提高分辨率也带来成本。判断时用同图换问、遮挡目标和关系交换等具体对照，而不把流畅回答本身当作细节保留的证明。","sha256":"611142a0bb5d0a7d4cf516bc65f90ef3dc4787c6b456d246bb1b2683bfce1556","chapter":{"id":"course","title":"组合接口与选读边界"},"unit":{"id":"notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20","title":"组合接口与选读边界"},"page":{"id":"reading-opening","title":"组合接口与选读边界","number":1},"sectionTitle":"组合接口与选读边界","currentQuestion":"组合接口与选读边界","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-302df97051f1b7256dbc75d9","nextBlock":"b-163c881802cd6b4c300d2d38","prerequisites":[{"id":"b-ad4e814034ea21197a4b4eb6","url":"#b-ad4e814034ea21197a4b4eb6","role":"section-title"},{"id":"b-c63443de68904d79446816c1","url":"#b-c63443de68904d79446816c1","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-163c881802cd6b4c300d2d38","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20.html#b-163c881802cd6b4c300d2d38","content":"[BLIP-2 原论文](https://arxiv.org/abs/2301.12597) · [Flamingo 原论文](https://arxiv.org/abs/2204.14198)。回到 [三条主线的组件复用](/course/pages/l3-r2-p66.html)，然后进入 [第四讲 · 从条件多解走向生成模型](https://deepnote.com/project/6f83a923-d155-47be-be58-680db701ff7f/notebook/694b7a43ea4948f5a197c014c2fbad3b?utm_source=openai&utm_medium=mcp&utm_campaign=openaimcp&utm_content=694b7a43ea4948f5a197c014c2fbad3b&utm_term=get_notebook)。","sha256":"6bd6829cbb9417b5e60425f8792e5d6b4dd1d1e3ca7aa225e6e61fe3b920ebfd","chapter":{"id":"course","title":"组合接口与选读边界"},"unit":{"id":"notebook-blocks/30c185dacfcf4fb59e6a8b9c3ae7be20","title":"组合接口与选读边界"},"page":{"id":"reading-opening","title":"组合接口与选读边界","number":1},"sectionTitle":"组合接口与选读边界","currentQuestion":"组合接口与选读边界","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-aad1562cccc20fff3fe17861","nextBlock":null,"prerequisites":[{"id":"b-ad4e814034ea21197a4b4eb6","url":"#b-ad4e814034ea21197a4b4eb6","role":"section-title"},{"id":"b-c63443de68904d79446816c1","url":"#b-c63443de68904d79446816c1","role":"problem-context"}],"relatedBlocks":[]}]}