YOLO2/3 can only regression the number of bounding box less than 30?

pjreddie / darknet

Convolutional Neural Networks

http://pjreddie.com/darknet/

Other

25.82k stars 21.33k forks source link

YOLO2/3 can only regression the number of bounding box less than 30? #600

Open VisionZQ opened 6 years ago

VisionZQ commented 6 years ago

Hi , Thanks for you share. I had view some source code and had some question about it. If the number of GT boxes more than 30 , yolo only deal with 30 boxes?

AlexeyAB commented 6 years ago

Yolo v2 = 30 boxes: https://github.com/pjreddie/darknet/blob/259be3217b0331d07c8c5c1995d90ab9f3b768c1/src/region_layer.c#L33
Yolo v3 = 90 boxes: https://github.com/pjreddie/darknet/blob/259be3217b0331d07c8c5c1995d90ab9f3b768c1/src/yolo_layer.c#L41

VisionZQ commented 6 years ago

ok. If the number of boxes more than 90, how do yolo deal with？ I try to change the l.max_boxes through setting max = 100 in cfg file , but yolo will crash (core dump)？？？

AlexeyAB commented 6 years ago

I changed it in my fork: https://github.com/AlexeyAB/darknet

Yolo v2 - set max=200 in the [region] layer in cfg-file: https://github.com/AlexeyAB/darknet/blob/35cc0aaa15b991b348cc8d9623eed5d4f8a1e435/src/region_layer.c#L31
Yolo v3 - set max=200 in the each of 3 [yolo] layers in cfg-file: https://github.com/AlexeyAB/darknet/blob/35cc0aaa15b991b348cc8d9623eed5d4f8a1e435/src/yolo_layer.c#L43

VisionZQ commented 6 years ago

thanks a lot. l will try again.

RyanCV commented 6 years ago

@AlexeyAB

Yolo v3 - set max=200 in the each of 3 [yolo] layers in cfg-file:

When max=200, does it mean the number of ground truth boxes be >= 200? For example, if there is only 150 ground truth boxes, can I set max=200 or higher value, i.e. max=200?