Welcome to NFNets PyTorch’s documentation!
NFNets-PyTorch is an implementation of the paper: “High-Performance Large-Scale Image Recognition Without Normalization “. Original paper can be found at arxiv. You can find other implementations at PapersWithCode. If you’re looking for an explanation, look at this blogpost.
Install
Stable release
pip3 install nfnets-pytorch
Latest code
pip3 install git+https://github.com/vballoli/nfnets-pytorch
Sample usage
import torch
from torch import nn
from torchvision.models import vgg16
from nfnets import replace_conv, AGC, WSConv2d, ScaledStdConv2d
model = vgg16()
replace_conv(model, WSConv2d) # Original repo's implementation
replace_conv(model, ScaledStdConv2d) # timm
optim = torch.optim.SGD(model.parameters(), 1e-3) # Or any of your favourite optimizer
optim = AGC(model.parameters(), optim)