{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/blocks/pixel-model-positive-result.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"同一组分割实验里，简单模型为何值得留下","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-model-positive-result.html","markdown":"https://codingai-lec03.pages.dev/course/blocks/pixel-model-positive-result.md","pages":[{"id":"reading-opening","title":"同一组分割实验里，简单模型为何值得留下","unit":"哪些信息真的改善了像素预测？","unitId":"pixel-detail-and-baselines","chapter":"II","number":1}],"aliases":[{"oldId":"b-5ed00934a33ed69da84fac93","targetId":"b-5ed00934a33ed69da84fac93","page":"reading-opening","operation":"move","beforeSHA256":"394a75d7c0ce822f695abd8560266ed0b5abca06938e62a73ec057709c05ffd5","afterSHA256":"394a75d7c0ce822f695abd8560266ed0b5abca06938e62a73ec057709c05ffd5","reason":"Same source body in authored semantic page"},{"oldId":"b-c2e03686d872b12626737921","targetId":"b-c2e03686d872b12626737921","page":"reading-opening","operation":"move","beforeSHA256":"183d41f9f2dd4315d6651bd668ce69b215479fec798b3e287bba7d446b7f4a31","afterSHA256":"183d41f9f2dd4315d6651bd668ce69b215479fec798b3e287bba7d446b7f4a31","reason":"Same source body in authored semantic page"},{"oldId":"b-4d43435f5cea2d785f1097e7","targetId":"b-4d43435f5cea2d785f1097e7","page":"reading-opening","operation":"move","beforeSHA256":"6c0114296cc35b10699175b4c89039f6f9876e3185b83a7f02e7fa67ae9448e9","afterSHA256":"6c0114296cc35b10699175b4c89039f6f9876e3185b83a7f02e7fa67ae9448e9","reason":"Same source body in authored semantic page"},{"oldId":"b-51c7159382538bc2feca63f1","targetId":"b-51c7159382538bc2feca63f1","page":"reading-opening","operation":"move","beforeSHA256":"01b07d138cf5170bb6f10b5ab91b026d706baa6e021a922543b13d85cf3271ac","afterSHA256":"01b07d138cf5170bb6f10b5ab91b026d706baa6e021a922543b13d85cf3271ac","reason":"Same source body in authored semantic page"},{"oldId":"b-c06c47f8c5a499a0ecb659b4","targetId":"b-c06c47f8c5a499a0ecb659b4","page":"reading-opening","operation":"move","beforeSHA256":"53b208f148eb09e2dac3c533a2268e625a9dd513bee413b9a41b925e40631934","afterSHA256":"53b208f148eb09e2dac3c533a2268e625a9dd513bee413b9a41b925e40631934","reason":"Same source body in authored semantic page"},{"oldId":"b-5b3aa4d3d0daf56b2dfd3ba8","targetId":"b-5b3aa4d3d0daf56b2dfd3ba8","page":"reading-opening","operation":"move","beforeSHA256":"c31efb764a8e17b83caed97f41768f2bcebbb1a1964a23bcc8cbf6462d1b09a0","afterSHA256":"c31efb764a8e17b83caed97f41768f2bcebbb1a1964a23bcc8cbf6462d1b09a0","reason":"Same source body in authored semantic page"},{"oldId":"b-c8d48cc6cd71912e5e458056","targetId":"b-c8d48cc6cd71912e5e458056","page":"reading-opening","operation":"move","beforeSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","afterSHA256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","reason":"Same source body in authored semantic page"}],"blocks":[{"id":"b-5ed00934a33ed69da84fac93","type":"h1","scope":"pixel-model-positive-result","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-model-positive-result.html#b-5ed00934a33ed69da84fac93","content":"# 同一组分割实验里，简单模型为何值得留下","sha256":"394a75d7c0ce822f695abd8560266ed0b5abca06938e62a73ec057709c05ffd5","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"pixel-detail-and-baselines","title":"哪些信息真的改善了像素预测？"