知识卡片
合成角点可启动学习检测
书中以SuperPoint说明:先在合成图上定义明确关键点,训练检测器,再借图像变换及伪标签扩展到真实图像。几何增强必须同步变换点坐标。这种流程不要求所有真实图人工标注,也不能把任意学习检测器统一归为全手工监督。
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
原书出处 · 1 处
- · 查看原书依据
Figure 44.4 illustrates a general formulation for detecting keypoints using a classifier. We start by training a classifier using hand-annotated images with keypoints. This can be done by generating simple synthetic images were keypoints are unambiguously defined. The training set can be extended by running an initial keypoint detector on images and then augmenting the images with random geometric transformations. Then we can use the
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第44章《多视图几何与运动恢复结构》(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/c044.xhtml)
- 结论依据:摘录中关于合成角点可启动学习检测的定义、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:Figure 44.4 illustrates a general formulation for detecting keypoints using a classifier. We start by training a classifier using hand-annotated images with keypoints. This can be done by generating simple synthetic images were keypoints are unambiguously defined. The training set can be extended by running an initial keypoint detector on images and then augmenting the images with random geometric transformations. Then we can use the