The loss value is too large #6
Description
When I use my own dataset for training, the initial loss value is about 60k, when it drops to 6k I still can't get good results when testing. Is this loss value normal? How big does it need to drop to in order to have better results?