Source code for mmagic.models.editors.nafnet.nafnet_net
# Copyright (c) 2022 megvii-model. All Rights Reserved.
# Copyright (c) OpenMMLab. All rights reserved.
import torch
import torch.nn as nn
import torch.nn.functional as F
from mmengine.model import BaseModule
from mmagic.registry import MODELS
from .naf_avgpool2d import Local_Base
from .naf_layerNorm2d import LayerNorm2d
@MODELS.register_module()
[docs]class NAFNet(BaseModule):
"""NAFNet.
The original version of NAFNet in "Simple Baseline for Image Restoration".
Args:
img_channels (int): Channel number of inputs.
mid_channels (int): Channel number of intermediate features.
middle_blk_num (int): Number of middle blocks.
enc_blk_nums (List of int): Number of blocks for each encoder.
dec_blk_nums (List of int): Number of blocks for each decoder.
"""
def __init__(self,
img_channels=3,
mid_channels=16,
middle_blk_num=1,
enc_blk_nums=[],
dec_blk_nums=[]):
super().__init__()
self.intro = nn.Conv2d(
in_channels=img_channels,
out_channels=mid_channels,
kernel_size=3,
padding=1,
stride=1,
groups=1,
bias=True)
self.ending = nn.Conv2d(
in_channels=mid_channels,
out_channels=img_channels,
kernel_size=3,
padding=1,
stride=1,
groups=1,
bias=True)
self.encoders = nn.ModuleList()
self.decoders = nn.ModuleList()
self.middle_blks = nn.ModuleList()
self.ups = nn.ModuleList()
self.downs = nn.ModuleList()
chan = mid_channels
for num in enc_blk_nums:
self.encoders.append(
nn.Sequential(*[NAFBlock(chan) for _ in range(num)]))
self.downs.append(nn.Conv2d(chan, 2 * chan, 2, 2))
chan = chan * 2
self.middle_blks = \
nn.Sequential(
*[NAFBlock(chan) for _ in range(middle_blk_num)]
)
for num in dec_blk_nums:
self.ups.append(
nn.Sequential(
nn.Conv2d(chan, chan * 2, 1, bias=False),
nn.PixelShuffle(2)))
chan = chan // 2
self.decoders.append(
nn.Sequential(*[NAFBlock(chan) for _ in range(num)]))
self.padder_size = 2**len(self.encoders)
[docs] def forward(self, inp):
"""Forward function.
args:
inp: input tensor image with (B, C, H, W) shape
"""
B, C, H, W = inp.shape
inp = self.check_image_size(inp)
x = self.intro(inp)
encs = []
for encoder, down in zip(self.encoders, self.downs):
x = encoder(x)
encs.append(x)
x = down(x)
x = self.middle_blks(x)
for decoder, up, enc_skip in zip(self.decoders, self.ups, encs[::-1]):
x = up(x)
x = x + enc_skip
x = decoder(x)
x = self.ending(x)
x = x + inp
return x[:, :, :H, :W]
[docs] def check_image_size(self, x):
"""Check image size and pad images so that it has enough dimension do
downsample.
args:
x: input tensor image with (B, C, H, W) shape.
"""
_, _, h, w = x.size()
mod_pad_h = (self.padder_size -
h % self.padder_size) % self.padder_size
mod_pad_w = (self.padder_size -
w % self.padder_size) % self.padder_size
x = F.pad(x, (0, mod_pad_w, 0, mod_pad_h))
return x
@MODELS.register_module()
[docs]class NAFNetLocal(Local_Base, NAFNet):
"""The original version of NAFNetLocal in "Simple Baseline for Image
Restoration".
NAFNetLocal uses local average pooling modules than NAFNet.
Args:
img_channels (int): Channel number of inputs.
mid_channels (int): Channel number of intermediate features.
middle_blk_num (int): Number of middle blocks.
enc_blk_nums (List of int): Number of blocks for each encoder.
dec_blk_nums (List of int): Number of blocks for each decoder.
