IOPaint/lama_cleaner/model/mat.py

1945 lines
61 KiB
Python
Raw Normal View History

2022-08-22 17:24:02 +02:00
import os
import random
import cv2
2022-09-02 04:37:30 +02:00
import numpy as np
2022-08-22 17:24:02 +02:00
import torch
import torch.nn as nn
import torch.nn.functional as F
2022-09-02 04:37:30 +02:00
import torch.utils.checkpoint as checkpoint
2022-08-22 17:24:02 +02:00
2023-11-16 14:12:06 +01:00
from lama_cleaner.helper import (
load_model,
get_cache_path_by_url,
norm_img,
download_model,
)
2022-08-22 17:24:02 +02:00
from lama_cleaner.model.base import InpaintModel
2023-02-26 02:19:48 +01:00
from lama_cleaner.model.utils import (
setup_filter,
Conv2dLayer,
FullyConnectedLayer,
conv2d_resample,
bias_act,
upsample2d,
activation_funcs,
MinibatchStdLayer,
to_2tuple,
normalize_2nd_moment,
2023-03-25 14:29:13 +01:00
set_seed,
2023-02-26 02:19:48 +01:00
)
2022-08-22 17:24:02 +02:00
from lama_cleaner.schema import Config
class ModulatedConv2d(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
in_channels, # Number of input channels.
out_channels, # Number of output channels.
kernel_size, # Width and height of the convolution kernel.
style_dim, # dimension of the style code
demodulate=True, # perfrom demodulation
up=1, # Integer upsampling factor.
down=1, # Integer downsampling factor.
resample_filter=[
1,
3,
3,
1,
], # Low-pass filter to apply when resampling activations.
conv_clamp=None, # Clamp the output to +-X, None = disable clamping.
):
2022-08-22 17:24:02 +02:00
super().__init__()
self.demodulate = demodulate
2023-02-26 02:19:48 +01:00
self.weight = torch.nn.Parameter(
torch.randn([1, out_channels, in_channels, kernel_size, kernel_size])
)
2022-08-22 17:24:02 +02:00
self.out_channels = out_channels
self.kernel_size = kernel_size
2023-11-16 14:12:06 +01:00
self.weight_gain = 1 / np.sqrt(in_channels * (kernel_size**2))
2022-08-22 17:24:02 +02:00
self.padding = self.kernel_size // 2
self.up = up
self.down = down
2023-02-26 02:19:48 +01:00
self.register_buffer("resample_filter", setup_filter(resample_filter))
2022-08-22 17:24:02 +02:00
self.conv_clamp = conv_clamp
self.affine = FullyConnectedLayer(style_dim, in_channels, bias_init=1)
def forward(self, x, style):
batch, in_channels, height, width = x.shape
style = self.affine(style).view(batch, 1, in_channels, 1, 1)
weight = self.weight * self.weight_gain * style
if self.demodulate:
decoefs = (weight.pow(2).sum(dim=[2, 3, 4]) + 1e-8).rsqrt()
weight = weight * decoefs.view(batch, self.out_channels, 1, 1, 1)
2023-02-26 02:19:48 +01:00
weight = weight.view(
batch * self.out_channels, in_channels, self.kernel_size, self.kernel_size
)
2022-08-22 17:24:02 +02:00
x = x.view(1, batch * in_channels, height, width)
2023-02-26 02:19:48 +01:00
x = conv2d_resample(
x=x,
w=weight,
f=self.resample_filter,
up=self.up,
down=self.down,
padding=self.padding,
groups=batch,
)
2022-08-22 17:24:02 +02:00
out = x.view(batch, self.out_channels, *x.shape[2:])
return out
class StyleConv(torch.nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
in_channels, # Number of input channels.
out_channels, # Number of output channels.
style_dim, # Intermediate latent (W) dimensionality.
resolution, # Resolution of this layer.
kernel_size=3, # Convolution kernel size.
up=1, # Integer upsampling factor.
use_noise=False, # Enable noise input?
activation="lrelu", # Activation function: 'relu', 'lrelu', etc.
resample_filter=[
1,
3,
3,
1,
], # Low-pass filter to apply when resampling activations.
conv_clamp=None, # Clamp the output of convolution layers to +-X, None = disable clamping.
demodulate=True, # perform demodulation
):
2022-08-22 17:24:02 +02:00
super().__init__()
2023-02-26 02:19:48 +01:00
self.conv = ModulatedConv2d(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=kernel_size,
style_dim=style_dim,
demodulate=demodulate,
up=up,
resample_filter=resample_filter,
conv_clamp=conv_clamp,
)
2022-08-22 17:24:02 +02:00
self.use_noise = use_noise
self.resolution = resolution
if use_noise:
2023-02-26 02:19:48 +01:00
self.register_buffer("noise_const", torch.randn([resolution, resolution]))
2022-08-22 17:24:02 +02:00
self.noise_strength = torch.nn.Parameter(torch.zeros([]))
self.bias = torch.nn.Parameter(torch.zeros([out_channels]))
self.activation = activation
self.act_gain = activation_funcs[activation].def_gain
self.conv_clamp = conv_clamp
2023-02-26 02:19:48 +01:00
def forward(self, x, style, noise_mode="random", gain=1):
2022-08-22 17:24:02 +02:00
x = self.conv(x, style)
2023-02-26 02:19:48 +01:00
assert noise_mode in ["random", "const", "none"]
2022-08-22 17:24:02 +02:00
if self.use_noise:
2023-02-26 02:19:48 +01:00
if noise_mode == "random":
2022-08-22 17:24:02 +02:00
xh, xw = x.size()[-2:]
2023-02-26 02:19:48 +01:00
noise = (
torch.randn([x.shape[0], 1, xh, xw], device=x.device)
* self.noise_strength
)
if noise_mode == "const":
2022-08-22 17:24:02 +02:00
noise = self.noise_const * self.noise_strength
x = x + noise
act_gain = self.act_gain * gain
act_clamp = self.conv_clamp * gain if self.conv_clamp is not None else None
2023-02-26 02:19:48 +01:00
out = bias_act(
x, self.bias, act=self.activation, gain=act_gain, clamp=act_clamp
)
2022-08-22 17:24:02 +02:00
return out
class ToRGB(torch.nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
in_channels,
out_channels,
style_dim,
kernel_size=1,
resample_filter=[1, 3, 3, 1],
conv_clamp=None,
demodulate=False,
):
2022-08-22 17:24:02 +02:00
super().__init__()
2023-02-26 02:19:48 +01:00
self.conv = ModulatedConv2d(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=kernel_size,
style_dim=style_dim,
demodulate=demodulate,
resample_filter=resample_filter,
conv_clamp=conv_clamp,
)
2022-08-22 17:24:02 +02:00
self.bias = torch.nn.Parameter(torch.zeros([out_channels]))
2023-02-26 02:19:48 +01:00
self.register_buffer("resample_filter", setup_filter(resample_filter))
2022-08-22 17:24:02 +02:00
self.conv_clamp = conv_clamp
def forward(self, x, style, skip=None):
x = self.conv(x, style)
out = bias_act(x, self.bias, clamp=self.conv_clamp)
if skip is not None:
if skip.shape != out.shape:
skip = upsample2d(skip, self.resample_filter)
out = out + skip
return out
def get_style_code(a, b):
return torch.cat([a, b], dim=1)
class DecBlockFirst(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
in_channels,
out_channels,
activation,
style_dim,
use_noise,
demodulate,
img_channels,
):
2022-08-22 17:24:02 +02:00
super().__init__()
2023-02-26 02:19:48 +01:00
self.fc = FullyConnectedLayer(
in_features=in_channels * 2,
2023-11-16 14:12:06 +01:00
out_features=in_channels * 4**2,
2023-02-26 02:19:48 +01:00
activation=activation,
)
self.conv = StyleConv(
in_channels=in_channels,
out_channels=out_channels,
style_dim=style_dim,
resolution=4,
kernel_size=3,
use_noise=use_noise,
activation=activation,
demodulate=demodulate,
)
self.toRGB = ToRGB(
in_channels=out_channels,
out_channels=img_channels,
style_dim=style_dim,
kernel_size=1,
demodulate=False,
)
def forward(self, x, ws, gs, E_features, noise_mode="random"):
2022-08-22 17:24:02 +02:00
x = self.fc(x).view(x.shape[0], -1, 4, 4)
x = x + E_features[2]
style = get_style_code(ws[:, 0], gs)
x = self.conv(x, style, noise_mode=noise_mode)
style = get_style_code(ws[:, 1], gs)
img = self.toRGB(x, style, skip=None)
return x, img
class DecBlockFirstV2(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
