906 lines
30 KiB
Python
906 lines
30 KiB
Python
import functools
|
|
import math
|
|
import multiprocessing
|
|
from contextlib import contextmanager, ExitStack
|
|
from functools import partial
|
|
from math import log2
|
|
from random import random
|
|
|
|
import torch
|
|
import torch.nn.functional as F
|
|
import trainer.losses as L
|
|
import numpy as np
|
|
|
|
from kornia.filters import filter2D
|
|
from linear_attention_transformer import ImageLinearAttention
|
|
from torch import nn
|
|
from torch.autograd import grad as torch_grad
|
|
from vector_quantize_pytorch import VectorQuantize
|
|
|
|
from trainer.networks import register_model
|
|
from utils.util import checkpoint
|
|
|
|
try:
|
|
from apex import amp
|
|
|
|
APEX_AVAILABLE = True
|
|
except:
|
|
APEX_AVAILABLE = False
|
|
|
|
assert torch.cuda.is_available(), 'You need to have an Nvidia GPU with CUDA installed.'
|
|
|
|
num_cores = multiprocessing.cpu_count()
|
|
|
|
# constants
|
|
|
|
EPS = 1e-8
|
|
CALC_FID_NUM_IMAGES = 12800
|
|
|
|
|
|
# helper classes
|
|
|
|
def DiffAugment(x, types=[]):
|
|
for p in types:
|
|
for f in AUGMENT_FNS[p]:
|
|
x = f(x)
|
|
return x.contiguous()
|
|
|
|
def rand_brightness(x):
|
|
x = x + (torch.rand(x.size(0), 1, 1, 1, dtype=x.dtype, device=x.device) - 0.5)
|
|
return x
|
|
|
|
def rand_saturation(x):
|
|
x_mean = x.mean(dim=1, keepdim=True)
|
|
x = (x - x_mean) * (torch.rand(x.size(0), 1, 1, 1, dtype=x.dtype, device=x.device) * 2) + x_mean
|
|
return x
|
|
|
|
def rand_contrast(x):
|
|
x_mean = x.mean(dim=[1, 2, 3], keepdim=True)
|
|
x = (x - x_mean) * (torch.rand(x.size(0), 1, 1, 1, dtype=x.dtype, device=x.device) + 0.5) + x_mean
|
|
return x
|
|
|
|
def rand_translation(x, ratio=0.125):
|
|
shift_x, shift_y = int(x.size(2) * ratio + 0.5), int(x.size(3) * ratio + 0.5)
|
|
translation_x = torch.randint(-shift_x, shift_x + 1, size=[x.size(0), 1, 1], device=x.device)
|
|
translation_y = torch.randint(-shift_y, shift_y + 1, size=[x.size(0), 1, 1], device=x.device)
|
|
grid_batch, grid_x, grid_y = torch.meshgrid(
|
|
torch.arange(x.size(0), dtype=torch.long, device=x.device),
|
|
torch.arange(x.size(2), dtype=torch.long, device=x.device),
|
|
torch.arange(x.size(3), dtype=torch.long, device=x.device),
|
|
)
|
|
grid_x = torch.clamp(grid_x + translation_x + 1, 0, x.size(2) + 1)
|
|
grid_y = torch.clamp(grid_y + translation_y + 1, 0, x.size(3) + 1)
|
|
x_pad = F.pad(x, [1, 1, 1, 1, 0, 0, 0, 0])
|
|
x = x_pad.permute(0, 2, 3, 1).contiguous()[grid_batch, grid_x, grid_y].permute(0, 3, 1, 2)
|
|
return x
|
|
|
|
def rand_cutout(x, ratio=0.5):
|
|
cutout_size = int(x.size(2) * ratio + 0.5), int(x.size(3) * ratio + 0.5)
|
|
offset_x = torch.randint(0, x.size(2) + (1 - cutout_size[0] % 2), size=[x.size(0), 1, 1], device=x.device)
|
|
offset_y = torch.randint(0, x.size(3) + (1 - cutout_size[1] % 2), size=[x.size(0), 1, 1], device=x.device)
