{"version":2,"relationshipURLBase":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html","relationshipResolution":"Resolve relative relationship URLs against relationshipURLBase; complete records and absolute URLs are in blocks.","title":"不同大小的区域，怎样读成固定结构？","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html","markdown":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.md","pages":[{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","unit":"多个区域怎样复用同一份视觉计算？","unitId":"region-computation-reuse","chapter":"II","number":1}],"aliases":[{"oldId":"explanation","targetId":"region-fixed-grid-readout","operation":"split-explanation-range"},{"oldId":"b-3af39a1baae77e2a18591dde","targetId":"b-3af39a1baae77e2a18591dde","page":"region-fixed-grid-readout","operation":"move","beforeSHA256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","afterSHA256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","reason":"Same source body in authored semantic page"},{"oldId":"b-ce5f95165e174c468bccd39a","targetId":"b-ce5f95165e174c468bccd39a","page":"region-fixed-grid-readout","operation":"move","beforeSHA256":"6837be3319b8248054aa18e902f639a5b4ad94cf795a5dc23d6a74dc6349e233","afterSHA256":"6837be3319b8248054aa18e902f639a5b4ad94cf795a5dc23d6a74dc6349e233","reason":"Same source body in authored semantic page"},{"oldId":"b-ed5d3f5cc0f93253e8804706","targetId":"b-ed5d3f5cc0f93253e8804706","page":"region-fixed-grid-readout","operation":"move","beforeSHA256":"d3c2aff40d77f95573a8243ba5fffda912e8d3ed195b0fd849eadf14161c476d","afterSHA256":"d3c2aff40d77f95573a8243ba5fffda912e8d3ed195b0fd849eadf14161c476d","reason":"Same source body in authored semantic page"},{"oldId":"b-8253e0bd3b541e777b1ab311","targetId":"b-8253e0bd3b541e777b1ab311","page":"region-fixed-grid-readout","operation":"rewrite","beforeSHA256":"7f2dec9f24fc128c9097a8aa40286fb55def820726943e8d921871061cbf224e","afterSHA256":"e34cbfc483d6da2d5ef34c9b5ab64c2e89edec7509490f7ce842d385743c80a9","reason":"Same source identity with figure semantics or reading context enriched"},{"oldId":"b-efa6d45fcb21f08c9d188de0","targetId":"b-efa6d45fcb21f08c9d188de0","page":"region-fixed-grid-readout","operation":"move","beforeSHA256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","afterSHA256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","reason":"Same source body in authored semantic page"},{"oldId":"b-70b15c2345a27eb53c64c6bc","targetId":"b-70b15c2345a27eb53c64c6bc","page":"region-fixed-grid-readout","operation":"move","beforeSHA256":"842e062ba99099f4345595d783aaa3aad625e7e51d1060b34fec0079999c62b4","afterSHA256":"842e062ba99099f4345595d783aaa3aad625e7e51d1060b34fec0079999c62b4","reason":"Same source body in authored semantic