in_channels,
out_channels,
activation,
style_dim,
use_noise,
demodulate,
img_channels,
):
2022-08-22 17:24:02 +02:00
super().__init__()
2023-02-26 02:19:48 +01:00
self.conv0 = Conv2dLayer(
in_channels=in_channels,
out_channels=in_channels,
kernel_size=3,
activation=activation,
)
self.conv1 = StyleConv(
in_channels=in_channels,
out_channels=out_channels,
style_dim=style_dim,
resolution=4,
kernel_size=3,
use_noise=use_noise,
activation=activation,
demodulate=demodulate,
)
self.toRGB = ToRGB(
in_channels=out_channels,
out_channels=img_channels,
style_dim=style_dim,
kernel_size=1,
demodulate=False,
)
def forward(self, x, ws, gs, E_features, noise_mode="random"):
2022-08-22 17:24:02 +02:00
# x = self.fc(x).view(x.shape[0], -1, 4, 4)
x = self.conv0(x)
x = x + E_features[2]
style = get_style_code(ws[:, 0], gs)
x = self.conv1(x, style, noise_mode=noise_mode)
style = get_style_code(ws[:, 1], gs)
img = self.toRGB(x, style, skip=None)
return x, img
class DecBlock(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
res,
in_channels,
out_channels,
activation,
style_dim,
use_noise,
demodulate,
img_channels,
): # res = 2, ..., resolution_log2
2022-08-22 17:24:02 +02:00
super().__init__()
self.res = res
2023-02-26 02:19:48 +01:00
self.conv0 = StyleConv(
in_channels=in_channels,
out_channels=out_channels,
style_dim=style_dim,
2023-11-16 14:12:06 +01:00
resolution=2**res,
2023-02-26 02:19:48 +01:00
kernel_size=3,
up=2,
use_noise=use_noise,
activation=activation,
demodulate=demodulate,
)
self.conv1 = StyleConv(
in_channels=out_channels,
out_channels=out_channels,
style_dim=style_dim,
2023-11-16 14:12:06 +01:00
resolution=2**res,
2023-02-26 02:19:48 +01:00
kernel_size=3,
use_noise=use_noise,
activation=activation,
demodulate=demodulate,
)
self.toRGB = ToRGB(
in_channels=out_channels,
out_channels=img_channels,
style_dim=style_dim,
kernel_size=1,
demodulate=False,
)
def forward(self, x, img, ws, gs, E_features, noise_mode="random"):
2022-08-22 17:24:02 +02:00
style = get_style_code(ws[:, self.res * 2 - 5], gs)
x = self.conv0(x, style, noise_mode=noise_mode)
x = x + E_features[self.res]
style = get_style_code(ws[:, self.res * 2 - 4], gs)
x = self.conv1(x, style, noise_mode=noise_mode)
style = get_style_code(ws[:, self.res * 2 - 3], gs)
img = self.toRGB(x, style, skip=img)
return x, img
class MappingNet(torch.nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
z_dim, # Input latent (Z) dimensionality, 0 = no latent.
c_dim, # Conditioning label (C) dimensionality, 0 = no label.
w_dim, # Intermediate latent (W) dimensionality.
num_ws, # Number of intermediate latents to output, None = do not broadcast.
num_layers=8, # Number of mapping layers.
embed_features=None, # Label embedding dimensionality, None = same as w_dim.
layer_features=None, # Number of intermediate features in the mapping layers, None = same as w_dim.
activation="lrelu", # Activation function: 'relu', 'lrelu', etc.
lr_multiplier=0.01, # Learning rate multiplier for the mapping layers.
w_avg_beta=0.995, # Decay for tracking the moving average of W during training, None = do not track.
2023-03-25 14:29:13 +01:00
torch_dtype=torch.float32,
2023-02-26 02:19:48 +01:00
):
2022-08-22 17:24:02 +02:00
super().__init__()
self.z_dim = z_dim
self.c_dim = c_dim
self.w_dim = w_dim
self.num_ws = num_ws
self.num_layers = num_layers
self.w_avg_beta = w_avg_beta
2023-03-25 14:29:13 +01:00
self.torch_dtype = torch_dtype
2022-08-22 17:24:02 +02:00
if embed_features is None:
embed_features = w_dim
if c_dim == 0:
embed_features = 0
if layer_features is None:
layer_features = w_dim
2023-02-26 02:19:48 +01:00
features_list = (
[z_dim + embed_features] + [layer_features] * (num_layers - 1) + [w_dim]
)
2022-08-22 17:24:02 +02:00
if c_dim > 0:
self.embed = FullyConnectedLayer(c_dim, embed_features)
for idx in range(num_layers):
in_features = features_list[idx]
out_features = features_list[idx + 1]
2023-02-26 02:19:48 +01:00
layer = FullyConnectedLayer(
in_features,
out_features,
activation=activation,
lr_multiplier=lr_multiplier,
)
setattr(self, f"fc{idx}", layer)
2022-08-22 17:24:02 +02:00
if num_ws is not None and w_avg_beta is not None:
2023-02-26 02:19:48 +01:00
self.register_buffer("w_avg", torch.zeros([w_dim]))
2022-08-22 17:24:02 +02:00
2023-02-26 02:19:48 +01:00
def forward(
self, z, c, truncation_psi=1, truncation_cutoff=None, skip_w_avg_update=False
):
2022-08-22 17:24:02 +02:00
# Embed, normalize, and concat inputs.
x = None
2023-03-25 14:29:13 +01:00
if self.z_dim > 0:
x = normalize_2nd_moment(z)
if self.c_dim > 0:
y = normalize_2nd_moment(self.embed(c))
x = torch.cat([x, y], dim=1) if x is not None else y
2022-08-22 17:24:02 +02:00
# Main layers.
for idx in range(self.num_layers):
2023-02-26 02:19:48 +01:00
layer = getattr(self, f"fc{idx}")
2022-08-22 17:24:02 +02:00
x = layer(x)
# Update moving average of W.
if self.w_avg_beta is not None and self.training and not skip_w_avg_update:
2023-03-25 14:29:13 +01:00
self.w_avg.copy_(x.detach().mean(dim=0).lerp(self.w_avg, self.w_avg_beta))
2022-08-22 17:24:02 +02:00
# Broadcast.
if self.num_ws is not None:
2023-03-25 14:29:13 +01:00
x = x.unsqueeze(1).repeat([1, self.num_ws, 1])
2022-08-22 17:24:02 +02:00
# Apply truncation.
if truncation_psi != 1:
2023-03-25 14:29:13 +01:00
assert self.w_avg_beta is not None
if self.num_ws is None or truncation_cutoff is None:
x = self.w_avg.lerp(x, truncation_psi)
else:
x[:, :truncation_cutoff] = self.w_avg.lerp(
x[:, :truncation_cutoff], truncation_psi
)
2022-08-22 17:24:02 +02:00
return x
class DisFromRGB(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self, in_channels, out_channels, activation
): # res = 2, ..., resolution_log2
2022-08-22 17:24:02 +02:00
super().__init__()
2023-02-26 02:19:48 +01:00
self.conv = Conv2dLayer(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=1,
activation=activation,
)
2022-08-22 17:24:02 +02:00
def forward(self, x):
return self.conv(x)
class DisBlock(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self, in_channels, out_channels, activation
): # res = 2, ..., resolution_log2
2022-08-22 17:24:02 +02:00
super().__init__()
2023-02-26 02:19:48 +01:00
self.conv0 = Conv2dLayer(
in_channels=in_channels,
out_channels=in_channels,
kernel_size=3,
activation=activation,
)
self.conv1 = Conv2dLayer(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=3,
down=2,
activation=activation,
)
self.skip = Conv2dLayer(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=1,
down=2,
bias=False,
)
2022-08-22 17:24:02 +02:00
def forward(self, x):
skip = self.skip(x, gain=np.sqrt(0.5))
x = self.conv0(x)
x = self.conv1(x, gain=np.sqrt(0.5))
out = skip + x
return out
class Discriminator(torch.nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
c_dim, # Conditioning label (C) dimensionality.
img_resolution, # Input resolution.
img_channels, # Number of input color channels.
channel_base=32768, # Overall multiplier for the number of channels.
channel_max=512, # Maximum number of channels in any layer.
channel_decay=1,
cmap_dim=None, # Dimensionality of mapped conditioning label, None = default.
activation="lrelu",
mbstd_group_size=4, # Group size for the minibatch standard deviation layer, None = entire minibatch.
mbstd_num_channels=1, # Number of features for the minibatch standard deviation layer, 0 = disable.