},"page":{"id":"reading-opening","title":"同一组分割实验里，简单模型为何值得留下","number":1},"sectionTitle":"同一组分割实验里，简单模型为何值得留下","currentQuestion":"同一组分割实验里，简单模型为何值得留下","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-c2e03686d872b12626737921","prerequisites":[{"id":"b-c2e03686d872b12626737921","url":"#b-c2e03686d872b12626737921","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-c2e03686d872b12626737921","type":"p","scope":"pixel-model-positive-result","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-model-positive-result.html#b-c2e03686d872b12626737921","content":"FCN 从逐像素类别图出发，独立于区域检测路线。将 `[B,C,H,W]` logits 与 `[B,H,W]` 标签一一对应，理解分割损失和 ignore 像素。LAB11 从图像预测 mask；已知 mask 的缩放练习只检验采样，不检验视觉预测。","sha256":"183d41f9f2dd4315d6651bd668ce69b215479fec798b3e287bba7d446b7f4a31","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"pixel-detail-and-baselines","title":"哪些信息真的改善了像素预测？"},"page":{"id":"reading-opening","title":"同一组分割实验里，简单模型为何值得留下","number":1},"sectionTitle":"同一组分割实验里，简单模型为何值得留下","currentQuestion":"同一组分割实验里，简单模型为何值得留下","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-5ed00934a33ed69da84fac93","nextBlock":"b-4d43435f5cea2d785f1097e7","prerequisites":[{"id":"b-5ed00934a33ed69da84fac93","url":"#b-5ed00934a33ed69da84fac93","role":"section-title"}],"relatedBlocks":[]},{"id":"b-4d43435f5cea2d785f1097e7","type":"p","scope":"pixel-model-positive-result","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-model-positive-result.html#b-4d43435f5cea2d785f1097e7","content":"当前 LAB11 同时训练逐像素 RGB 线性读出和小型 FCN。输入 48×48 RGB，标签是背景、红矩形、蓝椭圆；384 张训练图与 128 张保留图按场景种子分开。FCN 用粗分数与细分支融合，RGB 基线则只用 `Conv2d(3,3,1)`，共 12 个参数，不能看邻域。","sha256":"6c0114296cc35b10699175b4c89039f6f9876e3185b83a7f02e7fa67ae9448e9","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"pixel-detail-and-baselines","title":"哪些信息真的改善了像素预测？"},"page":{"id":"reading-opening","title":"同一组分割实验里，简单模型为何值得留下","number":1},"sectionTitle":"同一组分割实验里，简单模型为何值得留下","currentQuestion":"同一组分割实验里，简单模型为何值得留下","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c2e03686d872b12626737921","nextBlock":"b-51c7159382538bc2feca63f1","prerequisites":[{"id":"b-5ed00934a33ed69da84fac93","url":"#b-5ed00934a33ed69da84fac93","role":"section-title"},{"id":"b-c2e03686d872b12626737921","url":"#b-c2e03686d872b12626737921","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-51c7159382538bc2feca63f1","type":"p","scope":"pixel-model-positive-result","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-model-positive-result.html#b-51c7159382538bc2feca63f1","content":"已有 2026-09-22 本地完整运行中，原始合成图的 RGB 基线三类 mIoU=1，FCN=.99041；对输入做 5×5 模糊后，二者变为 .84996 与 .92397。原分布上简单模型胜出说明颜色已足够决定类别；模糊使边缘颜色混合，FCN 利用邻域在该干预下更稳健。两项观察共同说明为什么需要基线：它既暴露任务太容易的地方，也帮助找出复杂结构实际增加能力的条件。","sha256":"01b07d138cf5170bb6f10b5ab91b026d706baa6e021a922543b13d85cf3271ac","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"pixel-detail-and-baselines","title":"哪些信息真的改善了像素预测？"