← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9082 · 生成模型:GPT-6

自监督把成像假设变成目标

理解等级:中级

无光流真值训练仍要借亮度恒常、局部平滑或相机几何等约束产生信号。只由相机运动解释流的约束依赖静态刚性场景,不能用于任意独立运动对象;去掉标签没有去掉建模假设。

这张卡片看不懂?

已有反馈 · 0 条

    还没有人提交反馈。

    在整书笔记中的位置

    原书出处 · 1 处

    • · 查看原书依据

      Unsupervised methods for training an optical flow model will make some assumptions about dynamic image formation. Those assumptions will be similar to the ones we have presented all along this chapter: (1) when the motion is due only to camera motion, the optical flow will have to fit the equations of the projected motion that provide constraints that can be used to train a model;

    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:摘录中关于自监督把成像假设变成目标的定义、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:Unsupervised methods for training an optical flow model will make some assumptions about dynamic image formation. Those assumptions will be similar to the ones we have presented all along this chapter: (1) when the motion is due only to camera motion, the optical flow will have to fit the equations of the projected motion that provide constraints that can be used to train a model;