):
2022-08-22 17:24:02 +02:00
super().__init__()
self.c_dim = c_dim
self.img_resolution = img_resolution
self.img_channels = img_channels
resolution_log2 = int(np.log2(img_resolution))
2023-11-16 14:12:06 +01:00
assert img_resolution == 2**resolution_log2 and img_resolution >= 4
2022-08-22 17:24:02 +02:00
self.resolution_log2 = resolution_log2
def nf(stage):
2023-02-26 02:19:48 +01:00
return np.clip(
int(channel_base / 2 ** (stage * channel_decay)), 1, channel_max
)
2022-08-22 17:24:02 +02:00
if cmap_dim == None:
cmap_dim = nf(2)
if c_dim == 0:
cmap_dim = 0
self.cmap_dim = cmap_dim
if c_dim > 0:
2023-02-26 02:19:48 +01:00
self.mapping = MappingNet(
z_dim=0, c_dim=c_dim, w_dim=cmap_dim, num_ws=None, w_avg_beta=None
)
2022-08-22 17:24:02 +02:00
Dis = [DisFromRGB(img_channels + 1, nf(resolution_log2), activation)]
for res in range(resolution_log2, 2, -1):
Dis.append(DisBlock(nf(res), nf(res - 1), activation))
if mbstd_num_channels > 0:
2023-02-26 02:19:48 +01:00
Dis.append(
MinibatchStdLayer(
group_size=mbstd_group_size, num_channels=mbstd_num_channels
)
)
Dis.append(
Conv2dLayer(
nf(2) + mbstd_num_channels, nf(2), kernel_size=3, activation=activation
)
)
2022-08-22 17:24:02 +02:00
self.Dis = nn.Sequential(*Dis)
2023-11-16 14:12:06 +01:00
self.fc0 = FullyConnectedLayer(nf(2) * 4**2, nf(2), activation=activation)
2022-08-22 17:24:02 +02:00
self.fc1 = FullyConnectedLayer(nf(2), 1 if cmap_dim == 0 else cmap_dim)
def forward(self, images_in, masks_in, c):
x = torch.cat([masks_in - 0.5, images_in], dim=1)
x = self.Dis(x)
x = self.fc1(self.fc0(x.flatten(start_dim=1)))
if self.c_dim > 0:
cmap = self.mapping(None, c)
if self.cmap_dim > 0:
x = (x * cmap).sum(dim=1, keepdim=True) * (1 / np.sqrt(self.cmap_dim))
return x
def nf(stage, channel_base=32768, channel_decay=1.0, channel_max=512):
NF = {512: 64, 256: 128, 128: 256, 64: 512, 32: 512, 16: 512, 8: 512, 4: 512}
2023-11-16 14:12:06 +01:00
return NF[2**stage]
2022-08-22 17:24:02 +02:00
class Mlp(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
in_features,
hidden_features=None,
out_features=None,
act_layer=nn.GELU,
drop=0.0,
):
2022-08-22 17:24:02 +02:00
super().__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
2023-02-26 02:19:48 +01:00
self.fc1 = FullyConnectedLayer(
in_features=in_features, out_features=hidden_features, activation="lrelu"
)
self.fc2 = FullyConnectedLayer(
in_features=hidden_features, out_features=out_features
)
2022-08-22 17:24:02 +02:00
def forward(self, x):
x = self.fc1(x)
x = self.fc2(x)
return x
def window_partition(x, window_size):
"""
Args:
x: (B, H, W, C)
window_size (int): window size
Returns:
windows: (num_windows*B, window_size, window_size, C)
"""
B, H, W, C = x.shape
x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)
2023-02-26 02:19:48 +01:00
windows = (
x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
)
2022-08-22 17:24:02 +02:00
return windows
def window_reverse(windows, window_size: int, H: int, W: int):
"""
Args:
windows: (num_windows*B, window_size, window_size, C)
window_size (int): Window size
H (int): Height of image
W (int): Width of image
Returns:
x: (B, H, W, C)
"""
B = int(windows.shape[0] / (H * W / window_size / window_size))
# B = windows.shape[0] / (H * W / window_size / window_size)
2023-02-26 02:19:48 +01:00
x = windows.view(
B, H // window_size, W // window_size, window_size, window_size, -1
)
2022-08-22 17:24:02 +02:00
x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
return x
class Conv2dLayerPartial(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
in_channels, # Number of input channels.
out_channels, # Number of output channels.
kernel_size, # Width and height of the convolution kernel.
bias=True, # Apply additive bias before the activation function?
activation="linear", # Activation function: 'relu', 'lrelu', etc.
up=1, # Integer upsampling factor.
down=1, # Integer downsampling factor.
resample_filter=[
1,
3,
3,
1,
], # Low-pass filter to apply when resampling activations.
conv_clamp=None, # Clamp the output to +-X, None = disable clamping.
trainable=True, # Update the weights of this layer during training?
):
2022-08-22 17:24:02 +02:00
super().__init__()
2023-02-26 02:19:48 +01:00
self.conv = Conv2dLayer(
in_channels,
out_channels,
kernel_size,
bias,
activation,
up,
down,
resample_filter,
conv_clamp,
trainable,
)
2022-08-22 17:24:02 +02:00
self.weight_maskUpdater = torch.ones(1, 1, kernel_size, kernel_size)
2023-11-16 14:12:06 +01:00
self.slide_winsize = kernel_size**2
2022-08-22 17:24:02 +02:00
self.stride = down
self.padding = kernel_size // 2 if kernel_size % 2 == 1 else 0
def forward(self, x, mask=None):
if mask is not None:
with torch.no_grad():
if self.weight_maskUpdater.type() != x.type():
self.weight_maskUpdater = self.weight_maskUpdater.to(x)
2023-02-26 02:19:48 +01:00
update_mask = F.conv2d(
mask,
self.weight_maskUpdater,
bias=None,
stride=self.stride,
padding=self.padding,
)
2023-03-25 15:46:28 +01:00
mask_ratio = self.slide_winsize / (update_mask.to(torch.float32) + 1e-8)
2022-08-22 17:24:02 +02:00
update_mask = torch.clamp(update_mask, 0, 1) # 0 or 1
2023-03-25 15:46:28 +01:00
mask_ratio = torch.mul(mask_ratio, update_mask).to(x.dtype)
2022-08-22 17:24:02 +02:00
x = self.conv(x)
x = torch.mul(x, mask_ratio)
return x, update_mask
else:
x = self.conv(x)
return x, None
class WindowAttention(nn.Module):
2023-02-26 02:19:48 +01:00
r"""Window based multi-head self attention (W-MSA) module with relative position bias.
2022-08-22 17:24:02 +02:00
It supports both of shifted and non-shifted window.
