Giters

cszn / KAIR

Image Restoration Toolbox (PyTorch). Training and testing codes for DPIR, USRNet, DnCNN, FFDNet, SRMD, DPSR, BSRGAN, SwinIR

Home Page:https://cszn.github.io/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Momentum for BatchNorm2d

ryanxingql opened this issue · comments

commented

I notice that the momentum for BatchNorm2d is 0.9.

KAIR/models/basicblock.py

Line 69 in 7e51c16

L.append(nn.BatchNorm2d(out_channels, momentum=0.9, eps=1e-04, affine=True))

However, the default momentum of BatchNorm2d is 0.1.
https://pytorch.org/docs/stable/generated/torch.nn.BatchNorm2d.html

Should we use momentum=0.9 to construct DnCNN?

commented

Both are OK for Gaussian denoising.