"""
def __init__(self,
*args,
train_size=(1, 3, 256, 256),
fast_imp=False,
**kwargs):
Local_Base.__init__(self)
NAFNet.__init__(self, *args, **kwargs)
N, C, H, W = train_size
base_size = (int(H * 1.5), int(W * 1.5))
self.eval()
with torch.no_grad():
self.convert(
base_size=base_size, train_size=train_size, fast_imp=fast_imp)
# Components for NAFNet
[docs]class NAFBlock(BaseModule):
"""NAFNet's Block in paper.
Simple gate will shrink the channel to a half.
To keep the number of channels,
it expands the channels first.
Args:
in_channels (int): number of channels
DW_Expand (int): channel expansion factor for part 1
FFN_Expand (int): channel expansion factor for part 2
drop_out_rate (float): drop out ratio
"""
def __init__(self,
in_channels,
DW_Expand=2,
FFN_Expand=2,
drop_out_rate=0.):
super().__init__()
# Part 1
dw_channel = in_channels * DW_Expand
self.conv1 = nn.Conv2d(
in_channels=in_channels,
out_channels=dw_channel,
kernel_size=1,
padding=0,
stride=1,
groups=1,
bias=True)
self.conv2 = nn.Conv2d(
in_channels=dw_channel,
out_channels=dw_channel,
kernel_size=3,
padding=1,
stride=1,
groups=dw_channel,
bias=True)
# Simplified Channel Attention
self.sca = nn.Sequential(
nn.AdaptiveAvgPool2d(1),
nn.Conv2d(
in_channels=dw_channel // 2,
out_channels=dw_channel // 2,
kernel_size=1,
padding=0,
stride=1,
groups=1,
bias=True),
)
self.conv3 = nn.Conv2d(
in_channels=dw_channel // 2,
out_channels=in_channels,
kernel_size=1,
padding=0,
stride=1,
groups=1,
bias=True)
# Part 2
ffn_channel = FFN_Expand * in_channels
self.conv4 = nn.Conv2d(
in_channels=in_channels,
out_channels=ffn_channel,
kernel_size=1,
padding=0,
stride=1,
groups=1,
bias=True)
self.conv5 = nn.Conv2d(
in_channels=ffn_channel // 2,
out_channels=in_channels,
kernel_size=1,
padding=0,
stride=1,
groups=1,
bias=True)
# Simple Gate
self.sg = SimpleGate()
# Layer Normalization
self.norm1 = LayerNorm2d(in_channels)
self.norm2 = LayerNorm2d(in_channels)
# Dropout
self.dropout1 = nn.Dropout(
drop_out_rate) if drop_out_rate > 0. else nn.Identity()
self.dropout2 = nn.Dropout(
drop_out_rate) if drop_out_rate > 0. else nn.Identity()
# Feature weight ratio
self.beta = nn.Parameter(
torch.zeros((1, in_channels, 1, 1)), requires_grad=True)
self.gamma = nn.Parameter(
torch.zeros((1, in_channels, 1, 1)), requires_grad=True)
[docs] def forward(self, inp):
"""Forward Function.
Args:
inp: input tensor image
"""
x = inp
# part 1
x = self.norm1(x)
x = self.conv1(x)
x = self.conv2(x)
x = self.sg(x)
x = x * self.sca(x)
x = self.conv3(x)
x = self.dropout1(x)
y = inp + x * self.beta
# part 2
x = self.norm2(y)
x = self.conv4(x)
x = self.sg(x)
x = self.conv5(x)
x = self.dropout2(x)
out = y + x * self.gamma
return out
[docs]class SimpleGate(BaseModule):
"""The Simple Gate in "Simple Baseline for Image Restoration".
Args:
x: input tensor feature map with (B, 2 * C, H, W)
Return:
x1 * x2
(where x1, x2 are two separate parts by simple split x to [B, C, H, W])
"""