page"},{"oldId":"b-9dfec0da8b2cf1467a8177cc","targetId":"b-9dfec0da8b2cf1467a8177cc","page":"region-fixed-grid-readout","operation":"move","beforeSHA256":"458598c561a796cf6c0884e284ec9f9d5b2933fcbdc20a43cd405fda3c19705f","afterSHA256":"458598c561a796cf6c0884e284ec9f9d5b2933fcbdc20a43cd405fda3c19705f","reason":"Same source body in authored semantic page"},{"oldId":"b-aacf3c05c2a7d1e3cd061a9f","targetId":"b-aacf3c05c2a7d1e3cd061a9f","page":"region-fixed-grid-readout","operation":"move","beforeSHA256":"cfc72e17de8c18086f6d7e26fa4eb9839f8d2a474a2ed8be9a0f522c9b642939","afterSHA256":"cfc72e17de8c18086f6d7e26fa4eb9839f8d2a474a2ed8be9a0f522c9b642939","reason":"Same source body in authored semantic page"},{"oldId":"b-e3ee2d9ac657df56f5bdba96","targetId":"b-e3ee2d9ac657df56f5bdba96","page":"region-fixed-grid-readout","operation":"move","beforeSHA256":"fd58c2404b317076a1cbb180addfce1e7d86d06562d238b0d24013fd38a73499","afterSHA256":"fd58c2404b317076a1cbb180addfce1e7d86d06562d238b0d24013fd38a73499","reason":"Same source body in authored semantic page"},{"oldId":"b-fb14d905826edf83b320aa1e","targetId":"b-fb14d905826edf83b320aa1e","page":"region-fixed-grid-readout","operation":"move","beforeSHA256":"b2184d704b577a9d5526c3d69abf339e4751bf9bc567a3ee87c026502f2df480","afterSHA256":"b2184d704b577a9d5526c3d69abf339e4751bf9bc567a3ee87c026502f2df480","reason":"Same source body in authored semantic page"},{"oldId":"b-d83fbf4b1bdcfb707b9c72d9","targetId":"b-d83fbf4b1bdcfb707b9c72d9","page":"region-fixed-grid-readout","operation":"move","beforeSHA256":"bdaf1f1b9d7ae2375cd34cb8efe97442640187bc4703a8fc74383ae4475308e1","afterSHA256":"bdaf1f1b9d7ae2375cd34cb8efe97442640187bc4703a8fc74383ae4475308e1","reason":"Same source body in authored semantic page"},{"oldId":"b-e1fe8ce0ec306e233167eeb8","targetId":"b-e1fe8ce0ec306e233167eeb8","page":"region-fixed-grid-readout","operation":"move","beforeSHA256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","afterSHA256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","reason":"Same source body in authored semantic page"},{"oldId":"b-c2c7215e9bfc677736945b24","targetId":"b-c2c7215e9bfc677736945b24","page":"region-fixed-grid-readout","operation":"move","beforeSHA256":"b8b69250624fd8c15e6f5db8690409848aaec027fc3b89222c6665c3044c9586","afterSHA256":"b8b69250624fd8c15e6f5db8690409848aaec027fc3b89222c6665c3044c9586","reason":"Same source body in authored semantic page"},{"oldId":"b-77bea269849bff02306cc9df","targetId":"b-77bea269849bff02306cc9df","page":"region-fixed-grid-readout","operation":"move","beforeSHA256":"e53e41cb3af866607e8baf6c5f6e3dec8444ca104e6dfe86891722e96d4b51ce","afterSHA256":"e53e41cb3af866607e8baf6c5f6e3dec8444ca104e6dfe86891722e96d4b51ce","reason":"Same source body in authored semantic page"},{"oldId":"region-fixed-grid-readout-title","targetId":"b-b02ad3171795e86846aa735f","operation":"retain-source-identity"}],"blocks":[{"id":"b-b02ad3171795e86846aa735f","type":"h2","scope":"region-fixed-grid-readout","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-b02ad3171795e86846aa735f","content":"## 