Closed White-Friday closed 3 months ago
This is only for training. You can see the report for some preliminary results.
If you want to finetune an image-based model from an already trained one, I would suggest 0.5 for not masking, and 0.5 for the rest. If you want to train from scratch, 0.1/0.2 is enough.
This is only for training. You can see the report for some preliminary results.