Args:
dim (int): Number of input channels.
window_size (tuple[int]): The height and width of the window.
num_heads (int): Number of attention heads.
qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set
attn_drop (float, optional): Dropout ratio of attention weight. Default: 0.0
proj_drop (float, optional): Dropout ratio of output. Default: 0.0
"""
2023-02-26 02:19:48 +01:00
def __init__(
self,
dim,
window_size,
num_heads,
down_ratio=1,
qkv_bias=True,
qk_scale=None,
attn_drop=0.0,
proj_drop=0.0,
):
2022-08-22 17:24:02 +02:00
super().__init__()
self.dim = dim
self.window_size = window_size # Wh, Ww
self.num_heads = num_heads
head_dim = dim // num_heads
2023-11-16 14:12:06 +01:00
self.scale = qk_scale or head_dim**-0.5
2022-08-22 17:24:02 +02:00
self.q = FullyConnectedLayer(in_features=dim, out_features=dim)
self.k = FullyConnectedLayer(in_features=dim, out_features=dim)
self.v = FullyConnectedLayer(in_features=dim, out_features=dim)
self.proj = FullyConnectedLayer(in_features=dim, out_features=dim)
self.softmax = nn.Softmax(dim=-1)
def forward(self, x, mask_windows=None, mask=None):
"""
Args:
x: input features with shape of (num_windows*B, N, C)
mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None
"""
B_, N, C = x.shape
2023-03-25 15:46:28 +01:00
norm_x = F.normalize(x, p=2.0, dim=-1, eps=torch.finfo(x.dtype).eps)
2023-02-26 02:19:48 +01:00
q = (
self.q(norm_x)
.reshape(B_, N, self.num_heads, C // self.num_heads)
.permute(0, 2, 1, 3)
)
k = (
self.k(norm_x)
.view(B_, -1, self.num_heads, C // self.num_heads)
.permute(0, 2, 3, 1)
)
v = (
self.v(x)
.view(B_, -1, self.num_heads, C // self.num_heads)
.permute(0, 2, 1, 3)
)
2022-08-22 17:24:02 +02:00
attn = (q @ k) * self.scale
if mask is not None:
nW = mask.shape[0]
2023-02-26 02:19:48 +01:00
attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(
1
).unsqueeze(0)
2022-08-22 17:24:02 +02:00
attn = attn.view(-1, self.num_heads, N, N)
if mask_windows is not None:
attn_mask_windows = mask_windows.squeeze(-1).unsqueeze(1).unsqueeze(1)
2023-02-26 02:19:48 +01:00
attn = attn + attn_mask_windows.masked_fill(
attn_mask_windows == 0, float(-100.0)
).masked_fill(attn_mask_windows == 1, float(0.0))
2022-08-22 17:24:02 +02:00
with torch.no_grad():
2023-02-26 02:19:48 +01:00
mask_windows = torch.clamp(
torch.sum(mask_windows, dim=1, keepdim=True), 0, 1
).repeat(1, N, 1)
2022-08-22 17:24:02 +02:00
attn = self.softmax(attn)
x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
x = self.proj(x)
return x, mask_windows
class SwinTransformerBlock(nn.Module):
2023-02-26 02:19:48 +01:00
r"""Swin Transformer Block.
2022-08-22 17:24:02 +02:00
Args:
dim (int): Number of input channels.
input_resolution (tuple[int]): Input resulotion.
num_heads (int): Number of attention heads.
window_size (int): Window size.
shift_size (int): Shift size for SW-MSA.
mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
drop (float, optional): Dropout rate. Default: 0.0
attn_drop (float, optional): Attention dropout rate. Default: 0.0
drop_path (float, optional): Stochastic depth rate. Default: 0.0
act_layer (nn.Module, optional): Activation layer. Default: nn.GELU
norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
"""
2023-02-26 02:19:48 +01:00
def __init__(
self,
dim,
input_resolution,
num_heads,
down_ratio=1,
window_size=7,
shift_size=0,
mlp_ratio=4.0,
qkv_bias=True,
qk_scale=None,
drop=0.0,
attn_drop=0.0,
drop_path=0.0,
act_layer=nn.GELU,
norm_layer=nn.LayerNorm,
):
2022-08-22 17:24:02 +02:00
super().__init__()
self.dim = dim
self.input_resolution = input_resolution
self.num_heads = num_heads
self.window_size = window_size
self.shift_size = shift_size
self.mlp_ratio = mlp_ratio
if min(self.input_resolution) <= self.window_size:
# if window size is larger than input resolution, we don't partition windows
self.shift_size = 0
self.window_size = min(self.input_resolution)
2023-02-26 02:19:48 +01:00
assert (
0 <= self.shift_size < self.window_size
), "shift_size must in 0-window_size"
2022-08-22 17:24:02 +02:00
if self.shift_size > 0:
down_ratio = 1
2023-02-26 02:19:48 +01:00
self.attn = WindowAttention(
dim,
window_size=to_2tuple(self.window_size),
num_heads=num_heads,
down_ratio=down_ratio,
qkv_bias=qkv_bias,
qk_scale=qk_scale,
attn_drop=attn_drop,
proj_drop=drop,
)
2022-08-22 17:24:02 +02:00
2023-02-26 02:19:48 +01:00
self.fuse = FullyConnectedLayer(
in_features=dim * 2, out_features=dim, activation="lrelu"
)
2022-08-22 17:24:02 +02:00
mlp_hidden_dim = int(dim * mlp_ratio)
2023-02-26 02:19:48 +01:00
self.mlp = Mlp(
in_features=dim,
hidden_features=mlp_hidden_dim,
act_layer=act_layer,
drop=drop,
)
2022-08-22 17:24:02 +02:00
if self.shift_size > 0:
attn_mask = self.calculate_mask(self.input_resolution)
else:
attn_mask = None
self.register_buffer("attn_mask", attn_mask)
def calculate_mask(self, x_size):
# calculate attention mask for SW-MSA
H, W = x_size
img_mask = torch.zeros((1, H, W, 1)) # 1 H W 1
2023-02-26 02:19:48 +01:00
h_slices = (
slice(0, -self.window_size),
slice(-self.window_size, -self.shift_size),
slice(-self.shift_size, None),
)
w_slices = (
slice(0, -self.window_size),
slice(-self.window_size, -self.shift_size),
slice(-self.shift_size, None),
)
2022-08-22 17:24:02 +02:00
cnt = 0
for h in h_slices:
for w in w_slices:
img_mask[:, h, w, :] = cnt
cnt += 1
2023-02-26 02:19:48 +01:00
mask_windows = window_partition(
img_mask, self.window_size
) # nW, window_size, window_size, 1
2022-08-22 17:24:02 +02:00
mask_windows = mask_windows.view(-1, self.window_size * self.window_size)
attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
2023-02-26 02:19:48 +01:00
attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(
attn_mask == 0, float(0.0)
)
2022-08-22 17:24:02 +02:00
return attn_mask
def forward(self, x, x_size, mask=None):
# H, W = self.input_resolution
H, W = x_size
B, L, C = x.shape
# assert L == H * W, "input feature has wrong size"
shortcut = x
x = x.view(B, H, W, C)
if mask is not None:
mask = mask.view(B, H, W, 1)
# cyclic shift
if self.shift_size > 0:
2023-02-26 02:19:48 +01:00
shifted_x = torch.roll(
x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2)
)
2022-08-22 17:24:02 +02:00
if mask is not None:
2023-02-26 02:19:48 +01:00
shifted_mask = torch.roll(
mask, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2)
)
2022-08-22 17:24:02 +02:00
else:
shifted_x = x
if mask is not None:
shifted_mask = mask
# partition windows
2023-02-26 02:19:48 +01:00
x_windows = window_partition(
shifted_x, self.window_size
) # nW*B, window_size, window_size, C
x_windows = x_windows.view(
-1, self.window_size * self.window_size, C
) # nW*B, window_size*window_size, C
2022-08-22 17:24:02 +02:00
if mask is not None:
mask_windows = window_partition(shifted_mask, self.window_size)
mask_windows = mask_windows.view(-1, self.window_size * self.window_size, 1)
else:
mask_windows = None
# W-MSA/SW-MSA (to be compatible for testing on images whose shapes are the multiple of window size
if self.input_resolution == x_size:
2023-02-26 02:19:48 +01:00
attn_windows, mask_windows = self.attn(
x_windows, mask_windows, mask=self.attn_mask
) # nW*B, window_size*window_size, C
2022-08-22 17:24:02 +02:00
else:
2023-02-26 02:19:48 +01:00
attn_windows, mask_windows = self.attn(
2023-03-25 15:46:28 +01:00
x_windows,
mask_windows,
mask=self.calculate_mask(x_size).to(x.dtype).to(x.device),
2023-02-26 02:19:48 +01:00
) # nW*B, window_size*window_size, C
2022-08-22 17:24:02 +02:00
# merge windows
attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)
shifted_x = window_reverse(attn_windows, self.window_size, H, W) # B H' W' C
if mask is not None:
mask_windows = mask_windows.view(-1, self.window_size, self.window_size, 1)
shifted_mask = window_reverse(mask_windows, self.window_size, H, W)
# reverse cyclic shift
if self.shift_size > 0:
2023-02-26 02:19:48 +01:00
x = torch.roll(
shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2)
)
2022-08-22 17:24:02 +02:00
if mask is not None:
2023-02-26 02:19:48 +01:00
mask = torch.roll(
shifted_mask, shifts=(self.shift_size, self.shift_size), dims=(1, 2)
)
2022-08-22 17:24:02 +02:00
else:
x = shifted_x
if mask is not None:
mask = shifted_mask
x = x.view(B, H * W, C)
if mask is not None:
mask = mask.view(B, H * W, 1)
# FFN
x = self.fuse(torch.cat([shortcut, x], dim=-1))
x = self.mlp(x)
return x, mask
class PatchMerging(nn.Module):
def __init__(self, in_channels, out_channels, down=2):
super().__init__()
2023-02-26 02:19:48 +01:00
self.conv = Conv2dLayerPartial(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=3,
activation="lrelu",
down=down,
)
2022-08-22 17:24:02 +02:00
self.down = down
def forward(self, x, x_size, mask=None):
x = token2feature(x, x_size)
if mask is not None:
mask = token2feature(mask, x_size)
x, mask = self.conv(x, mask)
if self.down != 1:
ratio = 1 / self.down
x_size = (int(x_size[0] * ratio), int(x_size[1] * ratio))
x = feature2token(x)
if mask is not None:
mask = feature2token(mask)
return x, x_size, mask
class PatchUpsampling(nn.Module):
def __init__(self, in_channels, out_channels, up=2):
super().__init__()
2023-02-26 02:19:48 +01:00
self.conv = Conv2dLayerPartial(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=3,
activation="lrelu",
up=up,
)
2022-08-22 17:24:02 +02:00
self.up = up
def forward(self, x, x_size, mask=None):
x = token2feature(x, x_size)
if mask is not None:
mask = token2feature(mask, x_size)
x, mask = self.conv(x, mask)
if self.up != 1:
x_size = (int(x_size[0] * self.up), int(x_size[1] * self.up))
x = feature2token(x)
if mask is not None:
mask = feature2token(mask)
return x, x_size, mask
class BasicLayer(nn.Module):
2023-02-26 02:19:48 +01:00
"""A basic Swin Transformer layer for one stage.