|
|
grid_batch, grid_x, grid_y = torch.meshgrid(
|
|
torch.arange(x.size(0), dtype=torch.long, device=x.device),
|
|
torch.arange(cutout_size[0], dtype=torch.long, device=x.device),
|
|
torch.arange(cutout_size[1], dtype=torch.long, device=x.device),
|
|
)
|
|
grid_x = torch.clamp(grid_x + offset_x - cutout_size[0] // 2, min=0, max=x.size(2) - 1)
|
|
grid_y = torch.clamp(grid_y + offset_y - cutout_size[1] // 2, min=0, max=x.size(3) - 1)
|
|
mask = torch.ones(x.size(0), x.size(2), x.size(3), dtype=x.dtype, device=x.device)
|
|
mask[grid_batch, grid_x, grid_y] = 0
|
|
x = x * mask.unsqueeze(1)
|
|
return x
|
|
|
|
AUGMENT_FNS = {
|
|
'color': [rand_brightness, rand_saturation, rand_contrast],
|
|
'translation': [rand_translation],
|
|
'cutout': [rand_cutout],
|
|
}
|
|
|
|
class NanException(Exception):
|
|
pass
|
|
|
|
|
|
class EMA():
|
|
def __init__(self, beta):
|
|
super().__init__()
|
|
self.beta = beta
|
|
|
|
def update_average(self, old, new):
|
|
if not exists(old):
|
|
return new
|
|
return old * self.beta + (1 - self.beta) * new
|
|
|
|
|
|
class Flatten(nn.Module):
|
|
def forward(self, x):
|
|
return x.reshape(x.shape[0], -1)
|
|
|
|
|
|
class Residual(nn.Module):
|
|
def __init__(self, fn):
|
|
super().__init__()
|
|
self.fn = fn
|
|
|
|
def forward(self, x):
|
|
return self.fn(x) + x
|
|
|
|
|
|
class Rezero(nn.Module):
|
|
def __init__(self, fn):
|
|
super().__init__()
|
|
self.fn = fn
|
|
self.g = nn.Parameter(torch.zeros(1))
|
|
|
|
def forward(self, x):
|
|
return self.fn(x) * self.g
|
|
|
|
|
|
class PermuteToFrom(nn.Module):
|
|
def __init__(self, fn):
|
|
super().__init__()
|
|
self.fn = fn
|
|
|
|
def forward(self, x):
|
|
x = x.permute(0, 2, 3, 1)
|
|
out, loss = self.fn(x)
|
|
out = out.permute(0, 3, 1, 2)
|
|
return out, loss
|
|
|
|
|
|
class Blur(nn.Module):
|
|
def __init__(self):
|
|
super().__init__()
|
|
f = torch.Tensor([1, 2, 1])
|
|
self.register_buffer('f', f)
|
|
|
|
def forward(self, x):
|
|
f = self.f
|
|
f = f[None, None, :] * f[None, :, None]
|
|
return filter2D(x, f, normalized=True)
|
|
|
|
|
|
# one layer of self-attention and feedforward, for images
|
|
|
|
attn_and_ff = lambda chan: nn.Sequential(*[
|
|
Residual(Rezero(ImageLinearAttention(chan, norm_queries=True))),
|
|
Residual(Rezero(nn.Sequential(nn.Conv2d(chan, chan * 2, 1), leaky_relu(), nn.Conv2d(chan * 2, chan, 1))))
|
|
])
|
|
|
|
|
|
# helpers
|
|
|
|
def exists(val):
|
|
return val is not None
|
|
|
|
|
|
@contextmanager
|
|
def null_context():
|
|
yield
|
|
|
|
|
|
def combine_contexts(contexts):
|
|
@contextmanager
|
|
def multi_contexts():
|
|
with ExitStack() as stack:
|
|
yield [stack.enter_context(ctx()) for ctx in contexts]
|
|
|
|
return multi_contexts
|
|
|
|
|
|
def default(value, d):
|
|
return value if exists(value) else d
|
|
|
|
|
|
