IOPaint/lama_cleaner/model/fcf.py

1738 lines
56 KiB
Python
Raw Normal View History

2022-09-02 04:37:30 +02:00
import os
import random
import cv2
import torch
import numpy as np
import torch.fft as fft
2023-12-30 16:36:44 +01:00
from lama_cleaner.schema import InpaintRequest
2022-09-02 04:37:30 +02:00
2023-02-26 02:19:48 +01:00
from lama_cleaner.helper import (
load_model,
get_cache_path_by_url,
norm_img,
boxes_from_mask,
resize_max_size,
2023-11-16 14:12:06 +01:00
download_model,
2023-02-26 02:19:48 +01:00
)
2022-09-02 04:37:30 +02:00
from lama_cleaner.model.base import InpaintModel
from torch import conv2d, nn
import torch.nn.functional as F
2023-02-26 02:19:48 +01:00
from lama_cleaner.model.utils import (
setup_filter,
_parse_scaling,
_parse_padding,
Conv2dLayer,
FullyConnectedLayer,
MinibatchStdLayer,
activation_funcs,
conv2d_resample,
bias_act,
upsample2d,
normalize_2nd_moment,
downsample2d,
)
2022-09-02 04:37:30 +02:00
2023-02-26 02:19:48 +01:00
def upfirdn2d(x, f, up=1, down=1, padding=0, flip_filter=False, gain=1, impl="cuda"):
2022-09-02 04:37:30 +02:00
assert isinstance(x, torch.Tensor)
2023-02-26 02:19:48 +01:00
return _upfirdn2d_ref(
x, f, up=up, down=down, padding=padding, flip_filter=flip_filter, gain=gain
)
2022-09-02 04:37:30 +02:00
def _upfirdn2d_ref(x, f, up=1, down=1, padding=0, flip_filter=False, gain=1):
2023-02-26 02:19:48 +01:00
"""Slow reference implementation of `upfirdn2d()` using standard PyTorch ops."""
2022-09-02 04:37:30 +02:00
# Validate arguments.
assert isinstance(x, torch.Tensor) and x.ndim == 4
if f is None:
f = torch.ones([1, 1], dtype=torch.float32, device=x.device)
assert isinstance(f, torch.Tensor) and f.ndim in [1, 2]
assert f.dtype == torch.float32 and not f.requires_grad
batch_size, num_channels, in_height, in_width = x.shape
upx, upy = _parse_scaling(up)
downx, downy = _parse_scaling(down)
padx0, padx1, pady0, pady1 = _parse_padding(padding)
# Upsample by inserting zeros.
x = x.reshape([batch_size, num_channels, in_height, 1, in_width, 1])
x = torch.nn.functional.pad(x, [0, upx - 1, 0, 0, 0, upy - 1])
x = x.reshape([batch_size, num_channels, in_height * upy, in_width * upx])
# Pad or crop.
2023-02-26 02:19:48 +01:00
x = torch.nn.functional.pad(
x, [max(padx0, 0), max(padx1, 0), max(pady0, 0), max(pady1, 0)]
)
x = x[
:,
:,
max(-pady0, 0) : x.shape[2] - max(-pady1, 0),
max(-padx0, 0) : x.shape[3] - max(-padx1, 0),
]
2022-09-02 04:37:30 +02:00
# Setup filter.
f = f * (gain ** (f.ndim / 2))
f = f.to(x.dtype)
if not flip_filter:
f = f.flip(list(range(f.ndim)))
# Convolve with the filter.
f = f[np.newaxis, np.newaxis].repeat([num_channels, 1] + [1] * f.ndim)
if f.ndim == 4:
x = conv2d(input=x, weight=f, groups=num_channels)
else:
x = conv2d(input=x, weight=f.unsqueeze(2), groups=num_channels)
x = conv2d(input=x, weight=f.unsqueeze(3), groups=num_channels)
# Downsample by throwing away pixels.
x = x[:, :, ::downy, ::downx]
return x
class EncoderEpilogue(torch.nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
in_channels, # Number of input channels.
cmap_dim, # Dimensionality of mapped conditioning label, 0 = no label.
z_dim, # Output Latent (Z) dimensionality.
resolution, # Resolution of this block.
img_channels, # Number of input color channels.
architecture="resnet", # Architecture: 'orig', 'skip', 'resnet'.
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.
activation="lrelu", # Activation function: 'relu', 'lrelu', etc.
conv_clamp=None, # Clamp the output of convolution layers to +-X, None = disable clamping.
):
assert architecture in ["orig", "skip", "resnet"]
2022-09-02 04:37:30 +02:00
super().__init__()
self.in_channels = in_channels
self.cmap_dim = cmap_dim
self.resolution = resolution
self.img_channels = img_channels
self.architecture = architecture
2023-02-26 02:19:48 +01:00
if architecture == "skip":
self.fromrgb = Conv2dLayer(
self.img_channels, in_channels, kernel_size=1, activation=activation
)
self.mbstd = (
MinibatchStdLayer(
group_size=mbstd_group_size, num_channels=mbstd_num_channels
)
if mbstd_num_channels > 0
else None
)
self.conv = Conv2dLayer(
in_channels + mbstd_num_channels,
in_channels,
kernel_size=3,
activation=activation,
conv_clamp=conv_clamp,
)
self.fc = FullyConnectedLayer(
in_channels * (resolution**2), z_dim, activation=activation
)
2022-09-02 04:37:30 +02:00
self.dropout = torch.nn.Dropout(p=0.5)
def forward(self, x, cmap, force_fp32=False):
_ = force_fp32 # unused
dtype = torch.float32
memory_format = torch.contiguous_format
# FromRGB.
x = x.to(dtype=dtype, memory_format=memory_format)
# Main layers.
if self.mbstd is not None:
x = self.mbstd(x)
const_e = self.conv(x)
x = self.fc(const_e.flatten(1))
x = self.dropout(x)
# Conditioning.
if self.cmap_dim > 0:
x = (x * cmap).sum(dim=1, keepdim=True) * (1 / np.sqrt(self.cmap_dim))
assert x.dtype == dtype
return x, const_e
class EncoderBlock(torch.nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
in_channels, # Number of input channels, 0 = first block.
tmp_channels, # Number of intermediate channels.
out_channels, # Number of output channels.
resolution, # Resolution of this block.
img_channels, # Number of input color channels.
first_layer_idx, # Index of the first layer.
architecture="skip", # Architecture: 'orig', 'skip', 'resnet'.
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.
use_fp16=False, # Use FP16 for this block?
fp16_channels_last=False, # Use channels-last memory format with FP16?
freeze_layers=0, # Freeze-D: Number of layers to freeze.