2022-08-22 17:24:02 +02:00
Args:
dim (int): Number of input channels.
input_resolution (tuple[int]): Input resolution.
depth (int): Number of blocks.
num_heads (int): Number of attention heads.
window_size (int): Local window size.
mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
drop (float, optional): Dropout rate. Default: 0.0
attn_drop (float, optional): Attention dropout rate. Default: 0.0
drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0
norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None
use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False.
"""
2023-02-26 02:19:48 +01:00
def __init__(
self,
dim,
input_resolution,
depth,
num_heads,
window_size,
down_ratio=1,
mlp_ratio=2.0,
qkv_bias=True,
qk_scale=None,
drop=0.0,
attn_drop=0.0,
drop_path=0.0,
norm_layer=nn.LayerNorm,
downsample=None,
use_checkpoint=False,
):
2022-08-22 17:24:02 +02:00
super().__init__()
self.dim = dim
self.input_resolution = input_resolution
self.depth = depth
self.use_checkpoint = use_checkpoint
# patch merging layer
if downsample is not None:
# self.downsample = downsample(input_resolution, dim=dim, norm_layer=norm_layer)
self.downsample = downsample
else:
self.downsample = None
# build blocks
2023-02-26 02:19:48 +01:00
self.blocks = nn.ModuleList(
[
SwinTransformerBlock(
dim=dim,
input_resolution=input_resolution,
num_heads=num_heads,
down_ratio=down_ratio,
window_size=window_size,
shift_size=0 if (i % 2 == 0) else window_size // 2,
mlp_ratio=mlp_ratio,
qkv_bias=qkv_bias,
qk_scale=qk_scale,
drop=drop,
attn_drop=attn_drop,
drop_path=drop_path[i]
if isinstance(drop_path, list)
else drop_path,
norm_layer=norm_layer,
)
for i in range(depth)
]
)
self.conv = Conv2dLayerPartial(
in_channels=dim, out_channels=dim, kernel_size=3, activation="lrelu"
)
2022-08-22 17:24:02 +02:00
def forward(self, x, x_size, mask=None):
if self.downsample is not None:
x, x_size, mask = self.downsample(x, x_size, mask)
identity = x
for blk in self.blocks:
if self.use_checkpoint:
x, mask = checkpoint.checkpoint(blk, x, x_size, mask)
else:
x, mask = blk(x, x_size, mask)
if mask is not None:
mask = token2feature(mask, x_size)
x, mask = self.conv(token2feature(x, x_size), mask)
x = feature2token(x) + identity
if mask is not None:
mask = feature2token(mask)
return x, x_size, mask
class ToToken(nn.Module):
def __init__(self, in_channels=3, dim=128, kernel_size=5, stride=1):
super().__init__()
2023-02-26 02:19:48 +01:00
self.proj = Conv2dLayerPartial(
in_channels=in_channels,
out_channels=dim,
kernel_size=kernel_size,
activation="lrelu",
)
2022-08-22 17:24:02 +02:00
def forward(self, x, mask):
x, mask = self.proj(x, mask)
return x, mask
class EncFromRGB(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self, in_channels, out_channels, activation
): # res = 2, ..., resolution_log2
2022-08-22 17:24:02 +02:00
super().__init__()
2023-02-26 02:19:48 +01:00
self.conv0 = Conv2dLayer(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=1,
activation=activation,
)
self.conv1 = Conv2dLayer(
in_channels=out_channels,
out_channels=out_channels,
kernel_size=3,
activation=activation,
)
2022-08-22 17:24:02 +02:00
def forward(self, x):
x = self.conv0(x)
x = self.conv1(x)
return x
class ConvBlockDown(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self, in_channels, out_channels, activation
): # res = 2, ..., resolution_log
2022-08-22 17:24:02 +02:00
super().__init__()
2023-02-26 02:19:48 +01:00
self.conv0 = Conv2dLayer(
in_channels=in_channels,
out_channels=out_channels,
kernel_size=3,
activation=activation,
down=2,
)
self.conv1 = Conv2dLayer(
in_channels=out_channels,
out_channels=out_channels,
kernel_size=3,
activation=activation,
)
2022-08-22 17:24:02 +02:00
def forward(self, x):
x = self.conv0(x)
x = self.conv1(x)
return x
def token2feature(x, x_size):
B, N, C = x.shape
h, w = x_size
x = x.permute(0, 2, 1).reshape(B, C, h, w)
return x
def feature2token(x):
B, C, H, W = x.shape
x = x.view(B, C, -1).transpose(1, 2)
return x
class Encoder(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
res_log2,
img_channels,
activation,
patch_size=5,
channels=16,
drop_path_rate=0.1,
):
2022-08-22 17:24:02 +02:00
super().__init__()
self.resolution = []
for idx, i in enumerate(range(res_log2, 3, -1)): # from input size to 16x16
2023-11-16 14:12:06 +01:00
res = 2**i
2022-08-22 17:24:02 +02:00
self.resolution.append(res)
if i == res_log2:
block = EncFromRGB(img_channels * 2 + 1, nf(i), activation)
else:
block = ConvBlockDown(nf(i + 1), nf(i), activation)
2023-02-26 02:19:48 +01:00
setattr(self, "EncConv_Block_%dx%d" % (res, res), block)
2022-08-22 17:24:02 +02:00
def forward(self, x):
out = {}
for res in self.resolution:
res_log2 = int(np.log2(res))
2023-02-26 02:19:48 +01:00
x = getattr(self, "EncConv_Block_%dx%d" % (res, res))(x)
2022-08-22 17:24:02 +02:00
out[res_log2] = x
return out
class ToStyle(nn.Module):
def __init__(self, in_channels, out_channels, activation, drop_rate):
super().__init__()
self.conv = nn.Sequential(
2023-02-26 02:19:48 +01:00
Conv2dLayer(
in_channels=in_channels,
out_channels=in_channels,
kernel_size=3,
activation=activation,
down=2,
),
Conv2dLayer(
in_channels=in_channels,
out_channels=in_channels,
kernel_size=3,
activation=activation,
down=2,
),
Conv2dLayer(
in_channels=in_channels,
out_channels=in_channels,
kernel_size=3,
activation=activation,
down=2,
),
2022-08-22 17:24:02 +02:00
)
self.pool = nn.AdaptiveAvgPool2d(1)
2023-02-26 02:19:48 +01:00
self.fc = FullyConnectedLayer(
in_features=in_channels, out_features=out_channels, activation=activation
)
2022-08-22 17:24:02 +02:00
# self.dropout = nn.Dropout(drop_rate)
def forward(self, x):
x = self.conv(x)
x = self.pool(x)
x = self.fc(x.flatten(start_dim=1))
# x = self.dropout(x)
return x
class DecBlockFirstV2(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
res,
in_channels,
out_channels,
activation,
style_dim,
use_noise,
demodulate,
img_channels,
):
2022-08-22 17:24:02 +02:00
super().__init__()
self.res = res
2023-02-26 02:19:48 +01:00
self.conv0 = Conv2dLayer(
in_channels=in_channels,
out_channels=in_channels,
kernel_size=3,
activation=activation,
)
self.conv1 = StyleConv(
in_channels=in_channels,
out_channels=out_channels,
style_dim=style_dim,
2023-11-16 14:12:06 +01:00
resolution=2**res,
2023-02-26 02:19:48 +01:00
kernel_size=3,
use_noise=use_noise,