def cycle(iterable):
|
|
while True:
|
|
for i in iterable:
|
|
yield i
|
|
|
|
|
|
def cast_list(el):
|
|
return el if isinstance(el, list) else [el]
|
|
|
|
|
|
def is_empty(t):
|
|
if isinstance(t, torch.Tensor):
|
|
return t.nelement() == 0
|
|
return not exists(t)
|
|
|
|
|
|
def raise_if_nan(t):
|
|
if torch.isnan(t):
|
|
raise NanException
|
|
|
|
|
|
def gradient_accumulate_contexts(gradient_accumulate_every, is_ddp, ddps):
|
|
if is_ddp:
|
|
num_no_syncs = gradient_accumulate_every - 1
|
|
head = [combine_contexts(map(lambda ddp: ddp.no_sync, ddps))] * num_no_syncs
|
|
tail = [null_context]
|
|
contexts = head + tail
|
|
else:
|
|
contexts = [null_context] * gradient_accumulate_every
|
|
|
|
for context in contexts:
|
|
with context():
|
|
yield
|
|
|
|
|
|
def loss_backwards(fp16, loss, optimizer, loss_id, **kwargs):
|
|
if fp16:
|
|
with amp.scale_loss(loss, optimizer, loss_id) as scaled_loss:
|
|
scaled_loss.backward(**kwargs)
|
|
else:
|
|
loss.backward(**kwargs)
|
|
|
|
|
|
def gradient_penalty(images, output, weight=10):
|
|
batch_size = images.shape[0]
|
|
gradients = torch_grad(outputs=output, inputs=images,
|
|
grad_outputs=torch.ones(output.size(), device=images.device),
|
|
create_graph=True, retain_graph=True, only_inputs=True)[0]
|
|
|
|
gradients = gradients.reshape(batch_size, -1)
|
|
return weight * ((gradients.norm(2, dim=1) - 1) ** 2).mean()
|
|
|
|
|
|
def calc_pl_lengths(styles, images):
|
|
device = images.device
|
|
num_pixels = images.shape[2] * images.shape[3]
|
|
pl_noise = torch.randn(images.shape, device=device) / math.sqrt(num_pixels)
|
|
outputs = (images * pl_noise).sum()
|
|
|
|
pl_grads = torch_grad(outputs=outputs, inputs=styles,
|
|
grad_outputs=torch.ones(outputs.shape, device=device),
|
|
create_graph=True, retain_graph=True, only_inputs=True)[0]
|
|
|
|
return (pl_grads ** 2).sum(dim=2).mean(dim=1).sqrt()
|
|
|
|
|
|
def image_noise(n, im_size, device):
|
|
return torch.FloatTensor(n, im_size, im_size, 1).uniform_(0., 1.).cuda(device)
|
|
|
|
|
|
def leaky_relu(p=0.2):
|
|
return nn.LeakyReLU(p, inplace=True)
|
|
|
|
|
|
def evaluate_in_chunks(max_batch_size, model, *args):
|
|
split_args = list(zip(*list(map(lambda x: x.split(max_batch_size, dim=0), args))))
|
|
chunked_outputs = [model(*i) for i in split_args]
|
|
if len(chunked_outputs) == 1:
|
|
return chunked_outputs[0]
|
|
return torch.cat(chunked_outputs, dim=0)
|
|
|
|
|
|
def set_requires_grad(model, bool):
|
|
for p in model.parameters():
|
|
p.requires_grad = bool
|
|
|
|
|
|
def slerp(val, low, high):
|
|
low_norm = low / torch.norm(low, dim=1, keepdim=True)
|
|
high_norm = high / torch.norm(high, dim=1, keepdim=True)
|
|
omega = torch.acos((low_norm * high_norm).sum(1))
|
|
so = torch.sin(omega)
|
|
res = (torch.sin((1.0 - val) * omega) / so).unsqueeze(1) * low + (torch.sin(val * omega) / so).unsqueeze(1) * high
