目标检测(object detection)=what + where
Localization+Recongnition
类别标签(category label)
置信度得分(confidence score)
最小外接矩形(bounding box)
定位是找到检测图像中带有一个给定标签的单个目标;
检测是寻找到图像中带有给定标签的所有目标
目标检测性能指标=检测精度+检测速度
Precision,Recall,Fl score
IoU(Intersection over Union)
P-R curve (Precison-Recall cureve)
AP(Average Precision)
mAP(mean Average Precision)
前传耗时
每秒帧数FPS(Frames Per Second)
浮点运算量(FLOPS)
混淆矩阵(confusion matrix)
对于边界框的分类用混淆矩阵衡量
边界框框的准不准用IoU来衡量
IoU=1 predicted and the ground-truth bounding boxes perfectly overlap.
you can set a threshold value(阈值) for the IoU to determine if the objext detection is valid or not.
Let‘s say you set IoU to 0.5,in that case:
if IoU≥0.5,-->True Positive(TP)
if IoU<0.5, it is a wring detection and classify it as False Positive(FP)
When a ground truth is present in the image and modell failed to detect the object, claasify it as False Negative(FN).
True Negative(TN):TN is every part of the image where we did not predict an object. This metrics is not useful for ibject detection, hence we ignore TN.
AP衡量的是学习出来的模型在每个类别上的好坏。
mAP衡量的是学出的模型在所有类别上的好坏。mAP就是取所有类别上AP的平均值。
IoU阈值越大,对检测框要求就越紧,recall召回率就相对小
前传耗时(ms):从输入一张图像到输出最终结果所消耗的时间,包括前处理耗时(如图像归一化)、网络前传耗时、后处理耗时(如非极大值抑制)
每秒帧数FPS(Frames Per Second):每秒钟能处理的图像数量
浮点运算量(FLOPS):处理一张图像所需要的浮点运算数量,跟具体软硬件没有关系,可以公平地比较不同算法之间的检测速度。
最适合新手入门的【YOLOV5目标实战】教程!基于Pytorch搭建YOLOV5目标检测平台!环境部署+项目实战(深度学习/计算机视觉)_哔哩哔哩_bilibili