activation=activation,
demodulate=demodulate,
)
self.toRGB = ToRGB(
in_channels=out_channels,
out_channels=img_channels,
style_dim=style_dim,
kernel_size=1,
demodulate=False,
)
def forward(self, x, ws, gs, E_features, noise_mode="random"):
2022-08-22 17:24:02 +02:00
# x = self.fc(x).view(x.shape[0], -1, 4, 4)
x = self.conv0(x)
x = x + E_features[self.res]
style = get_style_code(ws[:, 0], gs)
x = self.conv1(x, style, noise_mode=noise_mode)
style = get_style_code(ws[:, 1], gs)
img = self.toRGB(x, style, skip=None)
return x, img
class DecBlock(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
res,
in_channels,
out_channels,
activation,
style_dim,
use_noise,
demodulate,
img_channels,
): # res = 4, ..., resolution_log2
2022-08-22 17:24:02 +02:00
super().__init__()
self.res = res
2023-02-26 02:19:48 +01:00
self.conv0 = StyleConv(
in_channels=in_channels,
out_channels=out_channels,
style_dim=style_dim,
2023-11-16 14:12:06 +01:00
resolution=2**res,
2023-02-26 02:19:48 +01:00
kernel_size=3,
up=2,
use_noise=use_noise,
activation=activation,
demodulate=demodulate,
)
self.conv1 = StyleConv(
in_channels=out_channels,
out_channels=out_channels,
style_dim=style_dim,
2023-11-16 14:12:06 +01:00
resolution=2**res,
2023-02-26 02:19:48 +01:00
kernel_size=3,
use_noise=use_noise,
activation=activation,
demodulate=demodulate,
)
self.toRGB = ToRGB(
in_channels=out_channels,
out_channels=img_channels,
style_dim=style_dim,
kernel_size=1,
demodulate=False,
)
def forward(self, x, img, ws, gs, E_features, noise_mode="random"):
2022-08-22 17:24:02 +02:00
style = get_style_code(ws[:, self.res * 2 - 9], gs)
x = self.conv0(x, style, noise_mode=noise_mode)
x = x + E_features[self.res]
style = get_style_code(ws[:, self.res * 2 - 8], gs)
x = self.conv1(x, style, noise_mode=noise_mode)
style = get_style_code(ws[:, self.res * 2 - 7], gs)
img = self.toRGB(x, style, skip=img)
return x, img
class Decoder(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self, res_log2, activation, style_dim, use_noise, demodulate, img_channels
):
2022-08-22 17:24:02 +02:00
super().__init__()
2023-02-26 02:19:48 +01:00
self.Dec_16x16 = DecBlockFirstV2(
4, nf(4), nf(4), activation, style_dim, use_noise, demodulate, img_channels
)
2022-08-22 17:24:02 +02:00
for res in range(5, res_log2 + 1):
2023-02-26 02:19:48 +01:00
setattr(
self,
2023-11-16 14:12:06 +01:00
"Dec_%dx%d" % (2**res, 2**res),
2023-02-26 02:19:48 +01:00
DecBlock(
res,
nf(res - 1),
nf(res),
activation,
style_dim,
use_noise,
demodulate,
img_channels,
),
)
2022-08-22 17:24:02 +02:00
self.res_log2 = res_log2
2023-02-26 02:19:48 +01:00
def forward(self, x, ws, gs, E_features, noise_mode="random"):
2022-08-22 17:24:02 +02:00
x, img = self.Dec_16x16(x, ws, gs, E_features, noise_mode=noise_mode)
for res in range(5, self.res_log2 + 1):
2023-11-16 14:12:06 +01:00
block = getattr(self, "Dec_%dx%d" % (2**res, 2**res))
2022-08-22 17:24:02 +02:00
x, img = block(x, img, ws, gs, E_features, noise_mode=noise_mode)
return img
class DecStyleBlock(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
res,
in_channels,
out_channels,
activation,
style_dim,
use_noise,
demodulate,
img_channels,
):
2022-08-22 17:24:02 +02:00
super().__init__()
self.res = res
2023-02-26 02:19:48 +01:00
self.conv0 = StyleConv(
in_channels=in_channels,
out_channels=out_channels,
style_dim=style_dim,
2023-11-16 14:12:06 +01:00
resolution=2**res,
2023-02-26 02:19:48 +01:00
kernel_size=3,
up=2,
use_noise=use_noise,
activation=activation,
demodulate=demodulate,
)
self.conv1 = StyleConv(
in_channels=out_channels,
out_channels=out_channels,
style_dim=style_dim,
2023-11-16 14:12:06 +01:00
resolution=2**res,
2023-02-26 02:19:48 +01:00
kernel_size=3,
use_noise=use_noise,
activation=activation,
demodulate=demodulate,
)
self.toRGB = ToRGB(
in_channels=out_channels,
out_channels=img_channels,
style_dim=style_dim,
kernel_size=1,
demodulate=False,
)
def forward(self, x, img, style, skip, noise_mode="random"):
2022-08-22 17:24:02 +02:00
x = self.conv0(x, style, noise_mode=noise_mode)
x = x + skip
x = self.conv1(x, style, noise_mode=noise_mode)
img = self.toRGB(x, style, skip=img)
return x, img
class FirstStage(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
img_channels,
img_resolution=256,
dim=180,
w_dim=512,
use_noise=False,
demodulate=True,
activation="lrelu",
):
2022-08-22 17:24:02 +02:00
super().__init__()
res = 64
2023-02-26 02:19:48 +01:00
self.conv_first = Conv2dLayerPartial(
in_channels=img_channels + 1,
out_channels=dim,
kernel_size=3,
activation=activation,
)
2022-08-22 17:24:02 +02:00
self.enc_conv = nn.ModuleList()
down_time = int(np.log2(img_resolution // res))
# 根据图片尺寸构建 swim transformer 的层数
for i in range(down_time): # from input size to 64
self.enc_conv.append(
2023-02-26 02:19:48 +01:00
Conv2dLayerPartial(
in_channels=dim,
out_channels=dim,
kernel_size=3,
down=2,
activation=activation,
)
2022-08-22 17:24:02 +02:00
)
# from 64 -> 16 -> 64
depths = [2, 3, 4, 3, 2]
ratios = [1, 1 / 2, 1 / 2, 2, 2]
num_heads = 6
window_sizes = [8, 16, 16, 16, 8]
drop_path_rate = 0.1
dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))]
self.tran = nn.ModuleList()
for i, depth in enumerate(depths):
res = int(res * ratios[i])
if ratios[i] < 1:
merge = PatchMerging(dim, dim, down=int(1 / ratios[i]))
elif ratios[i] > 1:
merge = PatchUpsampling(dim, dim, up=ratios[i])
else:
merge = None
self.tran.append(
2023-02-26 02:19:48 +01:00
BasicLayer(
dim=dim,
input_resolution=[res, res],
depth=depth,
num_heads=num_heads,
window_size=window_sizes[i],
drop_path=dpr[sum(depths[:i]) : sum(depths[: i + 1])],
downsample=merge,
)
2022-08-22 17:24:02 +02:00
)
# global style
down_conv = []
for i in range(int(np.log2(16))):
down_conv.append(
2023-02-26 02:19:48 +01:00
Conv2dLayer(
in_channels=dim,
out_channels=dim,
kernel_size=3,
down=2,
activation=activation,
)
)
2022-08-22 17:24:02 +02:00
down_conv.append(nn.AdaptiveAvgPool2d((1, 1)))
self.down_conv = nn.Sequential(*down_conv)
2023-02-26 02:19:48 +01:00
self.to_style = FullyConnectedLayer(
in_features=dim, out_features=dim * 2, activation=activation
)
self.ws_style = FullyConnectedLayer(
in_features=w_dim, out_features=dim, activation=activation
)
self.to_square = FullyConnectedLayer(
in_features=dim, out_features=16 * 16, activation=activation
)
2022-08-22 17:24:02 +02:00
style_dim = dim * 3
self.dec_conv = nn.ModuleList()