):
2022-09-02 04:37:30 +02:00
assert in_channels in [0, tmp_channels]
2023-02-26 02:19:48 +01:00
assert architecture in ["orig", "skip", "resnet"]
2022-09-02 04:37:30 +02:00
super().__init__()
self.in_channels = in_channels
self.resolution = resolution
self.img_channels = img_channels + 1
self.first_layer_idx = first_layer_idx
self.architecture = architecture
self.use_fp16 = use_fp16
2023-02-26 02:19:48 +01:00
self.channels_last = use_fp16 and fp16_channels_last
self.register_buffer("resample_filter", setup_filter(resample_filter))
2022-09-02 04:37:30 +02:00
self.num_layers = 0
def trainable_gen():
while True:
layer_idx = self.first_layer_idx + self.num_layers
2023-02-26 02:19:48 +01:00
trainable = layer_idx >= freeze_layers
2022-09-02 04:37:30 +02:00
self.num_layers += 1
yield trainable
trainable_iter = trainable_gen()
if in_channels == 0:
2023-02-26 02:19:48 +01:00
self.fromrgb = Conv2dLayer(
self.img_channels,
tmp_channels,
kernel_size=1,
activation=activation,
trainable=next(trainable_iter),
conv_clamp=conv_clamp,
channels_last=self.channels_last,
)
self.conv0 = Conv2dLayer(
tmp_channels,
tmp_channels,
kernel_size=3,
activation=activation,
trainable=next(trainable_iter),
conv_clamp=conv_clamp,
channels_last=self.channels_last,
)
2022-09-02 04:37:30 +02:00
2023-02-26 02:19:48 +01:00
self.conv1 = Conv2dLayer(
tmp_channels,
out_channels,
kernel_size=3,
activation=activation,
down=2,
trainable=next(trainable_iter),
resample_filter=resample_filter,
conv_clamp=conv_clamp,
channels_last=self.channels_last,
)
2022-09-02 04:37:30 +02:00
2023-02-26 02:19:48 +01:00
if architecture == "resnet":
self.skip = Conv2dLayer(
tmp_channels,
out_channels,
kernel_size=1,
bias=False,
down=2,
trainable=next(trainable_iter),
resample_filter=resample_filter,
channels_last=self.channels_last,
)
2022-09-02 04:37:30 +02:00
def forward(self, x, img, force_fp32=False):
# dtype = torch.float16 if self.use_fp16 and not force_fp32 else torch.float32
dtype = torch.float32
2023-02-26 02:19:48 +01:00
memory_format = (
torch.channels_last
if self.channels_last and not force_fp32
else torch.contiguous_format
)
2022-09-02 04:37:30 +02:00
# Input.
if x is not None:
x = x.to(dtype=dtype, memory_format=memory_format)
# FromRGB.
if self.in_channels == 0:
img = img.to(dtype=dtype, memory_format=memory_format)
y = self.fromrgb(img)
x = x + y if x is not None else y
2023-02-26 02:19:48 +01:00
img = (
downsample2d(img, self.resample_filter)
if self.architecture == "skip"
else None
)
2022-09-02 04:37:30 +02:00
# Main layers.
2023-02-26 02:19:48 +01:00
if self.architecture == "resnet":
2022-09-02 04:37:30 +02:00
y = self.skip(x, gain=np.sqrt(0.5))
x = self.conv0(x)
feat = x.clone()
x = self.conv1(x, gain=np.sqrt(0.5))
x = y.add_(x)
else:
x = self.conv0(x)
feat = x.clone()
x = self.conv1(x)
assert x.dtype == dtype
return x, img, feat
class EncoderNetwork(torch.nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
c_dim, # Conditioning label (C) dimensionality.
z_dim, # Input latent (Z) dimensionality.
img_resolution, # Input resolution.
img_channels, # Number of input color channels.
architecture="orig", # Architecture: 'orig', 'skip', 'resnet'.
channel_base=16384, # Overall multiplier for the number of channels.
channel_max=512, # Maximum number of channels in any layer.
num_fp16_res=0, # Use FP16 for the N highest resolutions.
conv_clamp=None, # Clamp the output of convolution layers to +-X, None = disable clamping.
cmap_dim=None, # Dimensionality of mapped conditioning label, None = default.
block_kwargs={}, # Arguments for DiscriminatorBlock.
mapping_kwargs={}, # Arguments for MappingNetwork.
epilogue_kwargs={}, # Arguments for EncoderEpilogue.
):
2022-09-02 04:37:30 +02:00
super().__init__()
self.c_dim = c_dim
self.z_dim = z_dim
self.img_resolution = img_resolution
self.img_resolution_log2 = int(np.log2(img_resolution))
self.img_channels = img_channels
2023-02-26 02:19:48 +01:00
self.block_resolutions = [
2**i for i in range(self.img_resolution_log2, 2, -1)
]
channels_dict = {
res: min(channel_base // res, channel_max)
for res in self.block_resolutions + [4]
}
2022-09-02 04:37:30 +02:00
fp16_resolution = max(2 ** (self.img_resolution_log2 + 1 - num_fp16_res), 8)
if cmap_dim is None:
cmap_dim = channels_dict[4]
if c_dim == 0:
cmap_dim = 0
2023-02-26 02:19:48 +01:00
common_kwargs = dict(
img_channels=img_channels, architecture=architecture, conv_clamp=conv_clamp
)
2022-09-02 04:37:30 +02:00
cur_layer_idx = 0
for res in self.block_resolutions:
in_channels = channels_dict[res] if res < img_resolution else 0
tmp_channels = channels_dict[res]
out_channels = channels_dict[res // 2]
2023-02-26 02:19:48 +01:00
use_fp16 = res >= fp16_resolution
2022-09-02 04:37:30 +02:00
use_fp16 = False
2023-02-26 02:19:48 +01:00
block = EncoderBlock(
in_channels,
tmp_channels,
out_channels,
resolution=res,
first_layer_idx=cur_layer_idx,
use_fp16=use_fp16,
**block_kwargs,
**common_kwargs,
)
setattr(self, f"b{res}", block)
2022-09-02 04:37:30 +02:00
cur_layer_idx += block.num_layers
if c_dim > 0:
2023-02-26 02:19:48 +01:00
self.mapping = MappingNetwork(
z_dim=0,
c_dim=c_dim,
w_dim=cmap_dim,
num_ws=None,
w_avg_beta=None,
**mapping_kwargs,
)
self.b4 = EncoderEpilogue(
channels_dict[4],
cmap_dim=cmap_dim,
z_dim=z_dim * 2,
resolution=4,
**epilogue_kwargs,
**common_kwargs,
)
2022-09-02 04:37:30 +02:00
def forward(self, img, c, **block_kwargs):
x = None
feats = {}
for res in self.block_resolutions:
2023-02-26 02:19:48 +01:00
block = getattr(self, f"b{res}")
2022-09-02 04:37:30 +02:00
x, img, feat = block(x, img, **block_kwargs)
feats[res] = feat
cmap = None
if self.c_dim > 0:
cmap = self.mapping(None, c)
x, const_e = self.b4(x, cmap)
feats[4] = const_e
B, _ = x.shape
2023-02-26 02:19:48 +01:00
z = torch.zeros(
(B, self.z_dim), requires_grad=False, dtype=x.dtype, device=x.device
) ## Noise for Co-Modulation
2022-09-02 04:37:30 +02:00
return x, z, feats
def fma(a, b, c): # => a * b + c
return _FusedMultiplyAdd.apply(a, b, c)
class _FusedMultiplyAdd(torch.autograd.Function): # a * b + c
@staticmethod
def forward(ctx, a, b, c): # pylint: disable=arguments-differ
out = torch.addcmul(c, a, b)
ctx.save_for_backward(a, b)
ctx.c_shape = c.shape
return out
@staticmethod
def backward(ctx, dout): # pylint: disable=arguments-differ
a, b = ctx.saved_tensors
c_shape = ctx.c_shape
da = None
db = None
dc = None
if ctx.needs_input_grad[0]:
da = _unbroadcast(dout * b, a.shape)
if ctx.needs_input_grad[1]:
db = _unbroadcast(dout * a, b.shape)
if ctx.needs_input_grad[2]:
dc = _unbroadcast(dout, c_shape)
return da, db, dc
def _unbroadcast(x, shape):
extra_dims = x.ndim - len(shape)
assert extra_dims >= 0
2023-02-26 02:19:48 +01:00
dim = [
i
for i in range(x.ndim)
if x.shape[i] > 1 and (i < extra_dims or shape[i - extra_dims] == 1)
]
2022-09-02 04:37:30 +02:00
if len(dim):
x = x.sum(dim=dim, keepdim=True)
if extra_dims:
2023-02-26 02:19:48 +01:00
x = x.reshape(-1, *x.shape[extra_dims + 1 :])
2022-09-02 04:37:30 +02:00
assert x.shape == shape
return x
def modulated_conv2d(
x, # Input tensor of shape [batch_size, in_channels, in_height, in_width].
