{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"从共享特征到两个输出头","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87.html","markdown":"https://codingai-lec03.pages.dev/course/notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87.md","pages":[{"id":"reading-opening","title":"从共享特征到两个输出头","unit":"从共享特征到两个输出头","unitId":"notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87","chapter":"course","number":1}],"aliases":[{"oldId":"b-2ea704d75709621903e9b873","targetId":"b-2ea704d75709621903e9b873","page":"reading-opening","operation":"move","beforeSHA256":"68ab73f0c1cbd872efb5dd565e57f07f1b4a2b56c5ea0ce2c6ed91ff60ff5a3f","afterSHA256":"68ab73f0c1cbd872efb5dd565e57f07f1b4a2b56c5ea0ce2c6ed91ff60ff5a3f","reason":"Same source body in authored semantic page"},{"oldId":"b-2f18a08e8a51e7c177ed5311","targetId":"b-2f18a08e8a51e7c177ed5311","page":"reading-opening","operation":"move","beforeSHA256":"85fd92cee7e79ec7c8b244e974ae5b9ff253f1c0e07c531b0aa878fc1ae59a82","afterSHA256":"85fd92cee7e79ec7c8b244e974ae5b9ff253f1c0e07c531b0aa878fc1ae59a82","reason":"Same source body in authored semantic page"},{"oldId":"b-6a1ac4a37e65098d6c157d7d","targetId":"b-6a1ac4a37e65098d6c157d7d","page":"reading-opening","operation":"move","beforeSHA256":"42cba31deccba2a350b9b23bd0ca1dcaa4c4c5cb9d35c773e1fcb89124788ccd","afterSHA256":"42cba31deccba2a350b9b23bd0ca1dcaa4c4c5cb9d35c773e1fcb89124788ccd","reason":"Same source body in authored semantic page"},{"oldId":"b-55d963655a22ecd6c8b3e249","targetId":"b-55d963655a22ecd6c8b3e249","page":"reading-opening","operation":"move","beforeSHA256":"b1c765d7bf903321abd36e75bae49048d61faec2aa3fd11922b328c5970003b1","afterSHA256":"b1c765d7bf903321abd36e75bae49048d61faec2aa3fd11922b328c5970003b1","reason":"Same source body in authored semantic page"},{"oldId":"b-60f654dad61dadd2dfdd3ef9","targetId":"b-60f654dad61dadd2dfdd3ef9","page":"reading-opening","operation":"move","beforeSHA256":"ae6109b950401193f5256036ea197169997c9abf6e09018367d63b1b3b9697af","afterSHA256":"ae6109b950401193f5256036ea197169997c9abf6e09018367d63b1b3b9697af","reason":"Same source body in authored semantic page"},{"oldId":"b-4b3902c1cb5152c0fc73f848","targetId":"b-4b3902c1cb5152c0fc73f848","page":"reading-opening","operation":"move","beforeSHA256":"6f1f64585357fa0d50beee983ee99c16e80bdf6eff58262950001f171ab07a13","afterSHA256":"6f1f64585357fa0d50beee983ee99c16e80bdf6eff58262950001f171ab07a13","reason":"Same source body in authored semantic page"},{"oldId":"b-58b9deda9d9d2c4930edd1f4","targetId":"b-58b9deda9d9d2c4930edd1f4","page":"reading-opening","operation":"move","beforeSHA256":"a8de3244286e583a34410f672ae0a8773ec99c72b4b883c1f41fab942e792874","afterSHA256":"a8de3244286e583a34410f672ae0a8773ec99c72b4b883c1f41fab942e792874","reason":"Same source body in authored semantic page"},{"oldId":"block-950aab9e6fbf43b4aef8ee21b9268a87--从共享特征到两个输出头","targetId":"b-60f654dad61dadd2dfdd3ef9","operation":"retain-source-identity"}],"blocks":[{"id":"b-2ea704d75709621903e9b873","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87.html#b-2ea704d75709621903e9b873","content":"[LAB 10 · 从 RGB 图像训练小型检测器](/course/notebooks/a14237bfa9dd413d959ab4e458e11f42.html)","sha256":"68ab73f0c1cbd872efb5dd565e57f07f1b4a2b56c5ea0ce2c6ed91ff60ff5a3f","chapter":{"id":"course","title":"从共享特征到两个输出头"},"unit":{"id":"notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87","title":"从共享特征到两个输出头"},"page":{"id":"reading-opening","title":"从共享特征到两个输出头","number":1},"sectionTitle":"从共享特征到两个输出头","currentQuestion":"从共享特征到两个输出头","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-2f18a08e8a51e7c177ed5311","prerequisites":[],"relatedBlocks":[]},{"id":"b-2f18a08e8a51e7c177ed5311","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87.html#b-2f18a08e8a51e7c177ed5311","content":"# 