← 返回刚才阅读的位置

知识卡片

普通读书笔记卡 · 9079.b · 生成模型:GPT-6

监督光流依赖真值的有效范围

理解等级:中级

训练图像对到流场的映射,需要为有效像素提供可靠位移目标。合成渲染、特殊传感器或人工标注具有不同误差与覆盖范围,遮挡和稀疏标签也需掩码处理;不能只把数据集列名当作等精度的全图真值。

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

      The simplest formulation for learning to estimate optical flow is when we have available a dataset of image sequences with associated ground truth optical flow. Researchers have used synthetic data [68], using lidar [157] or human annotations [302] to build datasets with ground truth motion. In previous approaches, ground truth data could be used for evaluation; however, we will use it here to train a

    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:摘录中关于监督光流依赖真值的有效范围的定义、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 simplest formulation for learning to estimate optical flow is when we have available a dataset of image sequences with associated ground truth optical flow. Researchers have used synthetic data [68], using lidar [157] or human annotations [302] to build datasets with ground truth motion. In previous approaches, ground truth data could be used for evaluation; however, we will use it here to train a