import torch
from torch import Tensor
import torch.nn as nn
from typing import Type, Any, Callable, Union, List, Optional
from nfnets import WSConv2d, ScaledStdConv2d
from functools import partial
__all__ = ['nf_ResNet', 'nf_resnet18', 'nf_resnet34', 'nf_resnet50', 'nf_resnet101',
'nf_resnet152', 'nf_resnext50_32x4d', 'nf_resnext101_32x8d',
'nf_wide_resnet50_2', 'nf_wide_resnet101_2']
_nonlin_gamma = dict(
identity=1.0,
celu=1.270926833152771,
elu=1.2716004848480225,
gelu=1.7015043497085571,
leaky_relu=1.70590341091156,
log_sigmoid=1.9193484783172607,
log_softmax=1.0002083778381348,
relu=1.7139588594436646,
relu6=1.7131484746932983,
selu=1.0008515119552612,
sigmoid=4.803835391998291,
silu=1.7881293296813965,
softsign=2.338853120803833,
softplus=1.9203323125839233,
tanh=1.5939117670059204,
)
ignore_inplace = ['gelu', 'silu', 'softplus', ]
activation_fn = {
'identity': lambda x, *args, **kwargs: nn.Identity(*args, **kwargs)(x) * _nonlin_gamma['identity'],
'celu': lambda x, *args, **kwargs: nn.CELU(*args, **kwargs)(x) * _nonlin_gamma['celu'],
'elu': lambda x, *args, **kwargs: nn.ELU(*args, **kwargs)(x) * _nonlin_gamma['elu'],
'gelu': lambda x, *args, **kwargs: nn.GELU(*args, **kwargs)(x) * _nonlin_gamma['gelu'],
'leaky_relu': lambda x, *args, **kwargs: nn.LeakyReLU(*args, **kwargs)(x) * _nonlin_gamma['leaky_relu'],
'log_sigmoid': lambda x, *args, **kwargs: nn.LogSigmoid(*args, **kwargs)(x) * _nonlin_gamma['log_sigmoid'],
'log_softmax': lambda x, *args, **kwargs: nn.LogSoftmax(*args, **kwargs)(x) * _nonlin_gamma['log_softmax'],
'relu': lambda x, *args, **kwargs: nn.ReLU(*args, **kwargs)(x) * _nonlin_gamma['relu'],
'relu6': lambda x, *args, **kwargs: nn.ReLU6(*args, **kwargs)(x) * _nonlin_gamma['relu6'],
'selu': lambda x, *args, **kwargs: nn.SELU(*args, **kwargs)(x) * _nonlin_gamma['selu'],
'sigmoid': lambda x, *args, **kwargs: nn.Sigmoid(*args, **kwargs)(x) * _nonlin_gamma['sigmoid'],
'silu': lambda x, *args, **kwargs: nn.SiLU(*args, **kwargs)(x) * _nonlin_gamma['silu'],
'softplus': lambda x, *args, **kwargs: nn.Softplus(*args, **kwargs)(x) * _nonlin_gamma['softplus'],
'tanh': lambda x, *args, **kwargs: nn.Tanh(*args, **kwargs)(x) * _nonlin_gamma['tanh'],
}
def conv3x3(in_planes: int, out_planes: int, stride: int = 1, groups: int = 1, dilation: int = 1, base_conv: nn.Conv2d = ScaledStdConv2d) -> nn.Conv2d:
"""3x3 convolution with padding"""
return base_conv(in_planes, out_planes, kernel_size=3, stride=stride,
padding=dilation, groups=groups, bias=False, dilation=dilation)
def conv1x1(in_planes: int, out_planes: int, stride: int = 1, base_conv: nn.Conv2d = ScaledStdConv2d) -> nn.Conv2d:
"""1x1 convolution"""
return base_conv(in_planes, out_planes, kernel_size=1, stride=stride, bias=False)
class BasicBlock(nn.Module):
expansion: int = 1
def __init__(
self,
inplanes: int,
planes: int,
stride: int = 1,
downsample: Optional[nn.Module] = None,
groups: int = 1,
base_width: int = 64,
dilation: int = 1,
alpha: float = 0.2,
beta: float = 1.0,
activation: str = 'relu',
base_conv: nn.Conv2d = ScaledStdConv2d
) -> None:
super(BasicBlock, self).__init__()
if groups != 1 or base_width != 64:
raise ValueError(
'BasicBlock only supports groups=1 and base_width=64')
if dilation > 1:
raise NotImplementedError(
"Dilation > 1 not supported in BasicBlock")
# Both self.conv1 and self.downsample layers downsample the input when stride != 1
self.conv1 = conv3x3(inplanes, planes, stride, base_conv=base_conv)
self.activation = activation
if activation not in ignore_inplace:
self.act = partial(activation_fn[activation], inplace=True)
else:
self.act = partial(activation_fn[activation])
self.conv2 = conv3x3(planes, planes, base_conv=base_conv)
self.downsample = downsample
self.stride = stride
self.alpha = alpha
self.beta = beta
def forward(self, x: Tensor) -> Tensor:
identity = x
out = activation_fn[self.activation](x=x) * self.beta
out = self.conv1(out)
out = self.act(x=out)
out = self.conv2(out)
if self.downsample is not None:
identity = self.downsample(x)
out *= self.alpha
out += identity
return out
class Bottleneck(nn.Module):
