getting Nan Value while training yolov4 on custom data
arpitvw opened this issue · comments
Hello Team ,
I am using yolov4 for object detection and training it on custom dataset
after 300 iteration i am getting Nan value for weights
weights used:
- yolov4.conv.137 (after 6000 iteration also not able to detect object
- Tried to train it from scratch with below command but getting Nan value
!./darknet detector train data/obj.data cfg/yolo-obj.cfg yolov4.weights -clear -dont_show -map
Note: Training data is related to crops
solutions:
1.Decrease the learning rate suppose, for example, if you are using learning rate as 0.00001, use 0.0000001.
2.adjust the learning_decay
3.add clipnorm value (the last option but it is optional)
4. Increase the validation data