search

AlexeyAB / darknet

YOLOv4 / Scaled-YOLOv4 / YOLO - Neural Networks for Object Detection (Windows and Linux version of Darknet )
http://pjreddie.com/darknet/
Other
21.73k stars 7.96k forks source link

Yolo v4 fine-tuning #5695

Open sctrueew opened 4 years ago

sctrueew commented 4 years ago

@AlexeyAB Hi,

I've trained a model with ~ 300 classes and I'm gonna train a new model with 350 classes and I use this command to make a pre-trained model darknet partial my-yolov4-custom.cfg backup\my-yolov4-custom.weights my-yolov4-custom.conv.137 137

but when I use that pre-trained for new training, I get nan avg after about 1K iter.

aicukltd commented 4 years ago

Try using the config from the YOLO MARK repo.

I had this same problem with any variation of the V4 configs then as soon as I used the config from the YOLO MARK release folder everything started working perfectly.

https://github.com/AlexeyAB/Yolo_mark/blob/master/x64/Release/yolo-obj.cfg

You are not the first person to report a problem with V4 configs.