weight, # Weight tensor of shape [out_channels, in_channels, kernel_height, kernel_width].
styles, # Modulation coefficients of shape [batch_size, in_channels].
noise=None, # Optional noise tensor to add to the output activations.
up=1, # Integer upsampling factor.
down=1, # Integer downsampling factor.
padding=0, # Padding with respect to the upsampled image.
resample_filter=None,
# Low-pass filter to apply when resampling activations. Must be prepared beforehand by calling upfirdn2d.setup_filter().
demodulate=True, # Apply weight demodulation?
flip_weight=True, # False = convolution, True = correlation (matches torch.nn.functional.conv2d).
fused_modconv=True, # Perform modulation, convolution, and demodulation as a single fused operation?
):
batch_size = x.shape[0]
out_channels, in_channels, kh, kw = weight.shape
# Pre-normalize inputs to avoid FP16 overflow.
if x.dtype == torch.float16 and demodulate:
2023-02-26 02:19:48 +01:00
weight = weight * (
1
/ np.sqrt(in_channels * kh * kw)
/ weight.norm(float("inf"), dim=[1, 2, 3], keepdim=True)
) # max_Ikk
styles = styles / styles.norm(float("inf"), dim=1, keepdim=True) # max_I
2022-09-02 04:37:30 +02:00
# Calculate per-sample weights and demodulation coefficients.
w = None
dcoefs = None
if demodulate or fused_modconv:
w = weight.unsqueeze(0) # [NOIkk]
w = w * styles.reshape(batch_size, 1, -1, 1, 1) # [NOIkk]
if demodulate:
dcoefs = (w.square().sum(dim=[2, 3, 4]) + 1e-8).rsqrt() # [NO]
if demodulate and fused_modconv:
w = w * dcoefs.reshape(batch_size, -1, 1, 1, 1) # [NOIkk]
# Execute by scaling the activations before and after the convolution.
if not fused_modconv:
x = x * styles.to(x.dtype).reshape(batch_size, -1, 1, 1)
2023-02-26 02:19:48 +01:00
x = conv2d_resample.conv2d_resample(
x=x,
w=weight.to(x.dtype),
f=resample_filter,
up=up,
down=down,
padding=padding,
flip_weight=flip_weight,
)
2022-09-02 04:37:30 +02:00
if demodulate and noise is not None:
2023-02-26 02:19:48 +01:00
x = fma(
x, dcoefs.to(x.dtype).reshape(batch_size, -1, 1, 1), noise.to(x.dtype)
)
2022-09-02 04:37:30 +02:00
elif demodulate:
x = x * dcoefs.to(x.dtype).reshape(batch_size, -1, 1, 1)
elif noise is not None:
x = x.add_(noise.to(x.dtype))
return x
# Execute as one fused op using grouped convolution.
batch_size = int(batch_size)
x = x.reshape(1, -1, *x.shape[2:])
w = w.reshape(-1, in_channels, kh, kw)
2023-02-26 02:19:48 +01:00
x = conv2d_resample(
x=x,
w=w.to(x.dtype),
f=resample_filter,
up=up,
down=down,
padding=padding,
groups=batch_size,
flip_weight=flip_weight,
)
2022-09-02 04:37:30 +02:00
x = x.reshape(batch_size, -1, *x.shape[2:])
if noise is not None:
x = x.add_(noise)
return x
class SynthesisLayer(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.
w_dim, # Intermediate latent (W) dimensionality.
resolution, # Resolution of this layer.
kernel_size=3, # Convolution kernel size.
up=1, # Integer upsampling factor.
use_noise=True, # 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.
channels_last=False, # Use channels_last format for the weights?
):
2022-09-02 04:37:30 +02:00
super().__init__()
self.resolution = resolution
self.up = up
self.use_noise = use_noise
self.activation = activation
self.conv_clamp = conv_clamp
2023-02-26 02:19:48 +01:00
self.register_buffer("resample_filter", setup_filter(resample_filter))
2022-09-02 04:37:30 +02:00
self.padding = kernel_size // 2
self.act_gain = activation_funcs[activation].def_gain
self.affine = FullyConnectedLayer(w_dim, in_channels, bias_init=1)
2023-02-26 02:19:48 +01:00
memory_format = (
torch.channels_last if channels_last else torch.contiguous_format
)
2022-09-02 04:37:30 +02:00
self.weight = torch.nn.Parameter(
2023-02-26 02:19:48 +01:00
torch.randn([out_channels, in_channels, kernel_size, kernel_size]).to(
memory_format=memory_format
)
)
2022-09-02 04:37:30 +02:00
if use_noise:
2023-02-26 02:19:48 +01:00
self.register_buffer("noise_const", torch.randn([resolution, resolution]))
2022-09-02 04:37:30 +02:00
self.noise_strength = torch.nn.Parameter(torch.zeros([]))
self.bias = torch.nn.Parameter(torch.zeros([out_channels]))
2023-02-26 02:19:48 +01:00
def forward(self, x, w, noise_mode="none", fused_modconv=True, gain=1):
assert noise_mode in ["random", "const", "none"]
2022-09-02 04:37:30 +02:00
in_resolution = self.resolution // self.up
styles = self.affine(w)
noise = None
2023-02-26 02:19:48 +01:00
if self.use_noise and noise_mode == "random":
noise = (
torch.randn(
[x.shape[0], 1, self.resolution, self.resolution], device=x.device
)
* self.noise_strength
)
if self.use_noise and noise_mode == "const":
2022-09-02 04:37:30 +02:00
noise = self.noise_const * self.noise_strength
2023-02-26 02:19:48 +01:00
flip_weight = self.up == 1 # slightly faster
x = modulated_conv2d(
x=x,
weight=self.weight,
styles=styles,
noise=noise,
up=self.up,
padding=self.padding,
resample_filter=self.resample_filter,
flip_weight=flip_weight,
fused_modconv=fused_modconv,
)
2022-09-02 04:37:30 +02:00
act_gain = self.act_gain * gain
act_clamp = self.conv_clamp * gain if self.conv_clamp is not None else None
x = F.leaky_relu(x, negative_slope=0.2, inplace=False)
if act_gain != 1:
x = x * act_gain
if act_clamp is not None:
x = x.clamp(-act_clamp, act_clamp)
return x
class ToRGBLayer(torch.nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
in_channels,
out_channels,
w_dim,
kernel_size=1,
conv_clamp=None,
channels_last=False,
):
2022-09-02 04:37:30 +02:00
super().__init__()
self.conv_clamp = conv_clamp
self.affine = FullyConnectedLayer(w_dim, in_channels, bias_init=1)
2023-02-26 02:19:48 +01:00
memory_format = (
torch.channels_last if channels_last else torch.contiguous_format
)
2022-09-02 04:37:30 +02:00
self.weight = torch.nn.Parameter(
2023-02-26 02:19:48 +01:00
torch.randn([out_channels, in_channels, kernel_size, kernel_size]).to(
memory_format=memory_format
)
)
2022-09-02 04:37:30 +02:00
self.bias = torch.nn.Parameter(torch.zeros([out_channels]))
2023-02-26 02:19:48 +01:00
self.weight_gain = 1 / np.sqrt(in_channels * (kernel_size**2))
2022-09-02 04:37:30 +02:00
def forward(self, x, w, fused_modconv=True):
styles = self.affine(w) * self.weight_gain
2023-02-26 02:19:48 +01:00
x = modulated_conv2d(
x=x,
weight=self.weight,
styles=styles,
demodulate=False,
fused_modconv=fused_modconv,
)
2022-09-02 04:37:30 +02:00
x = bias_act(x, self.bias.to(x.dtype), clamp=self.conv_clamp)
return x
class SynthesisForeword(torch.nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
z_dim, # Output Latent (Z) dimensionality.