for i in range(down_time): # from 64 to input size
res = res * 2
self.dec_conv.append(
2023-02-26 02:19:48 +01:00
DecStyleBlock(
res,
dim,
dim,
activation,
style_dim,
use_noise,
demodulate,
img_channels,
)
)
2022-08-22 17:24:02 +02:00
2023-02-26 02:19:48 +01:00
def forward(self, images_in, masks_in, ws, noise_mode="random"):
2022-08-22 17:24:02 +02:00
x = torch.cat([masks_in - 0.5, images_in * masks_in], dim=1)
skips = []
x, mask = self.conv_first(x, masks_in) # input size
skips.append(x)
for i, block in enumerate(self.enc_conv): # input size to 64
x, mask = block(x, mask)
if i != len(self.enc_conv) - 1:
skips.append(x)
x_size = x.size()[-2:]
x = feature2token(x)
mask = feature2token(mask)
mid = len(self.tran) // 2
for i, block in enumerate(self.tran): # 64 to 16
if i < mid:
x, x_size, mask = block(x, x_size, mask)
skips.append(x)
elif i > mid:
x, x_size, mask = block(x, x_size, None)
x = x + skips[mid - i]
else:
x, x_size, mask = block(x, x_size, None)
mul_map = torch.ones_like(x) * 0.5
mul_map = F.dropout(mul_map, training=True)
ws = self.ws_style(ws[:, -1])
add_n = self.to_square(ws).unsqueeze(1)
2023-02-26 02:19:48 +01:00
add_n = (
F.interpolate(
add_n, size=x.size(1), mode="linear", align_corners=False
)
.squeeze(1)
.unsqueeze(-1)
)
2022-08-22 17:24:02 +02:00
x = x * mul_map + add_n * (1 - mul_map)
2023-02-26 02:19:48 +01:00
gs = self.to_style(
self.down_conv(token2feature(x, x_size)).flatten(start_dim=1)
)
2022-08-22 17:24:02 +02:00
style = torch.cat([gs, ws], dim=1)
x = token2feature(x, x_size).contiguous()
img = None
for i, block in enumerate(self.dec_conv):
2023-02-26 02:19:48 +01:00
x, img = block(
x, img, style, skips[len(self.dec_conv) - i - 1], noise_mode=noise_mode
)
2022-08-22 17:24:02 +02:00
# ensemble
img = img * (1 - masks_in) + images_in * masks_in
return img
class SynthesisNet(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
w_dim, # Intermediate latent (W) dimensionality.
img_resolution, # Output image resolution.
img_channels=3, # Number of color channels.
channel_base=32768, # Overall multiplier for the number of channels.
channel_decay=1.0,
channel_max=512, # Maximum number of channels in any layer.
activation="lrelu", # Activation function: 'relu', 'lrelu', etc.
drop_rate=0.5,
use_noise=False,
demodulate=True,
):
2022-08-22 17:24:02 +02:00
super().__init__()
resolution_log2 = int(np.log2(img_resolution))
2023-11-16 14:12:06 +01:00
assert img_resolution == 2**resolution_log2 and img_resolution >= 4
2022-08-22 17:24:02 +02:00
self.num_layers = resolution_log2 * 2 - 3 * 2
self.img_resolution = img_resolution
self.resolution_log2 = resolution_log2
# first stage
2023-02-26 02:19:48 +01:00
self.first_stage = FirstStage(
img_channels,
img_resolution=img_resolution,
w_dim=w_dim,
use_noise=False,
demodulate=demodulate,
)
2022-08-22 17:24:02 +02:00
# second stage
2023-02-26 02:19:48 +01:00
self.enc = Encoder(
resolution_log2, img_channels, activation, patch_size=5, channels=16
)
self.to_square = FullyConnectedLayer(
in_features=w_dim, out_features=16 * 16, activation=activation
)
self.to_style = ToStyle(
in_channels=nf(4),
out_channels=nf(2) * 2,
activation=activation,
drop_rate=drop_rate,
)
2022-08-22 17:24:02 +02:00
style_dim = w_dim + nf(2) * 2
2023-02-26 02:19:48 +01:00
self.dec = Decoder(
resolution_log2, activation, style_dim, use_noise, demodulate, img_channels
)
2022-08-22 17:24:02 +02:00
2023-02-26 02:19:48 +01:00
def forward(self, images_in, masks_in, ws, noise_mode="random", return_stg1=False):
2022-08-22 17:24:02 +02:00
out_stg1 = self.first_stage(images_in, masks_in, ws, noise_mode=noise_mode)
# encoder
x = images_in * masks_in + out_stg1 * (1 - masks_in)
x = torch.cat([masks_in - 0.5, x, images_in * masks_in], dim=1)
E_features = self.enc(x)
fea_16 = E_features[4]
mul_map = torch.ones_like(fea_16) * 0.5
mul_map = F.dropout(mul_map, training=True)
add_n = self.to_square(ws[:, 0]).view(-1, 16, 16).unsqueeze(1)
2023-02-26 02:19:48 +01:00
add_n = F.interpolate(
add_n, size=fea_16.size()[-2:], mode="bilinear", align_corners=False
)
2022-08-22 17:24:02 +02:00
fea_16 = fea_16 * mul_map + add_n * (1 - mul_map)
E_features[4] = fea_16
# style
gs = self.to_style(fea_16)
# decoder
img = self.dec(fea_16, ws, gs, E_features, noise_mode=noise_mode)
# ensemble
img = img * (1 - masks_in) + images_in * masks_in
if not return_stg1:
return img
else:
return img, out_stg1
class Generator(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
z_dim, # Input latent (Z) dimensionality, 0 = no latent.
c_dim, # Conditioning label (C) dimensionality, 0 = no label.
w_dim, # Intermediate latent (W) dimensionality.
img_resolution, # resolution of generated image
img_channels, # Number of input color channels.
synthesis_kwargs={}, # Arguments for SynthesisNetwork.
mapping_kwargs={}, # Arguments for MappingNetwork.
):
2022-08-22 17:24:02 +02:00
super().__init__()
self.z_dim = z_dim
self.c_dim = c_dim
self.w_dim = w_dim
self.img_resolution = img_resolution
self.img_channels = img_channels
2023-02-26 02:19:48 +01:00
self.synthesis = SynthesisNet(
w_dim=w_dim,
img_resolution=img_resolution,
img_channels=img_channels,
**synthesis_kwargs,
)
self.mapping = MappingNet(
z_dim=z_dim,
c_dim=c_dim,
w_dim=w_dim,
num_ws=self.synthesis.num_layers,
**mapping_kwargs,
)
def forward(
self,
images_in,
masks_in,
z,
c,
truncation_psi=1,
truncation_cutoff=None,
skip_w_avg_update=False,
noise_mode="none",
return_stg1=False,
):
ws = self.mapping(
z,
c,
truncation_psi=truncation_psi,
truncation_cutoff=truncation_cutoff,
skip_w_avg_update=skip_w_avg_update,
)
2022-08-22 17:24:02 +02:00
img = self.synthesis(images_in, masks_in, ws, noise_mode=noise_mode)
return img
class Discriminator(torch.nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
c_dim, # Conditioning label (C) dimensionality.
img_resolution, # Input resolution.
img_channels, # Number of input color channels.
channel_base=32768, # Overall multiplier for the number of channels.
channel_max=512, # Maximum number of channels in any layer.
channel_decay=1,
cmap_dim=None, # Dimensionality of mapped conditioning label, None = default.
activation="lrelu",
mbstd_group_size=4, # Group size for the minibatch standard deviation layer, None = entire minibatch.
mbstd_num_channels=1, # Number of features for the minibatch standard deviation layer, 0 = disable.