从同一矩阵读出两份 2×2 结果","sha256":"9cfbf73da5623470d49f5af29a23132ac4a3f7da934d1139a80c3ea446540520","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"从同一矩阵读出两份 2×2 结果","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":null,"formulas":[],"previousBlock":null,"nextBlock":"b-3af39a1baae77e2a18591dde","prerequisites":[{"id":"b-ed5d3f5cc0f93253e8804706","url":"#b-ed5d3f5cc0f93253e8804706","role":"problem-context"}],"relatedBlocks":[{"id":"b-8253e0bd3b541e777b1ab311","url":"#b-8253e0bd3b541e777b1ab311","relation":"figure-and-explanation"}]},{"id":"b-3af39a1baae77e2a18591dde","type":"h2","scope":"explanation","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-3af39a1baae77e2a18591dde","content":"## 解释、算例与证据","sha256":"d71cf15e707c8a3674b4c90a599efee783cb84da5b8c8236a1d01c3add7b4958","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"从同一矩阵读出两份 2×2 结果","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-b02ad3171795e86846aa735f","nextBlock":"b-ce5f95165e174c468bccd39a","prerequisites":[{"id":"b-b02ad3171795e86846aa735f","url":"#b-b02ad3171795e86846aa735f","role":"section-title"},{"id":"b-ed5d3f5cc0f93253e8804706","url":"#b-ed5d3f5cc0f93253e8804706","role":"problem-context"}],"relatedBlocks":[{"id":"b-8253e0bd3b541e777b1ab311","url":"#b-8253e0bd3b541e777b1ab311","relation":"figure-and-explanation"}]},{"id":"b-ce5f95165e174c468bccd39a","type":"h1","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-ce5f95165e174c468bccd39a","content":"# P30 · 不同大小的区域，怎样读成固定结构？","sha256":"6837be3319b8248054aa18e902f639a5b4ad94cf795a5dc23d6a74dc6349e233","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"从同一矩阵读出两份 2×2 结果","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-3af39a1baae77e2a18591dde","nextBlock":"b-ed5d3f5cc0f93253e8804706","prerequisites":[{"id":"b-3af39a1baae77e2a18591dde","url":"#b-3af39a1baae77e2a18591dde","role":"section-title"},{"id":"b-ed5d3f5cc0f93253e8804706","url":"#b-ed5d3f5cc0f93253e8804706","role":"problem-context"}],"relatedBlocks":[{"id":"b-8253e0bd3b541e777b1ab311","url":"#b-8253e0bd3b541e777b1ab311","relation":"figure-and-explanation"}]},{"id":"b-ed5d3f5cc0f93253e8804706","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-ed5d3f5cc0f93253e8804706","content":"照着 6×8 矩阵实际取两个区域的 2×2 最大值，确认固定输出形状不等于相同内容。SPP 的不同分格层保存不同相对布局。当前整数例故意避开边界量化，P41 再解释真实连续区域的难点。","sha256":"d3c2aff40d77f95573a8243ba5fffda912e8d3ed195b0fd849eadf14161c476d","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"从同一矩阵读出两份 2×2 