从共享特征到两个输出头","sha256":"85fd92cee7e79ec7c8b244e974ae5b9ff253f1c0e07c531b0aa878fc1ae59a82","chapter":{"id":"course","title":"从共享特征到两个输出头"},"unit":{"id":"notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87","title":"从共享特征到两个输出头"},"page":{"id":"reading-opening","title":"从共享特征到两个输出头","number":1},"sectionTitle":"从共享特征到两个输出头","currentQuestion":"从共享特征到两个输出头","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-2ea704d75709621903e9b873","nextBlock":"b-6a1ac4a37e65098d6c157d7d","prerequisites":[{"id":"b-2ea704d75709621903e9b873","url":"#b-2ea704d75709621903e9b873","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-6a1ac4a37e65098d6c157d7d","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87.html#b-6a1ac4a37e65098d6c157d7d","content":"主线阅读与优先实验","sha256":"42cba31deccba2a350b9b23bd0ca1dcaa4c4c5cb9d35c773e1fcb89124788ccd","chapter":{"id":"course","title":"从共享特征到两个输出头"},"unit":{"id":"notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87","title":"从共享特征到两个输出头"},"page":{"id":"reading-opening","title":"从共享特征到两个输出头","number":1},"sectionTitle":"从共享特征到两个输出头","currentQuestion":"从共享特征到两个输出头","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-2f18a08e8a51e7c177ed5311","nextBlock":"b-55d963655a22ecd6c8b3e249","prerequisites":[{"id":"b-2f18a08e8a51e7c177ed5311","url":"#b-2f18a08e8a51e7c177ed5311","role":"section-title"},{"id":"b-2ea704d75709621903e9b873","url":"#b-2ea704d75709621903e9b873","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-55d963655a22ecd6c8b3e249","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87.html#b-55d963655a22ecd6c8b3e249","content":"RGB图像、单特征层锚框检测器、分类/位置损失、NMS、独立检测评价","sha256":"b1c765d7bf903321abd36e75bae49048d61faec2aa3fd11922b328c5970003b1","chapter":{"id":"course","title":"从共享特征到两个输出头"},"unit":{"id":"notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87","title":"从共享特征到两个输出头"},"page":{"id":"reading-opening","title":"从共享特征到两个输出头","number":1},"sectionTitle":"从共享特征到两个输出头","currentQuestion":"从共享特征到两个输出头","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-6a1ac4a37e65098d6c157d7d","nextBlock":"b-60f654dad61dadd2dfdd3ef9","prerequisites":[{"id":"b-2f18a08e8a51e7c177ed5311","url":"#b-2f18a08e8a51e7c177ed5311","role":"section-title"},{"id":"b-2ea704d75709621903e9b873","url":"#b-2ea704d75709621903e9b873","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-60f654dad61dadd2dfdd3ef9","type":"h2","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87.html#b-60f654dad61dadd2dfdd3ef9","content":"## 从共享特征到两个输出头","sha256":"ae6109b950401193f5256036ea197169997c9abf6e09018367d63b1b3b9697af","chapter":{"id":"course","title":"从共享特征到两个输出头"},"unit":{"id":"notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87","title":"从共享特征到两个输出头"},"page":{"id":"reading-opening","title":"从共享特征到两个输出头","number":1},"sectionTitle":"从共享特征到两个输出头","currentQuestion":"从共享特征到两个输出头","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-55d963655a22ecd6c8b3e249","nextBlock":"b-4b3902c1cb5152c0fc73f848","prerequisites":[{"id":"b-2f18a08e8a51e7c177ed5311","url":"#b-2f18a08e8a51e7c177ed5311","role":"section-title"},{"id":"b-2ea704d75709621903e9b873","url":"#b-2ea704d75709621903e9b873","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-4b3902c1cb5152c0fc73f848","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87.html#b-4b3902c1cb5152c0fc73f848","content":"四个卷积层形成 45,806 参数的模型。分类 logits 为 `[B,288,3]`，框偏移为 `[B,288,4]`。两个头共享空间特征，分类头的“有目标”分数和框头的“偏移多少”要同时学好；分类分数接近 1 并不保证坐标足够准确。","sha256":"6f1f64585357fa0d50beee983ee99c16e80bdf6eff58262950001f171ab07a13","chapter":{"id":"course","title":"从共享特征到两个输出头"},"unit":{"id":"notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87","title":"从共享特征到两个输出头"},"page":{"id":"reading-opening","title":"从共享特征到两个输出头","number":1},"sectionTitle":"从共享特征到两个输出头","currentQuestion":"从共享特征到两个输出头","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-60f654dad61dadd2dfdd3ef9","nextBlock":"b-58b9deda9d9d2c4930edd1f4","prerequisites":[{"id":"b-60f654dad61dadd2dfdd3ef9","url":"#b-60f654dad61dadd2dfdd3ef9","role":"section-title"},{"id":"b-2ea704d75709621903e9b873","url":"#b-2ea704d75709621903e9b873","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-58b9deda9d9d2c4930edd1f4","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87.html#b-58b9deda9d9d2c4930edd1f4","content":"推理只接收 RGB，解码所有候选，对每一前景类别分别做 NMS，默认抑制 IoU>.35 的低分框。保存候选的分数下限为 .01，绘图与 precision/recall 的工作点为 .4。候选阈值、NMS 的 IoU 阈值、评价的 IoU 阈值是三个不同控制量。","sha256":"a8de3244286e583a34410f672ae0a8773ec99c72b4b883c1f41fab942e792874","chapter":{"id":"course","title":"从共享特征到两个输出头"},"unit":{"id":"notebook-blocks/950aab9e6fbf43b4aef8ee21b9268a87","title":"从共享特征到两个输出头"},"page":{"id":"reading-opening","title":"从共享特征到两个输出头","number":1},"sectionTitle":"从共享特征到两个输出头","currentQuestion":"从共享特征到两个输出头","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-4b3902c1cb5152c0fc73f848","nextBlock":null,"prerequisites":[{"id":"b-60f654dad61dadd2dfdd3ef9","url":"#b-60f654dad61dadd2dfdd3ef9","role":"section-title"},{"id":"b-2ea704d75709621903e9b873","url":"#b-2ea704d75709621903e9b873","role":"problem-context"}],"relatedBlocks":[]}]}