resolution, # Resolution of this block.
in_channels,
img_channels, # Number of input color channels.
architecture="skip", # Architecture: 'orig', 'skip', 'resnet'.
activation="lrelu", # Activation function: 'relu', 'lrelu', etc.
):
2022-09-02 04:37:30 +02:00
super().__init__()
self.in_channels = in_channels
self.z_dim = z_dim
self.resolution = resolution
self.img_channels = img_channels
self.architecture = architecture
2023-02-26 02:19:48 +01:00
self.fc = FullyConnectedLayer(
self.z_dim, (self.z_dim // 2) * 4 * 4, activation=activation
)
self.conv = SynthesisLayer(
self.in_channels, self.in_channels, w_dim=(z_dim // 2) * 3, resolution=4
)
2022-09-02 04:37:30 +02:00
2023-02-26 02:19:48 +01:00
if architecture == "skip":
self.torgb = ToRGBLayer(
self.in_channels,
self.img_channels,
kernel_size=1,
w_dim=(z_dim // 2) * 3,
)
2022-09-02 04:37:30 +02:00
def forward(self, x, ws, feats, img, force_fp32=False):
_ = force_fp32 # unused
dtype = torch.float32
memory_format = torch.contiguous_format
x_global = x.clone()
# ToRGB.
x = self.fc(x)
x = x.view(-1, self.z_dim // 2, 4, 4)
x = x.to(dtype=dtype, memory_format=memory_format)
# Main layers.
x_skip = feats[4].clone()
x = x + x_skip
mod_vector = []
mod_vector.append(ws[:, 0])
mod_vector.append(x_global.clone())
mod_vector = torch.cat(mod_vector, dim=1)
x = self.conv(x, mod_vector)
mod_vector = []
mod_vector.append(ws[:, 2 * 2 - 3])
mod_vector.append(x_global.clone())
mod_vector = torch.cat(mod_vector, dim=1)
2023-02-26 02:19:48 +01:00
if self.architecture == "skip":
2022-09-02 04:37:30 +02:00
img = self.torgb(x, mod_vector)
img = img.to(dtype=torch.float32, memory_format=torch.contiguous_format)
assert x.dtype == dtype
return x, img
class SELayer(nn.Module):
def __init__(self, channel, reduction=16):
super(SELayer, self).__init__()
self.avg_pool = nn.AdaptiveAvgPool2d(1)
self.fc = nn.Sequential(
nn.Linear(channel, channel // reduction, bias=False),
nn.ReLU(inplace=False),
nn.Linear(channel // reduction, channel, bias=False),
2023-02-26 02:19:48 +01:00
nn.Sigmoid(),
2022-09-02 04:37:30 +02:00
)
def forward(self, x):
b, c, _, _ = x.size()
y = self.avg_pool(x).view(b, c)
y = self.fc(y).view(b, c, 1, 1)
res = x * y.expand_as(x)
return res
class FourierUnit(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
in_channels,
out_channels,
groups=1,
spatial_scale_factor=None,
spatial_scale_mode="bilinear",
spectral_pos_encoding=False,
use_se=False,
se_kwargs=None,
ffc3d=False,
fft_norm="ortho",
):
2022-09-02 04:37:30 +02:00
# bn_layer not used
super(FourierUnit, self).__init__()
self.groups = groups
2023-02-26 02:19:48 +01:00
self.conv_layer = torch.nn.Conv2d(
in_channels=in_channels * 2 + (2 if spectral_pos_encoding else 0),
out_channels=out_channels * 2,
kernel_size=1,
stride=1,
padding=0,
groups=self.groups,
bias=False,
)
2022-09-02 04:37:30 +02:00
self.relu = torch.nn.ReLU(inplace=False)
# squeeze and excitation block
self.use_se = use_se
if use_se:
if se_kwargs is None:
se_kwargs = {}
self.se = SELayer(self.conv_layer.in_channels, **se_kwargs)
self.spatial_scale_factor = spatial_scale_factor
self.spatial_scale_mode = spatial_scale_mode
self.spectral_pos_encoding = spectral_pos_encoding
self.ffc3d = ffc3d
self.fft_norm = fft_norm
def forward(self, x):
batch = x.shape[0]
if self.spatial_scale_factor is not None:
orig_size = x.shape[-2:]
2023-02-26 02:19:48 +01:00
x = F.interpolate(
x,
scale_factor=self.spatial_scale_factor,
mode=self.spatial_scale_mode,
align_corners=False,
)
2022-09-02 04:37:30 +02:00
r_size = x.size()
# (batch, c, h, w/2+1, 2)
fft_dim = (-3, -2, -1) if self.ffc3d else (-2, -1)
ffted = fft.rfftn(x, dim=fft_dim, norm=self.fft_norm)
ffted = torch.stack((ffted.real, ffted.imag), dim=-1)
ffted = ffted.permute(0, 1, 4, 2, 3).contiguous() # (batch, c, 2, h, w/2+1)
2023-02-26 02:19:48 +01:00
ffted = ffted.view(
(
batch,
-1,
)
+ ffted.size()[3:]
)
2022-09-02 04:37:30 +02:00
if self.spectral_pos_encoding:
height, width = ffted.shape[-2:]
2023-02-26 02:19:48 +01:00
coords_vert = (
torch.linspace(0, 1, height)[None, None, :, None]
.expand(batch, 1, height, width)
.to(ffted)
)
coords_hor = (
torch.linspace(0, 1, width)[None, None, None, :]
.expand(batch, 1, height, width)
.to(ffted)
)
2022-09-02 04:37:30 +02:00
ffted = torch.cat((coords_vert, coords_hor, ffted), dim=1)
if self.use_se:
ffted = self.se(ffted)
ffted = self.conv_layer(ffted) # (batch, c*2, h, w/2+1)
ffted = self.relu(ffted)
2023-02-26 02:19:48 +01:00
ffted = (
ffted.view(
(
batch,
-1,
2,
)
+ ffted.size()[2:]
)
.permute(0, 1, 3, 4, 2)
.contiguous()
) # (batch,c, t, h, w/2+1, 2)
2022-09-02 04:37:30 +02:00
ffted = torch.complex(ffted[..., 0], ffted[..., 1])
ifft_shape_slice = x.shape[-3:] if self.ffc3d else x.shape[-2:]
2023-02-26 02:19:48 +01:00
output = torch.fft.irfftn(
ffted, s=ifft_shape_slice, dim=fft_dim, norm=self.fft_norm
)
2022-09-02 04:37:30 +02:00
if self.spatial_scale_factor is not None:
2023-02-26 02:19:48 +01:00
output = F.interpolate(
output,
size=orig_size,
mode=self.spatial_scale_mode,
align_corners=False,
)
2022-09-02 04:37:30 +02:00
return output
class SpectralTransform(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
in_channels,
out_channels,
stride=1,
groups=1,
enable_lfu=True,
**fu_kwargs,
):
2022-09-02 04:37:30 +02:00
# bn_layer not used
super(SpectralTransform, self).__init__()
self.enable_lfu = enable_lfu
if stride == 2:
self.downsample = nn.AvgPool2d(kernel_size=(2, 2), stride=2)
else:
self.downsample = nn.Identity()
self.stride = stride
self.conv1 = nn.Sequential(
2023-02-26 02:19:48 +01:00
nn.Conv2d(
in_channels, out_channels // 2, kernel_size=1, groups=groups, bias=False
),
2022-09-02 04:37:30 +02:00
# nn.BatchNorm2d(out_channels // 2),
2023-02-26 02:19:48 +01:00
nn.ReLU(inplace=True),
2022-09-02 04:37:30 +02:00
)
2023-02-26 02:19:48 +01:00
self.fu = FourierUnit(out_channels // 2, out_channels // 2, groups, **fu_kwargs)
2022-09-02 04:37:30 +02:00
if self.enable_lfu:
2023-02-26 02:19:48 +01:00
self.lfu = FourierUnit(out_channels // 2, out_channels // 2, groups)