结果","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-ce5f95165e174c468bccd39a","nextBlock":"b-8253e0bd3b541e777b1ab311","prerequisites":[{"id":"b-ce5f95165e174c468bccd39a","url":"#b-ce5f95165e174c468bccd39a","role":"section-title"}],"relatedBlocks":[{"id":"b-8253e0bd3b541e777b1ab311","url":"#b-8253e0bd3b541e777b1ab311","relation":"figure-and-explanation"}]},{"id":"b-8253e0bd3b541e777b1ab311","type":"figure","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-8253e0bd3b541e777b1ab311","content":"图内标签与数值：Lecture 03 课堂图解 P30；不同大小的 RoI · 每格取最大值；0；1；2；0；1；2；1；0；1；2；3；1；2；4；3；1；0；2；5；2；3；5；4；2；1；3；4；1；0；2；6；1；0；1；2；0；1；3；4；2；0；0；1；0；0；1；2；0；2；1；4；3；2；5；3；4；固定 2 × 2 布局；各格汇总自身区域；位置是相对的；区域范围仍需保留；SPP-net 也可以组合多种分格尺度；1×1 · 2×2 · 3×3（分格示意）\n\n6×8 可核对数值矩阵上的两个 RoI，各按 2×2 分箱取最大值；下方显示多层空间分格。\n\n- 不同大小的 RoI · 每格取最大值\n- 0\n- 1\n- 2\n- 0\n- 1\n- 2\n- 1\n- 0\n- 1\n- 2\n- 3\n- 1\n- 2\n- 4\n- 3\n- 1\n- 0\n- 2\n- 5\n- 2\n- 3\n- 5\n- 4\n- 2\n- 1\n- 3\n- 4\n- 1\n- 0\n- 2\n- 6\n- 1\n- 0\n- 1\n- 2\n- 0\n- 1\n- 3\n- 4\n- 2\n- 0\n- 0\n- 1\n- 0\n- 0\n- 1\n- 2\n- 0\n- 2\n- 1\n- 4\n- 3\n- 2\n- 5\n- 3\n- 4\n- 固定 2 × 2 布局\n- 各格汇总自身区域\n- 位置是相对的\n- 区域范围仍需保留\n- SPP-net 也可以组合多种分格尺度\n- 1×1 · 2×2 · 3×3（分格示意）6×8 可核对数值矩阵上的两个 RoI，各按 2×2 分箱取最大值；下方显示多层空间分格。 [查看对应图解状态](/#/l3-r2-p30/0)[查看完整图解](/course/figures/l3-r2-p30-0.html)","sha256":"e34cbfc483d6da2d5ef34c9b5ab64c2e89edec7509490f7ce842d385743c80a9","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"从同一矩阵读出两份 2×2 结果","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":null,"formulas":[],"figure":{"caption":"6×8 可核对数值矩阵上的两个 RoI，各按 2×2 分箱取最大值；下方显示多层空间分格。 查看对应图解状态","labels":["不同大小的 RoI · 每格取最大值","0","1","2","0","1","2","1","0","1","2","3","1","2","4","3","1","0","2","5","2","3","5","4","2","1","3","4","1","0","2","6","1","0","1","2","0","1","3","4","2","0","0","1","0","0","1","2","0","2","1","4","3","2","5","3","4","固定 2 × 2 布局","各格汇总自身区域","位置是相对的","区域范围仍需保留","SPP-net 也可以组合多种分格尺度","1×1 · 2×2 · 3×3（分格示意）"],"description":"6×8 可核对数值矩阵上的两个 RoI，各按 2×2 分箱取最大值；下方显示多层空间分格。不同大小的 RoI · 每格取最大值01201210123124310252354213410261012013420010012021432534固定 2 × 2 布局各格汇总自身区域位置是相对的区域范围仍需保留SPP-net 也可以组合多种分格尺度1×1 · 2×2 · 3×3（分格示意）","evidenceIdentity":"6×8 可核对数值矩阵上的两个 RoI，各按 2×2 分箱取最大值；下方显示多层空间分格。 查看对应图解状态","images":[]},"previousBlock":"b-ed5d3f5cc0f93253e8804706","nextBlock":"b-efa6d45fcb21f08c9d188de0","prerequisites":[{"id":"b-ce5f95165e174c468bccd39a","url":"#b-ce5f95165e174c468bccd39a","role":"section-title"},{"id":"b-ed5d3f5cc0f93253e8804706","url":"#b-ed5d3f5cc0f93253e8804706","role":"problem-context"}],"relatedBlocks":[]},{"id":"b-efa6d45fcb21f08c9d188de0","type":"h2","scope":"narrative","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-efa6d45fcb21f08c9d188de0","content":"## 课堂讲述","sha256":"786ce10a6b32a492259a3dccd0a54e61cd41d28a34382fb7b5386c5f03b7ad69","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"从同一矩阵读出两份 2×2 