# Bottleneck in torchvision places the stride for downsampling at 3x3 convolution(self.conv2)
# while original implementation places the stride at the first 1x1 convolution(self.conv1)
# according to "Deep residual learning for image recognition"https://arxiv.org/abs/1512.03385.
# This variant is also known as ResNet V1.5 and improves accuracy according to
# https://ngc.nvidia.com/catalog/model-scripts/nvidia:resnet_50_v1_5_for_pytorch.
expansion: int = 4
def __init__(
self,
inplanes: int,
planes: int,
stride: int = 1,
downsample: Optional[nn.Module] = None,
groups: int = 1,
base_width: int = 64,
dilation: int = 1,
alpha: float = 0.2,
beta: float = 1.0,
activation: str = 'relu',
base_conv: nn.Conv2d = ScaledStdConv2d,
) -> None:
super(Bottleneck, self).__init__()
width = int(planes * (base_width / 64.)) * groups
# Both self.conv2 and self.downsample layers downsample the input when stride != 1
self.conv1 = conv1x1(inplanes, width, base_conv=base_conv)
self.conv2 = conv3x3(width, width, stride, groups,
dilation, base_conv=base_conv)
self.conv3 = conv1x1(
width, planes * self.expansion, base_conv=base_conv)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
self.stride = stride
self.alpha = alpha
self.beta = beta
self.activation = activation
if activation not in ignore_inplace:
self.act = partial(activation_fn[activation], inplace=True)
else:
self.act = partial(activation_fn[activation])
def forward(self, x: Tensor) -> Tensor:
identity = x
out = activation_fn[self.activation](x) * self.beta
out = self.conv1(out)
out = self.act(x=out)
out = self.conv2(out)
out = self.act(x=out)
out = self.conv3(out)
if self.downsample is not None:
identity = self.downsample(x)
out *= self.alpha
out += identity
return out
class NFResNet(nn.Module):
def __init__(
self,
block: Type[Union[BasicBlock, Bottleneck]],
layers: List[int],
num_classes: int = 1000,
zero_init_residual: bool = False,
groups: int = 1,
width_per_group: int = 64,
replace_stride_with_dilation: Optional[List[bool]] = None,
alpha: float = 0.2,
beta: float = 1.0,
activation: str = 'relu',
base_conv: nn.Conv2d = ScaledStdConv2d
) -> None:
super(NFResNet, self).__init__()
assert activation in activation_fn.keys()
self.inplanes = 64
self.dilation = 1
if replace_stride_with_dilation is None:
# each element in the tuple indicates if we should replace
# the 2x2 stride with a dilated convolution instead
replace_stride_with_dilation = [False, False, False]
if len(replace_stride_with_dilation) != 3:
raise ValueError("replace_stride_with_dilation should be None "
"or a 3-element tuple, got {}".format(replace_stride_with_dilation))
self.groups = groups
self.base_width = width_per_group
self.conv1 = base_conv(3, self.inplanes, kernel_size=7, stride=2, padding=3,
bias=False)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(
block, 64, layers[0], alpha=alpha, beta=beta, activation=activation, base_conv=base_conv)
self.layer2 = self._make_layer(block, 128, layers[1], stride=2,
dilate=replace_stride_with_dilation[0], alpha=alpha, beta=beta, activation=activation, base_conv=base_conv)
self.layer3 = self._make_layer(block, 256, layers[2], stride=2,
dilate=replace_stride_with_dilation[1], alpha=alpha, beta=beta, activation=activation, base_conv=base_conv)
self.layer4 = self._make_layer(block, 512, layers[3], stride=2,
dilate=replace_stride_with_dilation[2], alpha=alpha, beta=beta, activation=activation, base_conv=base_conv)
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
self.fc = nn.Linear(512 * block.expansion, num_classes)
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(
m.weight, mode='fan_in', nonlinearity='linear')