2022-09-02 04:37:30 +02:00
self.conv2 = torch.nn.Conv2d(
2023-02-26 02:19:48 +01:00
out_channels // 2, out_channels, kernel_size=1, groups=groups, bias=False
)
2022-09-02 04:37:30 +02:00
def forward(self, x):
x = self.downsample(x)
x = self.conv1(x)
output = self.fu(x)
if self.enable_lfu:
n, c, h, w = x.shape
split_no = 2
split_s = h // split_no
2023-02-26 02:19:48 +01:00
xs = torch.cat(
torch.split(x[:, : c // 4], split_s, dim=-2), dim=1
).contiguous()
xs = torch.cat(torch.split(xs, split_s, dim=-1), dim=1).contiguous()
2022-09-02 04:37:30 +02:00
xs = self.lfu(xs)
xs = xs.repeat(1, 1, split_no, split_no).contiguous()
else:
xs = 0
output = self.conv2(x + output + xs)
return output
class FFC(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
in_channels,
out_channels,
kernel_size,
ratio_gin,
ratio_gout,
stride=1,
padding=0,
dilation=1,
groups=1,
bias=False,
enable_lfu=True,
padding_type="reflect",
gated=False,
**spectral_kwargs,
):
2022-09-02 04:37:30 +02:00
super(FFC, self).__init__()
assert stride == 1 or stride == 2, "Stride should be 1 or 2."
self.stride = stride
in_cg = int(in_channels * ratio_gin)
in_cl = in_channels - in_cg
out_cg = int(out_channels * ratio_gout)
out_cl = out_channels - out_cg
# groups_g = 1 if groups == 1 else int(groups * ratio_gout)
# groups_l = 1 if groups == 1 else groups - groups_g
self.ratio_gin = ratio_gin
self.ratio_gout = ratio_gout
self.global_in_num = in_cg
module = nn.Identity if in_cl == 0 or out_cl == 0 else nn.Conv2d
2023-02-26 02:19:48 +01:00
self.convl2l = module(
in_cl,
out_cl,
kernel_size,
stride,
padding,
dilation,
groups,
bias,
padding_mode=padding_type,
)
2022-09-02 04:37:30 +02:00
module = nn.Identity if in_cl == 0 or out_cg == 0 else nn.Conv2d
2023-02-26 02:19:48 +01:00
self.convl2g = module(
in_cl,
out_cg,
kernel_size,
stride,
padding,
dilation,
groups,
bias,
padding_mode=padding_type,
)
2022-09-02 04:37:30 +02:00
module = nn.Identity if in_cg == 0 or out_cl == 0 else nn.Conv2d
2023-02-26 02:19:48 +01:00
self.convg2l = module(
in_cg,
out_cl,
kernel_size,
stride,
padding,
dilation,
groups,
bias,
padding_mode=padding_type,
)
2022-09-02 04:37:30 +02:00
module = nn.Identity if in_cg == 0 or out_cg == 0 else SpectralTransform
self.convg2g = module(
2023-02-26 02:19:48 +01:00
in_cg,
out_cg,
stride,
1 if groups == 1 else groups // 2,
enable_lfu,
**spectral_kwargs,
)
2022-09-02 04:37:30 +02:00
self.gated = gated
2023-02-26 02:19:48 +01:00
module = (
nn.Identity if in_cg == 0 or out_cl == 0 or not self.gated else nn.Conv2d
)
2022-09-02 04:37:30 +02:00
self.gate = module(in_channels, 2, 1)
def forward(self, x, fname=None):
x_l, x_g = x if type(x) is tuple else (x, 0)
out_xl, out_xg = 0, 0
if self.gated:
total_input_parts = [x_l]
if torch.is_tensor(x_g):
total_input_parts.append(x_g)
total_input = torch.cat(total_input_parts, dim=1)
gates = torch.sigmoid(self.gate(total_input))
g2l_gate, l2g_gate = gates.chunk(2, dim=1)
else:
g2l_gate, l2g_gate = 1, 1
spec_x = self.convg2g(x_g)
if self.ratio_gout != 1:
out_xl = self.convl2l(x_l) + self.convg2l(x_g) * g2l_gate
if self.ratio_gout != 0:
out_xg = self.convl2g(x_l) * l2g_gate + spec_x
return out_xl, out_xg
class FFC_BN_ACT(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
in_channels,
out_channels,
kernel_size,
ratio_gin,
ratio_gout,
stride=1,
padding=0,
dilation=1,
groups=1,
bias=False,
norm_layer=nn.SyncBatchNorm,
activation_layer=nn.Identity,
padding_type="reflect",
enable_lfu=True,
**kwargs,
):
2022-09-02 04:37:30 +02:00
super(FFC_BN_ACT, self).__init__()
2023-02-26 02:19:48 +01:00
self.ffc = FFC(
in_channels,
out_channels,
kernel_size,
ratio_gin,
ratio_gout,
stride,
padding,
dilation,
groups,
bias,
enable_lfu,
padding_type=padding_type,
**kwargs,
)
2022-09-02 04:37:30 +02:00
lnorm = nn.Identity if ratio_gout == 1 else norm_layer
gnorm = nn.Identity if ratio_gout == 0 else norm_layer
global_channels = int(out_channels * ratio_gout)
# self.bn_l = lnorm(out_channels - global_channels)
# self.bn_g = gnorm(global_channels)
lact = nn.Identity if ratio_gout == 1 else activation_layer
gact = nn.Identity if ratio_gout == 0 else activation_layer
self.act_l = lact(inplace=True)
self.act_g = gact(inplace=True)
def forward(self, x, fname=None):
2023-02-26 02:19:48 +01:00
x_l, x_g = self.ffc(
x,
fname=fname,
)
2022-09-02 04:37:30 +02:00
x_l = self.act_l(x_l)
x_g = self.act_g(x_g)
return x_l, x_g
class FFCResnetBlock(nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
dim,
padding_type,
norm_layer,
activation_layer=nn.ReLU,
dilation=1,
spatial_transform_kwargs=None,
inline=False,
ratio_gin=0.75,
ratio_gout=0.75,
):
2022-09-02 04:37:30 +02:00
super().__init__()
2023-02-26 02:19:48 +01:00
self.conv1 = FFC_BN_ACT(
dim,
dim,
kernel_size=3,
padding=dilation,
dilation=dilation,
norm_layer=norm_layer,
activation_layer=activation_layer,
padding_type=padding_type,
ratio_gin=ratio_gin,
ratio_gout=ratio_gout,
)
self.conv2 = FFC_BN_ACT(
dim,
dim,
kernel_size=3,
padding=dilation,
dilation=dilation,
norm_layer=norm_layer,
activation_layer=activation_layer,
padding_type=padding_type,
ratio_gin=ratio_gin,
ratio_gout=ratio_gout,
)
2022-09-02 04:37:30 +02:00
self.inline = inline
def forward(self, x, fname=None):
if self.inline:
2023-02-26 02:19:48 +01:00
x_l, x_g = (
x[:, : -self.conv1.ffc.global_in_num],
x[:, -self.conv1.ffc.global_in_num :],
)
2022-09-02 04:37:30 +02:00
else:
x_l, x_g = x if type(x) is tuple else (x, 0)
id_l, id_g = x_l, x_g
x_l, x_g = self.conv1((x_l, x_g), fname=fname)
x_l, x_g = self.conv2((x_l, x_g), fname=fname)
x_l, x_g = id_l + x_l, id_g + x_g
out = x_l, x_g
if self.inline:
out = torch.cat(out, dim=1)
return out
class ConcatTupleLayer(nn.Module):
def forward(self, x):
assert isinstance(x, tuple)
x_l, x_g = x
assert torch.is_tensor(x_l) or torch.is_tensor(x_g)
if not torch.is_tensor(x_g):
return x_l
return torch.cat(x, dim=1)
class FFCBlock(torch.nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
dim, # Number of output/input channels.