结果","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-8253e0bd3b541e777b1ab311","nextBlock":"b-70b15c2345a27eb53c64c6bc","prerequisites":[{"id":"b-ce5f95165e174c468bccd39a","url":"#b-ce5f95165e174c468bccd39a","role":"section-title"},{"id":"b-ed5d3f5cc0f93253e8804706","url":"#b-ed5d3f5cc0f93253e8804706","role":"problem-context"}],"relatedBlocks":[{"id":"b-8253e0bd3b541e777b1ab311","url":"#b-8253e0bd3b541e777b1ab311","relation":"figure-and-explanation"}]},{"id":"b-70b15c2345a27eb53c64c6bc","type":"p","scope":"narrative","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-70b15c2345a27eb53c64c6bc","content":"SPP-net 等工作让区域读取成为明确的计算模块。区域可以大，也可以小，但可以按相对位置划分成固定数量的格子，在每格汇总特征，得到固定结构的输出。\n这不是把不同物体强行说成一样大，而是在为后面的预测器提供统一接口。后续仍然需要知道区域来自哪里，并预测它的类别或位置。\n这里有一个值得保留的研究积累：空间聚合并不是深度学习时代凭空出现的主意，新的卷积特征让原有的空间组织经验获得了新用途。需要注意，先裁图再编码和先编码再取区域并不严格相同，它们看到的上下文和几何变换不同；共享计算是取舍，不是无条件等价变形。","sha256":"842e062ba99099f4345595d783aaa3aad625e7e51d1060b34fec0079999c62b4","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"从同一矩阵读出两份 2×2 结果","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-efa6d45fcb21f08c9d188de0","nextBlock":"b-9dfec0da8b2cf1467a8177cc","prerequisites":[{"id":"b-efa6d45fcb21f08c9d188de0","url":"#b-efa6d45fcb21f08c9d188de0","role":"section-title"},{"id":"b-ed5d3f5cc0f93253e8804706","url":"#b-ed5d3f5cc0f93253e8804706","role":"problem-context"}],"relatedBlocks":[{"id":"b-8253e0bd3b541e777b1ab311","url":"#b-8253e0bd3b541e777b1ab311","relation":"figure-and-explanation"}]},{"id":"b-9dfec0da8b2cf1467a8177cc","type":"h3","scope":"roi-fixed-grid-example","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-9dfec0da8b2cf1467a8177cc","content":"### 同样是 2×2 输出，读到的内容可以不同","sha256":"458598c561a796cf6c0884e284ec9f9d5b2933fcbdc20a43cd405fda3c19705f","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"同样是 2×2 输出，读到的内容可以不同","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":"roi-fixed-grid-example","formulas":[],"previousBlock":"b-70b15c2345a27eb53c64c6bc","nextBlock":"b-aacf3c05c2a7d1e3cd061a9f","prerequisites":[{"id":"b-efa6d45fcb21f08c9d188de0","url":"#b-efa6d45fcb21f08c9d188de0","role":"section-title"},{"id":"b-ed5d3f5cc0f93253e8804706","url":"#b-ed5d3f5cc0f93253e8804706","role":"problem-context"}],"relatedBlocks":[{"id":"b-8253e0bd3b541e777b1ab311","url":"#b-8253e0bd3b541e777b1ab311","relation":"figure-and-explanation"},{"id":"b-aacf3c05c2a7d1e3cd061a9f","url":"#b-aacf3c05c2a7d1e3cd061a9f","relation":"same-reasoning"},{"id":"b-e3ee2d9ac657df56f5bdba96","url":"#b-e3ee2d9ac657df56f5bdba96","relation":"same-reasoning"},{"id":"b-fb14d905826edf83b320aa1e","url":"#b-fb14d905826edf83b320aa1e","relation":"same-reasoning"},{"id":"b-d83fbf4b1bdcfb707b9c72d9","url":"#b-d83fbf4b1bdcfb707b9c72d9","relation":"same-reasoning"}]},{"id":"b-aacf3c05c2a7d1e3cd061a9f","type":"p","scope":"roi-fixed-grid-example","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-aacf3c05c2a7d1e3cd061a9f","content":"区域大小不一，后面的固定形状预测头却希望得到一致长度的输入。把每个区域按相对位置分成 2×2 四格，再对每格取最大值，就能对每个通道得到四个数。