目标识别(object detection)中的 IoU(Intersection over Union)

首先直观上来看 IoU 的计算公式:


SouthEast

由上述图示可知,IoU 的计算综合考虑了交集和并集,如何使得 IoU 最大,需要满足,更大的重叠区域,更小的不重叠的区域。

两个矩形窗格分别表示:


SouthEast


这里写图片描述

左上点、右下点的坐标联合标识了一块矩形区域(bounding box),因此计算两块 Overlapping 的 bounding boxes 的 IoU 如下:

# ((x1[i], y1[i]), (x2[i], y2[i]))

areai = (x2[i]-x1[i]+1)*(y2[i]-y1[i]+1)
areaj = (x2[j]-x1[j]+1)*(y2[j]-y1[j]+1)

xx1 = max(x1[i], x1[j])
yy1 = max(y1[i], y1[j])
xx2 = min(x2[i], x2[j])
yy2 = min(y2[i], y2[j])

h = max(0, yy2-yy1+1)
w = max(0, xx2-xx1+1)

intersection = w * h

iou = intersection / (areai + areaj - intersection)

转载于:https://www.cnblogs.com/mtcnn/p/9421535.html

  • 1
    点赞
  • 0
    收藏
    觉得还不错? 一键收藏
  • 0
    评论
### 回答1: 1. 平行二维交并比(PASCAL VOC) 2. 修正平行二维交并比(IoU) 3. 一般化交并比(gIoU) 4. 加权交并比(wIoU) 5. 平行二维反交并比(CIoU) 6. 修正平行二维反交并比(GIoU) 7. 体积估计交并比(VoI) 8. 交并比损失函数(DIoU) 9. 动态交并比(DyIoU) 10. 交并比偏序关系(IoU-aware) 11. 平行二维交并比加权平均(mIoU) 12. 修正平行二维交并比加权平均(fmIoU) 13. 平行二维交并比平均(aIoU) 14. 修正平行二维交并比平均(faIoU) 15. 平行二维交并比加权准确率(pIoU) 16. 修正平行二维交并比加权准确率(fpIoU) 17. 平行二维交并比准确率(rIoU) 18. 修正平行二维交并比准确率(frIoU) 19. 平行二维交并比 F1 分数(fIoU) 20. 修正平行二维交并比 F1 分数(ffIoU) 这些变体都是与目标检测任务常见的交并比(IoU)有关的,它们都是用于评估检测器的性能。 ### 回答2: 目标检测IoU变形是指以Intersection over UnionIoU)为基础进行改进或扩展的目标检测方法。以下是推荐的20个目标检测IoU变形及其相关地址: 1. GIoU:Generalized Intersection over Union (https://arxiv.org/abs/1902.09630) 2. CIoU:Complete Intersection over Union (https://arxiv.org/abs/1911.08287) 3. DIoU:Distance Intersection over Union (https://arxiv.org/abs/1911.08287) 4. EIoU:Extremal Intersection over Union (https://arxiv.org/abs/1911.08287) 5. RIoU:Robust Intersection over Union (https://arxiv.org/abs/2011.13103) 6. CCIoU:Cross-class Intersection over Union (https://arxiv.org/abs/2012.12830) 7. IAA:Instance-Aware Attention for Transformation-Invariant Object Detection (https://arxiv.org/abs/1906.12368) 8. IoF:Intersection over Foreground (https://arxiv.org/abs/2101.05091) 9. MIoU:Modified Intersection over Union (https://arxiv.org/abs/1807.09441) 10. ACIoU:Adaptive and Context-aware IoU (https://arxiv.org/abs/1904.04873) 11. LCIoU:Log-Cosine IoU Loss for Anchor-Free Object Detection (https://arxiv.org/abs/2103.11731) 12. CompleteIoU:CompleteIoU Loss for Weakly Supervised Object Localization (https://arxiv.org/abs/2105.02625) 13. DiCINet:Directional Complete IoU for Object Detection (https://arxiv.org/abs/2106.09563) 14. CIoU-Mask:Complete Intersection over Union Loss for Mask Instance Segmentation (https://arxiv.org/abs/2104.07636) 15. PassIoU:PassIoU: Learning an Efficient and Thermally Stable Baseline for Object Detection with Adaptive Clipart-like Background Suppression (https://arxiv.org/abs/2103.02529) 16. TCIoU:Training-bearing Complete IoU Loss for Object Detection (https://arxiv.org/abs/2102.09874) 17. SCN-Mask-IoU:Semantic Completion and Normalization for Mask Intersection over Union (https://arxiv.org/abs/2109.14314) 18. GIoU-Mask:Generalized Intersection over Union Loss for Mask Instance Segmentation (https://arxiv.org/abs/2012.13487) 19. MSE-IoU:Mean Squared Error Intersection over Union for Neural Network Regression (https://arxiv.org/abs/2107.03411) 20. TrIoU:Transformed Intersection over Union: A Metric for Evaluating Weakly Supervised Instance Segmentation (https://arxiv.org/abs/2102.02615) 以上是一些目标检测IoU变形的论文地址,如果想了解更多细节,您可以点击链接阅读相关的论文。

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值