kernel_size, # Width and height of the convolution kernel.
padding,
ratio_gin=0.75,
ratio_gout=0.75,
activation="linear", # Activation function: 'relu', 'lrelu', etc.
):
2022-09-02 04:37:30 +02:00
super().__init__()
2023-02-26 02:19:48 +01:00
if activation == "linear":
2022-09-02 04:37:30 +02:00
self.activation = nn.Identity
else:
self.activation = nn.ReLU
self.padding = padding
self.kernel_size = kernel_size
2023-02-26 02:19:48 +01:00
self.ffc_block = FFCResnetBlock(
dim=dim,
padding_type="reflect",
norm_layer=nn.SyncBatchNorm,
activation_layer=self.activation,
dilation=1,
ratio_gin=ratio_gin,
ratio_gout=ratio_gout,
)
2022-09-02 04:37:30 +02:00
self.concat_layer = ConcatTupleLayer()
def forward(self, gen_ft, mask, fname=None):
x = gen_ft.float()
2023-02-26 02:19:48 +01:00
x_l, x_g = (
x[:, : -self.ffc_block.conv1.ffc.global_in_num],
x[:, -self.ffc_block.conv1.ffc.global_in_num :],
)
2022-09-02 04:37:30 +02:00
id_l, id_g = x_l, x_g
x_l, x_g = self.ffc_block((x_l, x_g), fname=fname)
x_l, x_g = id_l + x_l, id_g + x_g
x = self.concat_layer((x_l, x_g))
return x + gen_ft.float()
class FFCSkipLayer(torch.nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
dim, # Number of input/output channels.
kernel_size=3, # Convolution kernel size.
ratio_gin=0.75,
ratio_gout=0.75,
):
2022-09-02 04:37:30 +02:00
super().__init__()
self.padding = kernel_size // 2
2023-02-26 02:19:48 +01:00
self.ffc_act = FFCBlock(
dim=dim,
kernel_size=kernel_size,
activation=nn.ReLU,
padding=self.padding,
ratio_gin=ratio_gin,
ratio_gout=ratio_gout,
)
2022-09-02 04:37:30 +02:00
def forward(self, gen_ft, mask, fname=None):
x = self.ffc_act(gen_ft, mask, fname=fname)
return x
class SynthesisBlock(torch.nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
in_channels, # Number of input channels, 0 = first block.
out_channels, # Number of output channels.
w_dim, # Intermediate latent (W) dimensionality.
resolution, # Resolution of this block.
img_channels, # Number of output color channels.
is_last, # Is this the last block?
architecture="skip", # Architecture: 'orig', 'skip', 'resnet'.
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.
use_fp16=False, # Use FP16 for this block?
fp16_channels_last=False, # Use channels-last memory format with FP16?
**layer_kwargs, # Arguments for SynthesisLayer.
):
assert architecture in ["orig", "skip", "resnet"]
2022-09-02 04:37:30 +02:00
super().__init__()
self.in_channels = in_channels
self.w_dim = w_dim
self.resolution = resolution
self.img_channels = img_channels
self.is_last = is_last
self.architecture = architecture
self.use_fp16 = use_fp16
2023-02-26 02:19:48 +01:00
self.channels_last = use_fp16 and fp16_channels_last
self.register_buffer("resample_filter", setup_filter(resample_filter))
2022-09-02 04:37:30 +02:00
self.num_conv = 0
self.num_torgb = 0
self.res_ffc = {4: 0, 8: 0, 16: 0, 32: 1, 64: 1, 128: 1, 256: 1, 512: 1}
if in_channels != 0 and resolution >= 8:
self.ffc_skip = nn.ModuleList()
for _ in range(self.res_ffc[resolution]):
self.ffc_skip.append(FFCSkipLayer(dim=out_channels))
if in_channels == 0:
2023-02-26 02:19:48 +01:00
self.const = torch.nn.Parameter(
torch.randn([out_channels, resolution, resolution])
)
2022-09-02 04:37:30 +02:00
if in_channels != 0:
2023-02-26 02:19:48 +01:00
self.conv0 = SynthesisLayer(
in_channels,
out_channels,
w_dim=w_dim * 3,
resolution=resolution,
up=2,
resample_filter=resample_filter,
conv_clamp=conv_clamp,
channels_last=self.channels_last,
**layer_kwargs,
)
2022-09-02 04:37:30 +02:00
self.num_conv += 1
2023-02-26 02:19:48 +01:00
self.conv1 = SynthesisLayer(
out_channels,
out_channels,
w_dim=w_dim * 3,
resolution=resolution,
conv_clamp=conv_clamp,
channels_last=self.channels_last,
**layer_kwargs,
)
2022-09-02 04:37:30 +02:00
self.num_conv += 1
2023-02-26 02:19:48 +01:00
if is_last or architecture == "skip":
self.torgb = ToRGBLayer(
out_channels,
img_channels,
w_dim=w_dim * 3,
conv_clamp=conv_clamp,
channels_last=self.channels_last,
)
2022-09-02 04:37:30 +02:00
self.num_torgb += 1
2023-02-26 02:19:48 +01:00
if in_channels != 0 and architecture == "resnet":
self.skip = Conv2dLayer(
in_channels,
out_channels,
kernel_size=1,
bias=False,
up=2,
resample_filter=resample_filter,
channels_last=self.channels_last,
)
def forward(
self,
x,
mask,
feats,
img,
ws,
fname=None,
force_fp32=False,
fused_modconv=None,
**layer_kwargs,
):
2022-09-02 04:37:30 +02:00
dtype = torch.float16 if self.use_fp16 and not force_fp32 else torch.float32
dtype = torch.float32
2023-02-26 02:19:48 +01:00
memory_format = (
torch.channels_last
if self.channels_last and not force_fp32
else torch.contiguous_format
)
2022-09-02 04:37:30 +02:00
if fused_modconv is None:
2023-02-26 02:19:48 +01:00
fused_modconv = (not self.training) and (
dtype == torch.float32 or int(x.shape[0]) == 1
)
2022-09-02 04:37:30 +02:00
x = x.to(dtype=dtype, memory_format=memory_format)
2023-02-26 02:19:48 +01:00
x_skip = (
feats[self.resolution].clone().to(dtype=dtype, memory_format=memory_format)
)
