pytorch-slimming
pytorch-slimming copied to clipboard
Is not l1 loss the smooth l1 mentioned in the paper?
https://github.com/foolwood/pytorch-slimming/blob/7d13c090720a2e614bf638f0a1e66aa97bffee0b/main.py#L97-L101
is it right?
Looks like he's taking the subgradient approach