. When I tested with 2k weights, I noticed a lot of graininess. I would like to know if this is caused by insufficient training. If I want to fine-tune for this scenario, do you have any suggestions?
image
Correct. We only use 8A100, 200K images(4K steps not finishing one epoch during training.) You can further improve it with more training data and time.
. When I tested with 2k weights, I noticed a lot of graininess. I would like to know if this is caused by insufficient training. If I want to fine-tune for this scenario, do you have any suggestions? image