When I use yolov3 to train my dataset and there are more than 60 objects in a picture,do I need to change yolov3.cfg or other parameters?

Question

When I use yolov3 to train my dataset and there are more than 60 objects in a picture,and these objects are very dense, the final train-loss is 16.4,do I need to change yolov3.cfg or other parameters?or what can I do to reduce the train-loss?

Answer 1

From your question I understand that there are many small objects in your images. You could do a few things to improve your training-loss:

• set flag random=1 in your .cfg-file - it will increase precision by training Yolo for different resolutions.
• increase network resolution in your .cfg-file ( height=608, width=608 or any value multiple of 32) - it will increase precision.

Also I'd suggest you to have some test images and check your mAP while training to if the model's detection accuracy is increasing as the loss goes down.