若有 C 个通道，输出是 C×2×2，而非一律变成四个类别。","sha256":"cfc72e17de8c18086f6d7e26fa4eb9839f8d2a474a2ed8be9a0f522c9b642939","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"同样是 2×2 输出，读到的内容可以不同","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":"roi-fixed-grid-example","formulas":[],"previousBlock":"b-9dfec0da8b2cf1467a8177cc","nextBlock":"b-e3ee2d9ac657df56f5bdba96","prerequisites":[{"id":"b-9dfec0da8b2cf1467a8177cc","url":"#b-9dfec0da8b2cf1467a8177cc","role":"section-title"},{"id":"b-ed5d3f5cc0f93253e8804706","url":"#b-ed5d3f5cc0f93253e8804706","role":"problem-context"}],"relatedBlocks":[{"id":"b-8253e0bd3b541e777b1ab311","url":"#b-8253e0bd3b541e777b1ab311","relation":"figure-and-explanation"},{"id":"b-9dfec0da8b2cf1467a8177cc","url":"#b-9dfec0da8b2cf1467a8177cc","relation":"same-reasoning"},{"id":"b-e3ee2d9ac657df56f5bdba96","url":"#b-e3ee2d9ac657df56f5bdba96","relation":"same-reasoning"},{"id":"b-fb14d905826edf83b320aa1e","url":"#b-fb14d905826edf83b320aa1e","relation":"same-reasoning"},{"id":"b-d83fbf4b1bdcfb707b9c72d9","url":"#b-d83fbf4b1bdcfb707b9c72d9","relation":"same-reasoning"}]},{"id":"b-e3ee2d9ac657df56f5bdba96","type":"p","scope":"roi-fixed-grid-example","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-e3ee2d9ac657df56f5bdba96","content":"本页 6×8 教学矩阵提供两个整齐对齐的 RoI。橙色区域选择前两行末四列，四格最大值为 `[[2,1],[4,3]]`；青色区域选择第 2–5 行前四列，读出 `[[2,5],[3,4]]`。两份结果有相同形状，数值不同，且各自仍与原来的框配对。","sha256":"fd58c2404b317076a1cbb180addfce1e7d86d06562d238b0d24013fd38a73499","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"同样是 2×2 输出，读到的内容可以不同","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":"roi-fixed-grid-example","formulas":[],"previousBlock":"b-aacf3c05c2a7d1e3cd061a9f","nextBlock":"b-fb14d905826edf83b320aa1e","prerequisites":[{"id":"b-9dfec0da8b2cf1467a8177cc","url":"#b-9dfec0da8b2cf1467a8177cc","role":"section-title"},{"id":"b-ed5d3f5cc0f93253e8804706","url":"#b-ed5d3f5cc0f93253e8804706","role":"problem-context"}],"relatedBlocks":[{"id":"b-8253e0bd3b541e777b1ab311","url":"#b-8253e0bd3b541e777b1ab311","relation":"figure-and-explanation"},{"id":"b-9dfec0da8b2cf1467a8177cc","url":"#b-9dfec0da8b2cf1467a8177cc","relation":"same-reasoning"},{"id":"b-aacf3c05c2a7d1e3cd061a9f","url":"#b-aacf3c05c2a7d1e3cd061a9f","relation":"same-reasoning"},{"id":"b-fb14d905826edf83b320aa1e","url":"#b-fb14d905826edf83b320aa1e","relation":"same-reasoning"},{"id":"b-d83fbf4b1bdcfb707b9c72d9","url":"#b-d83fbf4b1bdcfb707b9c72d9","relation":"same-reasoning"}]},{"id":"b-fb14d905826edf83b320aa1e","type":"p","scope":"roi-fixed-grid-example","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-fb14d905826edf83b320aa1e","content":"空间金字塔进一步使用 1×1、2×2 等多个分格层。1×1 捕捉整区域统计，2×2 保留相对位置；把结果拼接后，原区域有多大都能得到固定长度。