FCIS
FCIS copied to clipboard
how to train the model without OHEM?
When I train the model without OHTM, the loss will become Nan. Can you give more details on how to train the model without OHEM?
Why I want to train the model without OHEM? Right now, I am developing a new algorithm based on your project. I wanna to train a baseline model with OHEM.
Can confirm that. The bbox deltas drop to Nan after a short while. Tried to figure it out but I gave up, just enabled it for now. Should be added to the readme I think, it actually took me quite a while to figure out that my error came from the OHEM setting. VOC btw doesn't use OHEM, that adds to the initial confusion because "false" works with VOC.