# Zero-initialize the last BN in each residual branch,
# so that the residual branch starts with zeros, and each residual block behaves like an identity.
# This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
if zero_init_residual:
for m in self.modules():
if isinstance(m, Bottleneck):
# type: ignore[arg-type]
nn.init.constant_(m.bn3.weight, 0)
elif isinstance(m, BasicBlock):
# type: ignore[arg-type]
nn.init.constant_(m.bn2.weight, 0)
def _make_layer(self, block: Type[Union[BasicBlock, Bottleneck]], planes: int, blocks: int,
stride: int = 1, dilate: bool = False, alpha: float = 0.2, beta: float = 1.0, activation: str = 'relu', base_conv: nn.Conv2d = ScaledStdConv2d) -> nn.Sequential:
downsample = None
previous_dilation = self.dilation
if dilate:
self.dilation *= stride
stride = 1
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = nn.Sequential(
conv1x1(self.inplanes, planes * block.expansion,
stride, base_conv=base_conv),
)
layers = []
layers.append(block(self.inplanes, planes, stride, downsample, self.groups,
self.base_width, previous_dilation, alpha=alpha, beta=beta, activation=activation, base_conv=base_conv))
self.inplanes = planes * block.expansion
for _ in range(1, blocks):
layers.append(block(self.inplanes, planes, groups=self.groups,
base_width=self.base_width, dilation=self.dilation,
alpha=alpha, beta=beta, activation=activation,
base_conv=base_conv))
return nn.Sequential(*layers)
def _forward_impl(self, x: Tensor) -> Tensor:
# See note [TorchScript super()]
x = self.conv1(x)
x = self.relu(x)
x = self.maxpool(x)
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
x = self.avgpool(x)
x = torch.flatten(x, 1)
x = self.fc(x)
return x
def forward(self, x: Tensor) -> Tensor:
return self._forward_impl(x)
def _nf_resnet(
arch: str,
block: Type[Union[BasicBlock, Bottleneck]],
layers: List[int],
alpha: float,
beta: float,
activation: str,
base_conv: nn.Conv2d,
**kwargs: Any
) -> NFResNet:
model = NFResNet(block, layers, alpha=alpha, beta=beta, activation=activation, base_conv=base_conv, **kwargs)
return model
[docs]def nf_resnet18(alpha: float = 0.2, beta: float = 1.0, activation: str = 'relu', base_conv: nn.Conv2d = ScaledStdConv2d, **kwargs: Any) -> NFResNet:
r"""ResNet-18 model from
`"Deep Residual Learning for Image Recognition" <https://arxiv.org/pdf/1512.03385.pdf>`_
and `"High-Performance Large-Scale Image Recognition Without Normalization" <https://arxiv.org/pdf/2102.06171v1>`.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
return _nf_resnet('resnet18', BasicBlock, [2, 2, 2, 2], alpha=alpha, beta=beta, activation=activation, base_conv=base_conv,
**kwargs)
[docs]def nf_resnet34(alpha: float = 0.2, beta: float = 1.0, activation: str = 'relu', base_conv: nn.Conv2d = ScaledStdConv2d, **kwargs: Any) -> NFResNet:
r"""ResNet-34 model from
`"Deep Residual Learning for Image Recognition" <https://arxiv.org/pdf/1512.03385.pdf>`_
and `"High-Performance Large-Scale Image Recognition Without Normalization" <https://arxiv.org/pdf/2102.06171v1>`.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
return _nf_resnet('resnet34', BasicBlock, [3, 4, 6, 3], alpha=alpha, beta=beta, activation=activation, base_conv=base_conv,
** kwargs)
[docs]def nf_resnet50(alpha: float = 0.2, beta: float = 1.0, activation: str = 'relu', base_conv: nn.Conv2d = ScaledStdConv2d, **kwargs: Any) -> NFResNet:
r"""ResNet-50 model from
`"Deep Residual Learning for Image Recognition" <https://arxiv.org/pdf/1512.03385.pdf>`_
and `"High-Performance Large-Scale Image Recognition Without Normalization" <https://arxiv.org/pdf/2102.06171v1>`.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
return _nf_resnet('resnet50', Bottleneck, [3, 4, 6, 3], alpha=alpha, beta=beta, activation=activation, base_conv=base_conv,
** kwargs)
[docs]def nf_resnet101(alpha: float = 0.2, beta: float = 1.0, activation: str = 'relu', base_conv: nn.Conv2d = ScaledStdConv2d, **kwargs: Any) -> NFResNet:
r"""ResNet-101 model from
`"Deep Residual Learning for Image Recognition" <https://arxiv.org/pdf/1512.03385.pdf>`_.