这里统一的是接口，不是抹掉区域之间的差异。","sha256":"b2184d704b577a9d5526c3d69abf339e4751bf9bc567a3ee87c026502f2df480","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"同样是 2×2 输出，读到的内容可以不同","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":"roi-fixed-grid-example","formulas":[],"previousBlock":"b-e3ee2d9ac657df56f5bdba96","nextBlock":"b-d83fbf4b1bdcfb707b9c72d9","prerequisites":[{"id":"b-9dfec0da8b2cf1467a8177cc","url":"#b-9dfec0da8b2cf1467a8177cc","role":"section-title"},{"id":"b-ed5d3f5cc0f93253e8804706","url":"#b-ed5d3f5cc0f93253e8804706","role":"problem-context"}],"relatedBlocks":[{"id":"b-8253e0bd3b541e777b1ab311","url":"#b-8253e0bd3b541e777b1ab311","relation":"figure-and-explanation"},{"id":"b-9dfec0da8b2cf1467a8177cc","url":"#b-9dfec0da8b2cf1467a8177cc","relation":"same-reasoning"},{"id":"b-aacf3c05c2a7d1e3cd061a9f","url":"#b-aacf3c05c2a7d1e3cd061a9f","relation":"same-reasoning"},{"id":"b-e3ee2d9ac657df56f5bdba96","url":"#b-e3ee2d9ac657df56f5bdba96","relation":"same-reasoning"},{"id":"b-d83fbf4b1bdcfb707b9c72d9","url":"#b-d83fbf4b1bdcfb707b9c72d9","relation":"same-reasoning"}]},{"id":"b-d83fbf4b1bdcfb707b9c72d9","type":"p","scope":"roi-fixed-grid-example","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-d83fbf4b1bdcfb707b9c72d9","content":"本页特意用整数边界，先隔离“大小不同却可用统一头”的想法。真实 RoI 边界不总落在格点；分箱取整带来的位置偏差，到 P41–43 再处理。","sha256":"bdaf1f1b9d7ae2375cd34cb8efe97442640187bc4703a8fc74383ae4475308e1","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"同样是 2×2 输出，读到的内容可以不同","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":"roi-fixed-grid-example","formulas":[],"previousBlock":"b-fb14d905826edf83b320aa1e","nextBlock":"b-e1fe8ce0ec306e233167eeb8","prerequisites":[{"id":"b-9dfec0da8b2cf1467a8177cc","url":"#b-9dfec0da8b2cf1467a8177cc","role":"section-title"},{"id":"b-ed5d3f5cc0f93253e8804706","url":"#b-ed5d3f5cc0f93253e8804706","role":"problem-context"}],"relatedBlocks":[{"id":"b-8253e0bd3b541e777b1ab311","url":"#b-8253e0bd3b541e777b1ab311","relation":"figure-and-explanation"},{"id":"b-9dfec0da8b2cf1467a8177cc","url":"#b-9dfec0da8b2cf1467a8177cc","relation":"same-reasoning"},{"id":"b-aacf3c05c2a7d1e3cd061a9f","url":"#b-aacf3c05c2a7d1e3cd061a9f","relation":"same-reasoning"},{"id":"b-e3ee2d9ac657df56f5bdba96","url":"#b-e3ee2d9ac657df56f5bdba96","relation":"same-reasoning"},{"id":"b-fb14d905826edf83b320aa1e","url":"#b-fb14d905826edf83b320aa1e","relation":"same-reasoning"}]},{"id":"b-e1fe8ce0ec306e233167eeb8","type":"h2","scope":"practice","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-e1fe8ce0ec306e233167eeb8","content":"## 沿这个问题继续","sha256":"cf44fcc62782fb4f6bdc42c16c5187d57556a03cb7ce6065e7bd9c36d8e7d733","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"从同一矩阵读出两份 2×2 