},"page":{"id":"reading-opening","title":"同一组分割实验里，简单模型为何值得留下","number":1},"sectionTitle":"同一组分割实验里，简单模型为何值得留下","currentQuestion":"同一组分割实验里，简单模型为何值得留下","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-4d43435f5cea2d785f1097e7","nextBlock":"b-c06c47f8c5a499a0ecb659b4","prerequisites":[{"id":"b-5ed00934a33ed69da84fac93","url":"#b-5ed00934a33ed69da84fac93","role":"section-title"},{"id":"b-c2e03686d872b12626737921","url":"#b-c2e03686d872b12626737921","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-c06c47f8c5a499a0ecb659b4","type":"p","scope":"pixel-model-positive-result","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-model-positive-result.html#b-c06c47f8c5a499a0ecb659b4","content":"同一 FCN 参数，推理时仅把双线性改为最近邻，mIoU 变成 .91180。这检验训练/推理采样约定的敏感性，没有分别重训两种网络，不能当成插值方式的普遍排名。","sha256":"53b208f148eb09e2dac3c533a2268e625a9dd513bee413b9a41b925e40631934","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"pixel-detail-and-baselines","title":"哪些信息真的改善了像素预测？"},"page":{"id":"reading-opening","title":"同一组分割实验里，简单模型为何值得留下","number":1},"sectionTitle":"同一组分割实验里，简单模型为何值得留下","currentQuestion":"同一组分割实验里，简单模型为何值得留下","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-51c7159382538bc2feca63f1","nextBlock":"b-5b3aa4d3d0daf56b2dfd3ba8","prerequisites":[{"id":"b-5ed00934a33ed69da84fac93","url":"#b-5ed00934a33ed69da84fac93","role":"section-title"},{"id":"b-c2e03686d872b12626737921","url":"#b-c2e03686d872b12626737921","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-5b3aa4d3d0daf56b2dfd3ba8","type":"p","scope":"pixel-model-positive-result","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-model-positive-result.html#b-5b3aa4d3d0daf56b2dfd3ba8","content":"本轮整理只复用已有记录，没有重跑。完整学习证据在 RGB→模型→logits→mask 链条；P37 的已知 mask 缩放热身则解释采样，二者都保留，作用分清。","sha256":"c31efb764a8e17b83caed97f41768f2bcebbb1a1964a23bcc8cbf6462d1b09a0","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"pixel-detail-and-baselines","title":"哪些信息真的改善了像素预测？"},"page":{"id":"reading-opening","title":"同一组分割实验里，简单模型为何值得留下","number":1},"sectionTitle":"同一组分割实验里，简单模型为何值得留下","currentQuestion":"同一组分割实验里，简单模型为何值得留下","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c06c47f8c5a499a0ecb659b4","nextBlock":"b-c8d48cc6cd71912e5e458056","prerequisites":[{"id":"b-5ed00934a33ed69da84fac93","url":"#b-5ed00934a33ed69da84fac93","role":"section-title"},{"id":"b-c2e03686d872b12626737921","url":"#b-c2e03686d872b12626737921","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-c8d48cc6cd71912e5e458056","type":"h2","scope":"pixel-model-positive-result","url":"https://codingai-lec03.pages.dev/course/blocks/pixel-model-positive-result.html#b-c8d48cc6cd71912e5e458056","content":"## 继续阅读与实验","sha256":"2b9ccef7e1220e2250cceff1abd8bdd736b42f8da5a223f26b029f44b962e807","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"pixel-detail-and-baselines","title":"哪些信息真的改善了像素预测？"},"page":{"id":"reading-opening","title":"同一组分割实验里，简单模型为何值得留下","number":1},"sectionTitle":"同一组分割实验里，简单模型为何值得留下","currentQuestion":"同一组分割实验里，简单模型为何值得留下","sourceSlides":[],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-5b3aa4d3d0daf56b2dfd3ba8","nextBlock":null,"prerequisites":[{"id":"b-5ed00934a33ed69da84fac93","url":"#b-5ed00934a33ed69da84fac93","role":"section-title"},{"id":"b-c2e03686d872b12626737921","url":"#b-c2e03686d872b12626737921","role":"problem-context"}],"relatedBlocks":[]}]}