知识卡片
检测命中需标签与位置都对
评估通常按类别和分数排序,在同图中按IoU阈值关联真值,每个真值最多贡献一个TP,重复检测是FP,未匹配真值为FN。具体忽略区域与难例规则依评测协议,不能只要和任意物体框相交就算识别正确。
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
- · 目标识别
原书出处 · 1 处
- · 查看原书依据
Assign detections to ground truth labels: For each image, sort all the detections by their score, ŷi. Then, loop over the sorted list in decreasing order. For each bounding box, compute the IoU with all the ground truth bounding boxes. Select the ground truth bounding box with the highest IoU. If the IoU is larger than a predefined threshold, (a typical value is 0.5) mark
参考来源
- 位置:《Foundations of Computer Vision (Antonio Torralba, Phillip Isola etc.)》第50章《目标识别》(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/c050b.xhtml)
- 结论依据:摘录中关于检测命中需标签与位置都对的定义、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:Assign detections to ground truth labels: For each image, sort all the detections by their score, ŷi. Then, loop over the sorted list in decreasing order. For each bounding box, compute the IoU with all the ground truth bounding boxes. Select the ground truth bounding box with the highest IoU. If the IoU is larger than a predefined threshold, (a typical value is 0.5) mark