):
2022-08-22 17:24:02 +02:00
super().__init__()
self.c_dim = c_dim
self.img_resolution = img_resolution
self.img_channels = img_channels
resolution_log2 = int(np.log2(img_resolution))
2023-11-16 14:12:06 +01:00
assert img_resolution == 2**resolution_log2 and img_resolution >= 4
2022-08-22 17:24:02 +02:00
self.resolution_log2 = resolution_log2
if cmap_dim == None:
cmap_dim = nf(2)
if c_dim == 0:
cmap_dim = 0
self.cmap_dim = cmap_dim
if c_dim > 0:
2023-02-26 02:19:48 +01:00
self.mapping = MappingNet(
z_dim=0, c_dim=c_dim, w_dim=cmap_dim, num_ws=None, w_avg_beta=None
)
2022-08-22 17:24:02 +02:00
Dis = [DisFromRGB(img_channels + 1, nf(resolution_log2), activation)]
for res in range(resolution_log2, 2, -1):
Dis.append(DisBlock(nf(res), nf(res - 1), activation))
if mbstd_num_channels > 0:
2023-02-26 02:19:48 +01:00
Dis.append(
MinibatchStdLayer(
group_size=mbstd_group_size, num_channels=mbstd_num_channels
)
)
Dis.append(
Conv2dLayer(
nf(2) + mbstd_num_channels, nf(2), kernel_size=3, activation=activation
)
)
2022-08-22 17:24:02 +02:00
self.Dis = nn.Sequential(*Dis)
2023-11-16 14:12:06 +01:00
self.fc0 = FullyConnectedLayer(nf(2) * 4**2, nf(2), activation=activation)
2022-08-22 17:24:02 +02:00
self.fc1 = FullyConnectedLayer(nf(2), 1 if cmap_dim == 0 else cmap_dim)
# for 64x64
Dis_stg1 = [DisFromRGB(img_channels + 1, nf(resolution_log2) // 2, activation)]
for res in range(resolution_log2, 2, -1):
Dis_stg1.append(DisBlock(nf(res) // 2, nf(res - 1) // 2, activation))
if mbstd_num_channels > 0:
2023-02-26 02:19:48 +01:00
Dis_stg1.append(
MinibatchStdLayer(
group_size=mbstd_group_size, num_channels=mbstd_num_channels
)
)
Dis_stg1.append(
Conv2dLayer(
nf(2) // 2 + mbstd_num_channels,
nf(2) // 2,
kernel_size=3,
activation=activation,
)
)
2022-08-22 17:24:02 +02:00
self.Dis_stg1 = nn.Sequential(*Dis_stg1)
2023-02-26 02:19:48 +01:00
self.fc0_stg1 = FullyConnectedLayer(
2023-11-16 14:12:06 +01:00
nf(2) // 2 * 4**2, nf(2) // 2, activation=activation
2023-02-26 02:19:48 +01:00
)
self.fc1_stg1 = FullyConnectedLayer(
nf(2) // 2, 1 if cmap_dim == 0 else cmap_dim
)
2022-08-22 17:24:02 +02:00
def forward(self, images_in, masks_in, images_stg1, c):
x = self.Dis(torch.cat([masks_in - 0.5, images_in], dim=1))
x = self.fc1(self.fc0(x.flatten(start_dim=1)))
x_stg1 = self.Dis_stg1(torch.cat([masks_in - 0.5, images_stg1], dim=1))
x_stg1 = self.fc1_stg1(self.fc0_stg1(x_stg1.flatten(start_dim=1)))
if self.c_dim > 0:
cmap = self.mapping(None, c)
if self.cmap_dim > 0:
x = (x * cmap).sum(dim=1, keepdim=True) * (1 / np.sqrt(self.cmap_dim))
2023-02-26 02:19:48 +01:00
x_stg1 = (x_stg1 * cmap).sum(dim=1, keepdim=True) * (
1 / np.sqrt(self.cmap_dim)
)
2022-08-22 17:24:02 +02:00
return x, x_stg1
MAT_MODEL_URL = os.environ.get(
"MAT_MODEL_URL",
"https://github.com/Sanster/models/releases/download/add_mat/Places_512_FullData_G.pth",
)
2023-02-26 02:19:48 +01:00
MAT_MODEL_MD5 = os.environ.get("MAT_MODEL_MD5", "8ca927835fa3f5e21d65ffcb165377ed")
2022-08-22 17:24:02 +02:00
class MAT(InpaintModel):
2023-02-11 06:30:09 +01:00
name = "mat"
2023-03-28 16:12:24 +02:00
min_size = 512
2022-08-22 17:24:02 +02:00
pad_mod = 512
pad_to_square = True
2022-09-15 16:21:27 +02:00
def init_model(self, device, **kwargs):
2022-08-22 17:24:02 +02:00
seed = 240 # pick up a random number
2023-03-25 14:29:13 +01:00
set_seed(seed)
2023-03-26 05:45:39 +02:00
fp16 = not kwargs.get("no_half", False)
use_gpu = "cuda" in str(device) and torch.cuda.is_available()
self.torch_dtype = torch.float16 if use_gpu and fp16 else torch.float32
2023-03-25 14:29:13 +01:00
G = Generator(
z_dim=512,
c_dim=0,
w_dim=512,
img_resolution=512,
img_channels=3,
mapping_kwargs={"torch_dtype": self.torch_dtype},
2023-03-25 15:46:28 +01:00
).to(self.torch_dtype)
2023-03-25 14:29:13 +01:00
# fmt: off
2023-03-25 15:46:28 +01:00
self.model = load_model(G, MAT_MODEL_URL, device, MAT_MODEL_MD5)
2023-03-25 14:29:13 +01:00
self.z = torch.from_numpy(np.random.randn(1, G.z_dim)).to(self.torch_dtype).to(device)
self.label = torch.zeros([1, self.model.c_dim], device=device).to(self.torch_dtype)
# fmt: on
2022-08-22 17:24:02 +02:00
2023-11-16 14:12:06 +01:00
@staticmethod
def download():
download_model(MAT_MODEL_URL, MAT_MODEL_MD5)
2022-08-22 17:24:02 +02:00
@staticmethod
def is_downloaded() -> bool:
return os.path.exists(get_cache_path_by_url(MAT_MODEL_URL))
def forward(self, image, mask, config: Config):
"""Input images and output images have same size
images: [H, W, C] RGB
masks: [H, W] mask area == 255
return: BGR IMAGE
"""
image = norm_img(image) # [0, 1]
image = image * 2 - 1 # [0, 1] -> [-1, 1]
mask = (mask > 127) * 255
mask = 255 - mask
mask = norm_img(mask)
2023-03-25 14:29:13 +01:00
image = (
torch.from_numpy(image).unsqueeze(0).to(self.torch_dtype).to(self.device)
)
mask = torch.from_numpy(mask).unsqueeze(0).to(self.torch_dtype).to(self.device)
2022-08-22 17:24:02 +02:00
2023-02-26 02:19:48 +01:00
output = self.model(
image, mask, self.z, self.label, truncation_psi=1, noise_mode="none"
)
output = (
(output.permute(0, 2, 3, 1) * 127.5 + 127.5)
.round()
.clamp(0, 255)
.to(torch.uint8)
)
2022-08-22 17:24:02 +02:00
output = output[0].cpu().numpy()
cur_res = cv2.cvtColor(output, cv2.COLOR_RGB2BGR)
return cur_res