and `"High-Performance Large-Scale Image Recognition Without Normalization" <https://arxiv.org/pdf/2102.06171v1>`.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
return _nf_resnet('resnet101', Bottleneck, [3, 4, 23, 3], alpha=alpha, beta=beta, activation=activation, base_conv=base_conv,
** kwargs)
[docs]def nf_resnet152(alpha: float = 0.2, beta: float = 1.0, activation: str = 'relu', base_conv: nn.Conv2d = ScaledStdConv2d, **kwargs: Any) -> NFResNet:
r"""ResNet-152 model from
`"Deep Residual Learning for Image Recognition" <https://arxiv.org/pdf/1512.03385.pdf>`_.
and `"High-Performance Large-Scale Image Recognition Without Normalization" <https://arxiv.org/pdf/2102.06171v1>`.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
return _nf_resnet('resnet152', Bottleneck, [3, 8, 36, 3], alpha=alpha, beta=beta, activation=activation, base_conv=base_conv,
** kwargs)
[docs]def nf_resnext50_32x4d(alpha: float = 0.2, beta: float = 1.0, activation: str = 'relu', base_conv: nn.Conv2d = ScaledStdConv2d, **kwargs: Any) -> NFResNet:
r"""ResNeXt-50 32x4d model from
`"Aggregated Residual Transformation for Deep Neural Networks" <https://arxiv.org/pdf/1611.05431.pdf>`_.
and `"High-Performance Large-Scale Image Recognition Without Normalization" <https://arxiv.org/pdf/2102.06171v1>`.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
kwargs['groups'] = 32
kwargs['width_per_group'] = 4
return _nf_resnet('resnext50_32x4d', Bottleneck, [3, 4, 6, 3],
alpha=alpha, beta=beta, activation=activation, base_conv=base_conv, **kwargs)
[docs]def nf_resnext101_32x8d(alpha: float = 0.2, beta: float = 1.0, activation: str = 'relu', base_conv: nn.Conv2d = ScaledStdConv2d, **kwargs: Any) -> NFResNet:
r"""ResNeXt-101 32x8d model from
`"Aggregated Residual Transformation for Deep Neural Networks" <https://arxiv.org/pdf/1611.05431.pdf>`_.
and `"High-Performance Large-Scale Image Recognition Without Normalization" <https://arxiv.org/pdf/2102.06171v1>`.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
kwargs['groups'] = 32
kwargs['width_per_group'] = 8
return _nf_resnet('resnext101_32x8d', Bottleneck, [3, 4, 23, 3],
alpha=alpha, beta=beta, activation=activation, base_conv=base_conv, **kwargs)
[docs]def nf_wide_resnet50_2(alpha: float = 0.2, beta: float = 1.0, activation: str = 'relu', base_conv: nn.Conv2d = ScaledStdConv2d, **kwargs: Any) -> NFResNet:
r"""Wide ResNet-50-2 model from
`"Wide Residual Networks" <https://arxiv.org/pdf/1605.07146.pdf>`_.
and `"High-Performance Large-Scale Image Recognition Without Normalization" <https://arxiv.org/pdf/2102.06171v1>`.
The model is the same as ResNet except for the bottleneck number of channels
which is twice larger in every block. The number of channels in outer 1x1
convolutions is the same, e.g. last block in ResNet-50 has 2048-512-2048
channels, and in Wide ResNet-50-2 has 2048-1024-2048.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
kwargs['width_per_group'] = 64 * 2
return _nf_resnet('wide_nf_resnet50_2', Bottleneck, [3, 4, 6, 3],
alpha=alpha, beta=beta, activation=activation, base_conv=base_conv, **kwargs)
[docs]def nf_wide_resnet101_2(alpha: float = 0.2, beta: float = 1.0, activation: str = 'relu', base_conv: nn.Conv2d = ScaledStdConv2d, **kwargs: Any) -> NFResNet:
r"""Wide ResNet-101-2 model from
`"Wide Residual Networks" <https://arxiv.org/pdf/1605.07146.pdf>`_.
and `"High-Performance Large-Scale Image Recognition Without Normalization" <https://arxiv.org/pdf/2102.06171v1>`.
The model is the same as ResNet except for the bottleneck number of channels
which is twice larger in every block. The number of channels in outer 1x1
convolutions is the same, e.g. last block in ResNet-50 has 2048-512-2048
channels, and in Wide ResNet-50-2 has 2048-1024-2048.
Args:
pretrained (bool): If True, returns a model pre-trained on ImageNet
"""
kwargs['width_per_group'] = 64 * 2
return _nf_resnet('wide_nf_resnet101_2', Bottleneck, [3, 4, 23, 3],
alpha=alpha, beta=beta, activation=activation, base_conv=base_conv, **kwargs)