Closed yc-cui closed 4 months ago
The 2 stage training is unstable because the sample of a video sequence may be causes diverse difficulties in training. It is normal if you train more than 5 epochs, the average loss will slowly decrease.
OK,I will check the results when training is finished.
If you need my training log, i am reproduce my base_256_got model, i will release that maybe 2 hours later?
If you need my training log, i am reproduce my base_256_got model, i will release that maybe 2 hours later?
That's very kind of you for doing so.
This is the logs for stage 2:
Stage 1 training is normal:
I only changed the number of GPUs from 4 to 2.