Back when it was initially implemented in 2019, the LR scheduler in the segmentation reference scripts couldn't be implemented with native PyTorch schedulers, so we had to resort to LambdaLR
|
lr_scheduler = torch.optim.lr_scheduler.LambdaLR( |
|
optimizer, |
|
lambda x: (1 - x / (len(data_loader) * args.epochs)) ** 0.9) |
It might be that this is now available in PyTorch natively, and this can be simplified.
cc @datumbox
Back when it was initially implemented in 2019, the LR scheduler in the segmentation reference scripts couldn't be implemented with native PyTorch schedulers, so we had to resort to
LambdaLRvision/references/segmentation/train.py
Lines 136 to 138 in 9275cc6
It might be that this is now available in PyTorch natively, and this can be simplified.
cc @datumbox