mmedit.models.backbones.sr_backbones.rrdb_net 源代码
# Copyright (c) OpenMMLab. All rights reserved.
import torch
import torch.nn as nn
import torch.nn.functional as F
from mmcv.runner import load_checkpoint
from mmedit.models.common import (default_init_weights, make_layer,
pixel_unshuffle)
from mmedit.models.registry import BACKBONES
from mmedit.utils import get_root_logger
class ResidualDenseBlock(nn.Module):
"""Residual Dense Block.
Used in RRDB block in ESRGAN.
Args:
mid_channels (int): Channel number of intermediate features.
growth_channels (int): Channels for each growth.
"""
def __init__(self, mid_channels=64, growth_channels=32):
super().__init__()
for i in range(5):
out_channels = mid_channels if i == 4 else growth_channels
self.add_module(
f'conv{i+1}',
nn.Conv2d(mid_channels + i * growth_channels, out_channels, 3,
1, 1))
self.lrelu = nn.LeakyReLU(negative_slope=0.2, inplace=True)
self.init_weights()
def init_weights(self):
"""Init weights for ResidualDenseBlock.
Use smaller std for better stability and performance. We empirically
use 0.1. See more details in "ESRGAN: Enhanced Super-Resolution
Generative Adversarial Networks"
"""
for i in range(5):
default_init_weights(getattr(self, f'conv{i+1}'), 0.1)
def forward(self, x):
"""Forward function.
Args:
x (Tensor): Input tensor with shape (n, c, h, w).
Returns:
Tensor: Forward results.
"""
x1 = self.lrelu(self.conv1(x))
x2 = self.lrelu(self.conv2(torch.cat((x, x1), 1)))
x3 = self.lrelu(self.conv3(torch.cat((x, x1, x2), 1)))
x4 = self.lrelu(self.conv4(torch.cat((x, x1, x2, x3), 1)))
x5 = self.conv5(torch.cat((x, x1, x2, x3, x4), 1))
# Emperically, we use 0.2 to scale the residual for better performance
return x5 * 0.2 + x
class RRDB(nn.Module):
"""Residual in Residual Dense Block.
Used in RRDB-Net in ESRGAN.
Args:
mid_channels (int): Channel number of intermediate features.
growth_channels (int): Channels for each growth.
"""
def __init__(self, mid_channels, growth_channels=32):
super().__init__()
self.rdb1 = ResidualDenseBlock(mid_channels, growth_channels)
self.rdb2 = ResidualDenseBlock(mid_channels, growth_channels)
self.rdb3 = ResidualDenseBlock(mid_channels, growth_channels)
def forward(self, x):
"""Forward function.
Args:
x (Tensor): Input tensor with shape (n, c, h, w).
Returns:
Tensor: Forward results.
"""
out = self.rdb1(x)
out = self.rdb2(out)
out = self.rdb3(out)
# Emperically, we use 0.2 to scale the residual for better performance
return out * 0.2 + x
[文档]@BACKBONES.register_module()
class RRDBNet(nn.Module):
"""Networks consisting of Residual in Residual Dense Block, which is used
in ESRGAN and Real-ESRGAN.
ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks.
Real-ESRGAN: Training Real-World Blind Super-Resolution with Pure Synthetic Data. # noqa: E501
Currently, it supports [x1/x2/x4] upsampling scale factor.
Args:
in_channels (int): Channel number of inputs.
out_channels (int): Channel number of outputs.
mid_channels (int): Channel number of intermediate features.
Default: 64
num_blocks (int): Block number in the trunk network. Defaults: 23
growth_channels (int): Channels for each growth. Default: 32.
upscale_factor (int): Upsampling factor. Support x1, x2 and x4.
Default: 4.
"""
_supported_upscale_factors = [1, 2, 4]
def __init__(self,
in_channels,
out_channels,
mid_channels=64,
num_blocks=23,
growth_channels=32,
upscale_factor=4):
super().__init__()
if upscale_factor in self._supported_upscale_factors:
in_channels = in_channels * ((4 // upscale_factor)**2)
else:
raise ValueError(f'Unsupported scale factor {upscale_factor}. '
f'Currently supported ones are '
f'{self._supported_upscale_factors}.')
self.upscale_factor = upscale_factor
self.conv_first = nn.Conv2d(in_channels, mid_channels, 3, 1, 1)
self.body = make_layer(
RRDB,
num_blocks,
mid_channels=mid_channels,
growth_channels=growth_channels)
self.conv_body = nn.Conv2d(mid_channels, mid_channels, 3, 1, 1)
# upsample
self.conv_up1 = nn.Conv2d(mid_channels, mid_channels, 3, 1, 1)
self.conv_up2 = nn.Conv2d(mid_channels, mid_channels, 3, 1, 1)
self.conv_hr = nn.Conv2d(mid_channels, mid_channels, 3, 1, 1)
self.conv_last = nn.Conv2d(mid_channels, out_channels, 3, 1, 1)
self.lrelu = nn.LeakyReLU(negative_slope=0.2, inplace=True)
[文档] def forward(self, x):
"""Forward function.
Args:
x (Tensor): Input tensor with shape (n, c, h, w).
Returns:
Tensor: Forward results.
"""
if self.upscale_factor in [1, 2]:
feat = pixel_unshuffle(x, scale=4 // self.upscale_factor)
else:
feat = x
feat = self.conv_first(feat)
body_feat = self.conv_body(self.body(feat))
feat = feat + body_feat
# upsample
feat = self.lrelu(
self.conv_up1(F.interpolate(feat, scale_factor=2, mode='nearest')))
feat = self.lrelu(
self.conv_up2(F.interpolate(feat, scale_factor=2, mode='nearest')))
out = self.conv_last(self.lrelu(self.conv_hr(feat)))
return out
[文档] def init_weights(self, pretrained=None, strict=True):
"""Init weights for models.
Args:
pretrained (str, optional): Path for pretrained weights. If given
None, pretrained weights will not be loaded. Defaults to None.
strict (boo, optional): Whether strictly load the pretrained model.
Defaults to True.
"""
if isinstance(pretrained, str):
logger = get_root_logger()
load_checkpoint(self, pretrained, strict=strict, logger=logger)
elif pretrained is None:
# Use smaller std for better stability and performance. We
# use 0.1. See more details in "ESRGAN: Enhanced Super-Resolution
# Generative Adversarial Networks"
for m in [
self.conv_first, self.conv_body, self.conv_up1,
self.conv_up2, self.conv_hr, self.conv_last
]:
default_init_weights(m, 0.1)
else:
raise TypeError(f'"pretrained" must be a str or None. '
f'But received {type(pretrained)}.')