Open ZOUYIyi opened 4 years ago
Yolov2 / v3 / v4 use NMS, with the same greedynms with the same nms_thresh.
it detected severe boxes aroud one object with same class id
This is just a False Positive detection of Yolov3 due bad accuracy.
Yolov3 without nms looks like:
In your paper, you said you used DIoU-NMS. But in your project,you just use greedynms not DIoU-NMS
Yolov2 / v3 / v4 use NMS, with the same greedynms with the same nms_thresh.
it detected severe boxes aroud one object with same class id
This is just a False Positive detection of Yolov3 due bad accuracy. Yolov3 without nms looks like:
yes, yesterday I see exactly the picture you put out , I don't know why ,but I recompile all program,it works fine now.
@Amore-HDU Set for each yolo-layer
[yolo]
nms_kind=diounms
@AlexeyAB I have two question about nms: 1) You set greedynms in cfg, Does the result of greedynms is better than diounms ? 2) When i run the command: ./darknet detector test cfg/coco.data cfg/yolov4.cfg yolov4.weights data/horses.jpg When i set nms_kind=diounms in cfg, i found that i got l.nms_kind = 0 = default (line 1628 in detector.c ) not l.nms_kind = 2 = diounms. how can i use diounms, could you give some help? thanks a lot!
when I use the code and net of yolov4,I got good result ,but when I use this code and inference a image with yolov3.cfg and yolov3.weights, the nms seems don't work , it detected severe boxes aroud one object with same class id . I noticed it is use
nms_kind=greedynms beta_nms=0.6 in yolo layer in yolov4.cfg should I modify the yolov3.cfg by adding these two lines to it ,or I have made mistake there ?