结果","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-d83fbf4b1bdcfb707b9c72d9","nextBlock":"b-c2c7215e9bfc677736945b24","prerequisites":[{"id":"b-9dfec0da8b2cf1467a8177cc","url":"#b-9dfec0da8b2cf1467a8177cc","role":"section-title"},{"id":"b-ed5d3f5cc0f93253e8804706","url":"#b-ed5d3f5cc0f93253e8804706","role":"problem-context"}],"relatedBlocks":[{"id":"b-8253e0bd3b541e777b1ab311","url":"#b-8253e0bd3b541e777b1ab311","relation":"figure-and-explanation"}]},{"id":"b-c2c7215e9bfc677736945b24","type":"p","scope":"practice","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-c2c7215e9bfc677736945b24","content":"先读本页解释与图，再到 LAB03 改一个坐标或对象。","sha256":"b8b69250624fd8c15e6f5db8690409848aaec027fc3b89222c6665c3044c9586","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"从同一矩阵读出两份 2×2 结果","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-e1fe8ce0ec306e233167eeb8","nextBlock":"b-34cb45f6f9bd690b60d6b82c","prerequisites":[{"id":"b-e1fe8ce0ec306e233167eeb8","url":"#b-e1fe8ce0ec306e233167eeb8","role":"section-title"},{"id":"b-ed5d3f5cc0f93253e8804706","url":"#b-ed5d3f5cc0f93253e8804706","role":"problem-context"}],"relatedBlocks":[{"id":"b-8253e0bd3b541e777b1ab311","url":"#b-8253e0bd3b541e777b1ab311","relation":"figure-and-explanation"}]},{"id":"b-34cb45f6f9bd690b60d6b82c","type":"h4","scope":"region-fixed-grid-readout","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-34cb45f6f9bd690b60d6b82c","content":"#### 图示条件与公式约定","sha256":"ef7f915c1ae63ed67bd23fecf050249273ba0efa8965d2c8d1c1c6b2d3178122","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"从同一矩阵读出两份 2×2 结果","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-c2c7215e9bfc677736945b24","nextBlock":"b-77bea269849bff02306cc9df","prerequisites":[{"id":"b-e1fe8ce0ec306e233167eeb8","url":"#b-e1fe8ce0ec306e233167eeb8","role":"section-title"},{"id":"b-ed5d3f5cc0f93253e8804706","url":"#b-ed5d3f5cc0f93253e8804706","role":"problem-context"}],"relatedBlocks":[{"id":"b-8253e0bd3b541e777b1ab311","url":"#b-8253e0bd3b541e777b1ab311","relation":"figure-and-explanation"}]},{"id":"b-77bea269849bff02306cc9df","type":"p","scope":"document","url":"https://codingai-lec03.pages.dev/course/pages/l3-r2-p30.html#b-77bea269849bff02306cc9df","content":"F 是 6×8 教学矩阵。橙色 RoI 取前两行末四列，按每格最大值产生 [[2,1],[4,3]]；青色 RoI 取第2–5行前四列，产生 [[2,5],[3,4]]。这是整齐对齐的简化分格，不模拟历史库的边界取整。","sha256":"e53e41cb3af866607e8baf6c5f6e3dec8444ca104e6dfe86891722e96d4b51ce","chapter":{"id":"II","title":"区域、像素与实例"},"unit":{"id":"region-computation-reuse","title":"多个区域怎样复用同一份视觉计算？"},"page":{"id":"region-fixed-grid-readout","title":"从同一矩阵读出两份 2×2 结果","number":1},"sectionTitle":"从同一矩阵读出两份 2×2 结果","currentQuestion":"从逐区域编码到共享整图特征，哪些计算被保留，哪些接口必须重建？","sourceSlides":["l3-r2-p30"],"knowledgeBlock":null,"formulas":[],"previousBlock":"b-34cb45f6f9bd690b60d6b82c","nextBlock":null,"prerequisites":[{"id":"b-34cb45f6f9bd690b60d6b82c","url":"#b-34cb45f6f9bd690b60d6b82c","role":"section-title"},{"id":"b-ed5d3f5cc0f93253e8804706","url":"#b-ed5d3f5cc0f93253e8804706","role":"problem-context"}],"relatedBlocks":[{"id":"b-8253e0bd3b541e777b1ab311","url":"#b-8253e0bd3b541e777b1ab311","relation":"figure-and-explanation"}]}]}