知识卡片
可学习遮挡掩码可能退化
遮挡权重可排除不存在真实对应的像素,但若掩码只以压低同一个重建损失训练,全部判为遮挡即可获得虚假低损失。需要额外约束、监督或一致性设计限制这种退化,并分别检查掩码覆盖与流质量。
这张卡片看不懂?
这张卡片已通过人工审查,不再接受失败反馈。
已有反馈 · 0 条
还没有人提交反馈。
在整书笔记中的位置
- · 学习运动估计
原书出处 · 1 处
- · 查看原书依据
The learning is done by searching over the parameter space for the parameters θ that minimize the photometric loss over a large collection of videos. The photometric loss can also be replaced by the L1 or other robust norms. If the network also predicts occlusions, the photometric loss can include a weight that cancels the contribution of occluded pixels to the loss.
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第49章《学习运动估计》(来源项目:a19c7933-4eda-4176-aebd-30fe7e1ad292;源文件:extracted/Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)/_epub-src/OPS/xhtml/c049.xhtml)
- 结论依据:摘录中关于可学习遮挡掩码可能退化的定义、比较或条件支撑卡片;原书表述不严谨的范围已在覆盖记录逐项限定,不把案例结果外推为普遍结论。
- 原始内容:The learning is done by searching over the parameter space for the parameters θ that minimize the photometric loss over a large collection of videos. The photometric loss can also be replaced by the L1 or other robust norms. If the network also predicts occlusions, the photometric loss can include a weight that cancels the contribution of occluded pixels to the loss.