2022-09-02 04:37:30 +02:00
# Main layers.
if self.in_channels == 0:
x = self.conv1(x, ws[1], fused_modconv=fused_modconv, **layer_kwargs)
2023-02-26 02:19:48 +01:00
elif self.architecture == "resnet":
2022-09-02 04:37:30 +02:00
y = self.skip(x, gain=np.sqrt(0.5))
2023-02-26 02:19:48 +01:00
x = self.conv0(
x, ws[0].clone(), fused_modconv=fused_modconv, **layer_kwargs
)
2022-09-02 04:37:30 +02:00
if len(self.ffc_skip) > 0:
2023-02-26 02:19:48 +01:00
mask = F.interpolate(
mask,
size=x_skip.shape[2:],
)
2022-09-02 04:37:30 +02:00
z = x + x_skip
for fres in self.ffc_skip:
z = fres(z, mask)
x = x + z
else:
x = x + x_skip
2023-02-26 02:19:48 +01:00
x = self.conv1(
x,
ws[1].clone(),
fused_modconv=fused_modconv,
gain=np.sqrt(0.5),
**layer_kwargs,
)
2022-09-02 04:37:30 +02:00
x = y.add_(x)
else:
2023-02-26 02:19:48 +01:00
x = self.conv0(
x, ws[0].clone(), fused_modconv=fused_modconv, **layer_kwargs
)
2022-09-02 04:37:30 +02:00
if len(self.ffc_skip) > 0:
2023-02-26 02:19:48 +01:00
mask = F.interpolate(
mask,
size=x_skip.shape[2:],
)
2022-09-02 04:37:30 +02:00
z = x + x_skip
for fres in self.ffc_skip:
z = fres(z, mask)
x = x + z
else:
x = x + x_skip
2023-02-26 02:19:48 +01:00
x = self.conv1(
x, ws[1].clone(), fused_modconv=fused_modconv, **layer_kwargs
)
2022-09-02 04:37:30 +02:00
# ToRGB.
if img is not None:
img = upsample2d(img, self.resample_filter)
2023-02-26 02:19:48 +01:00
if self.is_last or self.architecture == "skip":
2022-09-02 04:37:30 +02:00
y = self.torgb(x, ws[2].clone(), fused_modconv=fused_modconv)
y = y.to(dtype=torch.float32, memory_format=torch.contiguous_format)
img = img.add_(y) if img is not None else y
x = x.to(dtype=dtype)
assert x.dtype == dtype
assert img is None or img.dtype == torch.float32
return x, img
class SynthesisNetwork(torch.nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
w_dim, # Intermediate latent (W) dimensionality.
z_dim, # Output Latent (Z) dimensionality.
img_resolution, # Output image resolution.
img_channels, # Number of color channels.
channel_base=16384, # Overall multiplier for the number of channels.
channel_max=512, # Maximum number of channels in any layer.
num_fp16_res=0, # Use FP16 for the N highest resolutions.
**block_kwargs, # Arguments for SynthesisBlock.
):
2022-09-02 04:37:30 +02:00
assert img_resolution >= 4 and img_resolution & (img_resolution - 1) == 0
super().__init__()
self.w_dim = w_dim
self.img_resolution = img_resolution
self.img_resolution_log2 = int(np.log2(img_resolution))
self.img_channels = img_channels
2023-02-26 02:19:48 +01:00
self.block_resolutions = [
2**i for i in range(3, self.img_resolution_log2 + 1)
]
channels_dict = {
res: min(channel_base // res, channel_max) for res in self.block_resolutions
}
2022-09-02 04:37:30 +02:00
fp16_resolution = max(2 ** (self.img_resolution_log2 + 1 - num_fp16_res), 8)
2023-02-26 02:19:48 +01:00
self.foreword = SynthesisForeword(
img_channels=img_channels,
in_channels=min(channel_base // 4, channel_max),
z_dim=z_dim * 2,
resolution=4,
)
2022-09-02 04:37:30 +02:00
self.num_ws = self.img_resolution_log2 * 2 - 2
for res in self.block_resolutions:
if res // 2 in channels_dict.keys():
in_channels = channels_dict[res // 2] if res > 4 else 0
else:
in_channels = min(channel_base // (res // 2), channel_max)
out_channels = channels_dict[res]
2023-02-26 02:19:48 +01:00
use_fp16 = res >= fp16_resolution
2022-09-02 04:37:30 +02:00
use_fp16 = False
2023-02-26 02:19:48 +01:00
is_last = res == self.img_resolution
block = SynthesisBlock(
in_channels,
out_channels,
w_dim=w_dim,
resolution=res,
img_channels=img_channels,
is_last=is_last,
use_fp16=use_fp16,
**block_kwargs,
)
setattr(self, f"b{res}", block)
2022-09-02 04:37:30 +02:00
def forward(self, x_global, mask, feats, ws, fname=None, **block_kwargs):
img = None
x, img = self.foreword(x_global, ws, feats, img)
for res in self.block_resolutions:
2023-02-26 02:19:48 +01:00
block = getattr(self, f"b{res}")
2022-09-02 04:37:30 +02:00
mod_vector0 = []
mod_vector0.append(ws[:, int(np.log2(res)) * 2 - 5])
mod_vector0.append(x_global.clone())
mod_vector0 = torch.cat(mod_vector0, dim=1)
mod_vector1 = []
mod_vector1.append(ws[:, int(np.log2(res)) * 2 - 4])
mod_vector1.append(x_global.clone())
mod_vector1 = torch.cat(mod_vector1, dim=1)
mod_vector_rgb = []
mod_vector_rgb.append(ws[:, int(np.log2(res)) * 2 - 3])
mod_vector_rgb.append(x_global.clone())
mod_vector_rgb = torch.cat(mod_vector_rgb, dim=1)
2023-02-26 02:19:48 +01:00
x, img = block(
x,
mask,
feats,
img,
(mod_vector0, mod_vector1, mod_vector_rgb),
fname=fname,
**block_kwargs,
)
2022-09-02 04:37:30 +02:00
return img
class MappingNetwork(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.
