BCELoss was unstable so just get logits from model and use BCEwithLog… #309

deshwalmahesh · 2022-06-07T18:54:22Z

Use bias = False in Line 9 of model of u2net.py
Line 67-69: Usetorch.backends.cudnn.benchmark = Trueand use Mixed Precision Training
Use BceWithLogitLoss : Line 29 in place of BCELoss of `training
Return logits as d0-d6 in place of Sigmoids: Line 525 of u2net.py
Set torch.backends.cudnn.benchmark = True : Line 69 of training

…it. Use AMP and cuda benchmarks = True for fast training

deshwalmahesh and others added 15 commits June 8, 2022 00:13

BCELoss was unstable so just get logits from model and use BCEwithLog…

4f173fd

…it. Use AMP and cuda benchmarks = True for fast training

torch.channels_last added

c82c853

input tensors copying process change

25c12dc

added FocalLoss from focal_loss.py

64a2372

added dummy train data and Focal Loss defaults

d7f8556

changes in class U2Net for the last return statement

fb8a296

pushing and closing

4d21a2f

pushing api code

d596c3e

Update README.md

a1ce8ab

Update bg.py

a19c4fb

Update __init__.py

c7d5d5e

minimal code

1547715

minimal code

b87a113

Update bg.py

6d66ba5

Delete __init__.py

ff19c0b

Provide feedback