search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.65k stars 7.96k forks source link

Precision Recall F1 mAP calculation question #8054

Open cpsu00 opened 3 years ago

cpsu00 commented 3 years ago

To my knowledge

Precision, Recall, F1 use conf_thresh=0.25 and IoU_thresh=0.5 as default mAP use conf_thresh=0.005 and IoU_thresh=0.5 as default

Is this correct?

Fetulhak commented 3 years ago

@cpsu00 the conf_thresh value is 0.25 during training but they use 0.005 during inference as a default value. All the evaluation metrics use the same conf_thresh and IOU_thresh as you set it during inference time. you can use the ./darknet detector map..... to see evaluation results.