yolov2-yolov3_PyTorch
yolov2-yolov3_PyTorch copied to clipboard
yjh0410
About training yolov3 gradient explode problem
When I trained YOLOV3 at Batch Size = 4, ‘nan’ appeared. After Debug, I found the infinite number appears at models/yolov3.py 118th code,
This is because the output of the network txtytwth_pred [:::,:, 2:] some values are too large, resulting in an infinite exp value.
Therefore, this is not the author's implementation problem, it is the reason for the unstable training, so I reduced the learning rate to 1e-4, and the training was normal. Hope can help to other friend.
@kill2013110 Batch size =4 is too small to stable training. Lowering the learning rate is helpful.
You could try my another YOLO project:
https://github.com/yjh0410/PyTorch_YOLO-Family
or:
https://github.com/yjh0410/FreeYOLO
@kill2013110 Batch size =4 is too small to stable training. Lowering the learning rate is helpful.
You could try my another YOLO project:
https://github.com/yjh0410/PyTorch_YOLO-Family
or:
https://github.com/yjh0410/FreeYOLO
Thank you for your reply，i have solved. I have another question. Do you think the infer speed of this project is the best way to achieve?
@yjh0410 In other words, do you think there is room for improvement in inference speed based on pytorch？
@kill2013110 I don't think that. Although my way of testing inference speed isn't too serious, it's clearly not rigorous. I implemented a better infer speed measurement function in the benchmark.py file in my FreeYOLO project, referring to MMDetection.