|
|
return res
|
|
|
|
# augmentations
|
|
|
|
def random_hflip(tensor, prob):
|
|
if prob > random():
|
|
return tensor
|
|
return torch.flip(tensor, dims=(3,))
|
|
|
|
|
|
class StyleGan2Augmentor(nn.Module):
|
|
def __init__(self, D, image_size, types, prob):
|
|
super().__init__()
|
|
self.D = D
|
|
self.prob = prob
|
|
self.types = types
|
|
|
|
def forward(self, images, detach=False):
|
|
if random() < self.prob:
|
|
images = random_hflip(images, prob=0.5)
|
|
images = DiffAugment(images, types=self.types)
|
|
|
|
if detach:
|
|
images = images.detach()
|
|
|
|
# Save away for use elsewhere (e.g. unet loss)
|
|
self.aug_images = images
|
|
|
|
return self.D(images)
|
|
|
|
|
|
# stylegan2 classes
|
|
|
|
class EqualLinear(nn.Module):
|
|
def __init__(self, in_dim, out_dim, lr_mul=1, bias=True):
|
|
super().__init__()
|
|
self.weight = nn.Parameter(torch.randn(out_dim, in_dim))
|
|
if bias:
|
|
self.bias = nn.Parameter(torch.zeros(out_dim))
|
|
|
|
self.lr_mul = lr_mul
|
|
|
|
def forward(self, input):
|
|
return F.linear(input, self.weight * self.lr_mul, bias=self.bias * self.lr_mul)
|
|
|
|
|
|
class StyleVectorizer(nn.Module):
|
|
def __init__(self, emb, depth, lr_mul=0.1):
|
|
super().__init__()
|
|
|
|
layers = []
|
|
for i in range(depth):
|
|
layers.extend([EqualLinear(emb, emb, lr_mul), leaky_relu()])
|
|
|
|
self.net = nn.Sequential(*layers)
|
|
|
|
def forward(self, x):
|
|
x = F.normalize(x, dim=1)
|
|
return self.net(x)
|
|
|
|
|
|
class RGBBlock(nn.Module):
|
|
def __init__(self, latent_dim, input_channel, upsample, rgba=False):
|
|
super().__init__()
|
|
self.input_channel = input_channel
|
|
self.to_style = nn.Linear(latent_dim, input_channel)
|
|
|
|
out_filters = 3 if not rgba else 4
|
|
self.conv = Conv2DMod(input_channel, out_filters, 1, demod=False)
|
|
|
|
self.upsample = nn.Sequential(
|
|
nn.Upsample(scale_factor=2, mode='bilinear', align_corners=False),
|
|
Blur()
|
|
) if upsample else None
|
|
|
|
def forward(self, x, prev_rgb, istyle):
|
|
b, c, h, w = x.shape
|
|
style = self.to_style(istyle)
|
|
x = self.conv(x, style)
|
|
|
|
if exists(prev_rgb):
|
|
x = x + prev_rgb
|
|
|
|
if exists(self.upsample):
|
|
x = self.upsample(x)
|
|
|
|
return x
|
|
|
|
|
|
class AdaptiveInstanceNorm(nn.Module):
|
|
def __init__(self, in_channel, style_dim):
|
|
super().__init__()
|
|
from models.archs.arch_util import ConvGnLelu
|
|
self.style2scale = ConvGnLelu(style_dim, in_channel, kernel_size=1, norm=False, activation=False, bias=True)
|
|
self.style2bias = ConvGnLelu(style_dim, in_channel, kernel_size=1, norm=False, activation=False, bias=True, weight_init_factor=0)
|
|
self.norm = nn.InstanceNorm2d(in_channel)
|
|
|
|
def forward(self, input, style):
|
|
gamma = self.style2scale(style)
|
|
beta = self.style2bias(style)
|
|
out = self.norm(input)
|
|