):
2022-09-02 04:37:30 +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
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]
)
2022-09-02 04:37:30 +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)
2022-09-02 04:37:30 +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]))
2022-09-02 04:37:30 +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
):
2022-09-02 04:37:30 +02:00
# Embed, normalize, and concat inputs.
x = None
2023-02-26 02:19:48 +01:00
with torch.autograd.profiler.record_function("input"):
2022-09-02 04:37:30 +02:00
if self.z_dim > 0:
x = normalize_2nd_moment(z.to(torch.float32))
if self.c_dim > 0:
y = normalize_2nd_moment(self.embed(c.to(torch.float32)))
x = torch.cat([x, y], dim=1) if x is not None else y
# Main layers.
for idx in range(self.num_layers):
2023-02-26 02:19:48 +01:00
layer = getattr(self, f"fc{idx}")
2022-09-02 04:37:30 +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:
2023-02-26 02:19:48 +01:00
with torch.autograd.profiler.record_function("update_w_avg"):
self.w_avg.copy_(
x.detach().mean(dim=0).lerp(self.w_avg, self.w_avg_beta)
)
2022-09-02 04:37:30 +02:00
# Broadcast.
if self.num_ws is not None:
2023-02-26 02:19:48 +01:00
with torch.autograd.profiler.record_function("broadcast"):
2022-09-02 04:37:30 +02:00
x = x.unsqueeze(1).repeat([1, self.num_ws, 1])
# Apply truncation.
if truncation_psi != 1:
2023-02-26 02:19:48 +01:00
with torch.autograd.profiler.record_function("truncate"):
2022-09-02 04:37:30 +02:00
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:
2023-02-26 02:19:48 +01:00
x[:, :truncation_cutoff] = self.w_avg.lerp(
x[:, :truncation_cutoff], truncation_psi
)
2022-09-02 04:37:30 +02:00
return x
class Generator(torch.nn.Module):
2023-02-26 02:19:48 +01:00
def __init__(
self,
z_dim, # Input latent (Z) dimensionality.
c_dim, # Conditioning label (C) dimensionality.
w_dim, # Intermediate latent (W) dimensionality.
img_resolution, # Output resolution.
img_channels, # Number of output color channels.
encoder_kwargs={}, # Arguments for EncoderNetwork.
mapping_kwargs={}, # Arguments for MappingNetwork.
synthesis_kwargs={}, # Arguments for SynthesisNetwork.
):
2022-09-02 04:37:30 +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.encoder = EncoderNetwork(
c_dim=c_dim,
z_dim=z_dim,
img_resolution=img_resolution,
img_channels=img_channels,
**encoder_kwargs,
)
self.synthesis = SynthesisNetwork(
z_dim=z_dim,
w_dim=w_dim,
img_resolution=img_resolution,
img_channels=img_channels,
**synthesis_kwargs,
)
2022-09-02 04:37:30 +02:00
self.num_ws = self.synthesis.num_ws
2023-02-26 02:19:48 +01:00
self.mapping = MappingNetwork(
z_dim=z_dim, c_dim=c_dim, w_dim=w_dim, num_ws=self.num_ws, **mapping_kwargs
)
2022-09-02 04:37:30 +02:00
2023-02-26 02:19:48 +01:00
def forward(
self,
img,
c,
fname=None,
truncation_psi=1,
truncation_cutoff=None,
**synthesis_kwargs,
):
2022-09-02 04:37:30 +02:00
mask = img[:, -1].unsqueeze(1)
x_global, z, feats = self.encoder(img, c)
2023-02-26 02:19:48 +01:00
ws = self.mapping(
z, c, truncation_psi=truncation_psi, truncation_cutoff=truncation_cutoff
)
2022-09-02 04:37:30 +02:00
img = self.synthesis(x_global, mask, feats, ws, fname=fname, **synthesis_kwargs)
return img
FCF_MODEL_URL = os.environ.get(
"FCF_MODEL_URL",
2022-09-02 05:08:32 +02:00
"https://github.com/Sanster/models/releases/download/add_fcf/places_512_G.pth",
2022-09-02 04:37:30 +02:00
)
2023-02-26 02:19:48 +01:00
FCF_MODEL_MD5 = os.environ.get("FCF_MODEL_MD5", "3323152bc01bf1c56fd8aba74435a211")
2022-09-02 04:37:30 +02:00
class FcF(InpaintModel):
2023-02-11 06:30:09 +01:00
name = "fcf"
2022-09-02 04:37:30 +02:00
min_size = 512
pad_mod = 512
pad_to_square = True
2023-12-01 03:15:35 +01:00
is_erase_model = True
2022-09-02 04:37:30 +02:00
2022-09-15 16:21:27 +02:00
def init_model(self, device, **kwargs):
2022-09-02 04:37:30 +02:00
seed = 0
random.seed(seed)
np.random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
2023-02-26 02:19:48 +01:00
kwargs = {
"channel_base": 1 * 32768,
"channel_max": 512,
"num_fp16_res": 4,
"conv_clamp": 256,
}
G = Generator(
z_dim=512,
c_dim=0,
w_dim=512,
img_resolution=512,
img_channels=3,
synthesis_kwargs=kwargs,
encoder_kwargs=kwargs,
mapping_kwargs={"num_layers": 2},
)
self.model = load_model(G, FCF_MODEL_URL, device, FCF_MODEL_MD5)
2022-09-02 04:37:30 +02:00
self.label = torch.zeros([1, self.model.c_dim], device=device)
2023-11-16 14:12:06 +01:00
@staticmethod
def download():
download_model(FCF_MODEL_URL, FCF_MODEL_MD5)
2022-09-02 04:37:30 +02:00
@staticmethod
def is_downloaded() -> bool:
return os.path.exists(get_cache_path_by_url(FCF_MODEL_URL))
2022-09-04 10:00:42 +02:00
@torch.no_grad()
2023-12-30 16:36:44 +01:00
def __call__(self, image, mask, config: InpaintRequest):
2022-09-04 10:00:42 +02:00
"""
images: [H, W, C] RGB, not normalized
masks: [H, W]
return: BGR IMAGE
"""
2022-09-04 15:23:58 +02:00
if image.shape[0] == 512 and image.shape[1] == 512:
return self._pad_forward(image, mask, config)
2022-09-04 10:00:42 +02:00
boxes = boxes_from_mask(mask)
crop_result = []
config.hd_strategy_crop_margin = 128
for box in boxes:
crop_image, crop_mask, crop_box = self._crop_box(image, mask, box, config)
origin_size = crop_image.shape[:2]
resize_image = resize_max_size(crop_image, size_limit=512)
resize_mask = resize_max_size(crop_mask, size_limit=512)
inpaint_result = self._pad_forward(resize_image, resize_mask, config)
# only paste masked area result
2023-02-26 02:19:48 +01:00
inpaint_result = cv2.resize(
inpaint_result,
(origin_size[1], origin_size[0]),
interpolation=cv2.INTER_CUBIC,
)
2022-09-04 10:00:42 +02:00
original_pixel_indices = crop_mask < 127
2023-02-26 02:19:48 +01:00
inpaint_result[original_pixel_indices] = crop_image[:, :, ::-1][
original_pixel_indices
]
2022-09-04 10:00:42 +02:00
crop_result.append((inpaint_result, crop_box))
inpaint_result = image[:, :, ::-1]
for crop_image, crop_box in crop_result:
x1, y1, x2, y2 = crop_box
inpaint_result[y1:y2, x1:x2, :] = crop_image
return inpaint_result
2023-12-30 16:36:44 +01:00
def forward(self, image, mask, config: InpaintRequest):
2022-09-02 04:37:30 +02:00
"""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 > 120) * 255
mask = norm_img(mask)
image = torch.from_numpy(image).unsqueeze(0).to(self.device)
mask = torch.from_numpy(mask).unsqueeze(0).to(self.device)
erased_img = image * (1 - mask)
input_image = torch.cat([0.5 - mask, erased_img], dim=1)
2023-02-26 02:19:48 +01:00
output = self.model(
input_image, self.label, truncation_psi=0.1, noise_mode="none"
)
output = (
(output.permute(0, 2, 3, 1) * 127.5 + 127.5)
.round()
.clamp(0, 255)
.to(torch.uint8)
)
2022-09-02 04:37:30 +02:00
output = output[0].cpu().numpy()
cur_res = cv2.cvtColor(output, cv2.COLOR_RGB2BGR)
return cur_res