out = gamma * out + beta
|
|
return out
|
|
|
|
|
|
class NoiseInjection(nn.Module):
|
|
def __init__(self, channel):
|
|
super().__init__()
|
|
self.weight = nn.Parameter(torch.zeros(1, channel, 1, 1))
|
|
|
|
def forward(self, image, noise):
|
|
return image + self.weight * noise
|
|
|
|
|
|
class EqualLR:
|
|
def __init__(self, name):
|
|
self.name = name
|
|
|
|
def compute_weight(self, module):
|
|
weight = getattr(module, self.name + '_orig')
|
|
fan_in = weight.data.size(1) * weight.data[0][0].numel()
|
|
|
|
return weight * math.sqrt(2 / fan_in)
|
|
|
|
@staticmethod
|
|
def apply(module, name):
|
|
fn = EqualLR(name)
|
|
|
|
weight = getattr(module, name)
|
|
del module._parameters[name]
|
|
module.register_parameter(name + '_orig', nn.Parameter(weight.data))
|
|
module.register_forward_pre_hook(fn)
|
|
|
|
return fn
|
|
|
|
def __call__(self, module, input):
|
|
weight = self.compute_weight(module)
|
|
setattr(module, self.name, weight)
|
|
|
|
|
|
def equal_lr(module, name='weight'):
|
|
EqualLR.apply(module, name)
|
|
|
|
return module
|
|
|
|
|
|
class EqualConv2d(nn.Module):
|
|
def __init__(self, *args, **kwargs):
|
|
super().__init__()
|
|
|
|
conv = nn.Conv2d(*args, **kwargs)
|
|
conv.weight.data.normal_()
|
|
conv.bias.data.zero_()
|
|
self.conv = equal_lr(conv)
|
|
|
|
def forward(self, input):
|
|
return self.conv(input)
|
|
|
|
|
|
class Conv2DMod(nn.Module):
|
|
def __init__(self, in_chan, out_chan, kernel, demod=True, stride=1, dilation=1, **kwargs):
|
|
super().__init__()
|
|
self.filters = out_chan
|
|
self.demod = demod
|
|
self.kernel = kernel
|
|
self.stride = stride
|
|
self.dilation = dilation
|
|
self.weight = nn.Parameter(torch.randn((out_chan, in_chan, kernel, kernel)))
|
|
nn.init.kaiming_normal_(self.weight, a=0, mode='fan_in', nonlinearity='leaky_relu')
|
|
|
|
def _get_same_padding(self, size, kernel, dilation, stride):
|
|
return ((size - 1) * (stride - 1) + dilation * (kernel - 1)) // 2
|
|
|
|
def forward(self, x, y):
|
|
b, c, h, w = x.shape
|
|
|
|
w1 = y[:, None, :, None, None]
|
|
w2 = self.weight[None, :, :, :, :]
|
|
weights = w2 * (w1 + 1)
|
|
|
|
if self.demod:
|
|
d = torch.rsqrt((weights ** 2).sum(dim=(2, 3, 4), keepdim=True) + EPS)
|
|
weights = weights * d
|
|
|
|
x = x.reshape(1, -1, h, w)
|
|
|
|
_, _, *ws = weights.shape
|
|
weights = weights.reshape(b * self.filters, *ws)
|
|
|
|
padding = self._get_same_padding(h, self.kernel, self.dilation, self.stride)
|
|
x = F.conv2d(x, weights, padding=padding, groups=b)
|
|
|
|
x = x.reshape(-1, self.filters, h, w)
|
|
return x
|
|
|
|
|
|
class GeneratorBlockWithStructure(nn.Module):
|
|
def __init__(self, latent_dim, input_channels, filters, upsample=True, upsample_rgb=True, rgba=False):
|
|
super().__init__()
|
|
self.upsample = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=False) if upsample else None
|
|
|
|
# Uses stylegan1 style blocks for injecting structural latent.
|
|
self.conv0 = EqualConv2d(input_channels, filters, 3, padding=1)
|
|
self.to_noise0 = nn.Linear(1, filters)
|
|
self.noise0 = equal_lr(NoiseInjection(filters))
|
|
self.adain0 = AdaptiveInstanceNorm(filters, latent_dim)
|
|
|
|
self.to_style1 = nn.Linear(latent_dim, filters)
|
|
self.to_noise1 = nn.Linear(1, filters)
|
|
self.conv1 = Conv2DMod(filters, filters, 3)
|
|
|
|
self.to_style2 = nn.Linear(latent_dim, filters)
|
|
self.to_noise2 = nn.Linear(1, filters)
|
|
self.conv2 = Conv2DMod(filters, filters, 3)
|
|
|
|
self.activation = leaky_relu()
|
|
self.to_rgb = RGBBlock(latent_dim, filters, upsample_rgb, rgba)
|
|
|
|
def forward(self, x, prev_rgb, istyle, inoise, structure_input):
|
|
if exists(self.upsample):
|
|
x = self.upsample(x)
|
|
|
|
inoise = inoise[:, :x.shape[2], :x.shape[3], :]
|
|
noise0 = self.to_noise0(inoise).permute((0, 3, 1, 2))
|
|
noise1 = self.to_noise1(inoise).permute((0, 3, 1, 2))
|
|
noise2 = self.to_noise2(inoise).permute((0, 3, 1, 2))
|
|
|
|
structure = torch.nn.functional.interpolate(structure_input, size=x.shape[2:], mode="nearest")
|
|
x = self.conv0(x)
|
|
x = self.noise0(x, noise0)
|
|
x = self.adain0(x, structure)
|
|
|
|
style1 = self.to_style1(istyle)
|
|
x = self.conv1(x, style1)
|
|
x = self.activation(x + noise1)
|
|
|
|
style2 = self.to_style2(istyle)
|
|
x = self.conv2(x, style2)
|
|
x = self.activation(x + noise2)
|
|
|
|
rgb = self.to_rgb(x, prev_rgb, istyle)
|
|
return x, rgb
|
|
|
|
|
|
class GeneratorBlock(nn.Module):
|
|
def __init__(self, latent_dim, input_channels, filters, upsample=True, upsample_rgb=True, rgba=False, structure_input=False):
|
|
super().__init__()
|
|
self.upsample = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=False) if upsample else None
|
|
|
|
self.structure_input = structure_input
|
|
if self.structure_input:
|
|
self.structure_conv = nn.Conv2d(3, input_channels, 3, padding=1)
|
|
input_channels = input_channels * 2
|
|
|
|
self.to_style1 = nn.Linear(latent_dim, input_channels)
|
|
self.to_noise1 = nn.Linear(1, filters)
|
|
self.conv1 = Conv2DMod(input_channels, filters, 3)
|
|
|
|
self.to_style2 = nn.Linear(latent_dim, filters)
|
|
self.to_noise2 = nn.Linear(1, filters)
|
|
self.conv2 = Conv2DMod(filters, filters, 3)
|
|
|
|
self.activation = leaky_relu()
|
|
self.to_rgb = RGBBlock(latent_dim, filters, upsample_rgb, rgba)
|
|
|
|
def forward(self, x, prev_rgb, istyle, inoise, structure_input=None):
|
|
if exists(self.upsample):
|
|
x = self.upsample(x)
|
|
|
|
if self.structure_input:
|
|
s = self.structure_conv(structure_input)
|
|
x = torch.cat([x, s], dim=1)
|
|
|
|
inoise = inoise[:, :x.shape[2], :x.shape[3], :]
|
|
noise1 = self.to_noise1(inoise).permute((0, 3, 2, 1))
|
|
noise2 = self.to_noise2(inoise).permute((0, 3, 2, 1))
|
|
|
|
style1 = self.to_style1(istyle)
|
|
x = self.conv1(x, style1)
|
|
x = self.activation(x + noise1)
|
|
|
|
style2 = self.to_style2(istyle)
|
|
x = self.conv2(x, style2)
|
|
x = self.activation(x + noise2)
|
|
|
|
rgb = self.to_rgb(x, prev_rgb, istyle)
|
|
return x, rgb
|
|
|
|
|
|
class Generator(nn.Module):
|
|
def __init__(self, image_size, latent_dim, network_capacity=16, transparent=False, attn_layers=[], no_const=False,
|
|
fmap_max=512, structure_input=False):
|
|
super().__init__()
|
|
self.image_size = image_size
|
|
self.latent_dim = latent_dim
|
|
self.num_layers = int(log2(image_size) - 1)
|
|
|
|
filters = [network_capacity * (2 ** (i + 1)) for i in range(self.num_layers)][::-1]
|
|
|
|
set_fmap_max = partial(min, fmap_max)
|
|
filters = list(map(set_fmap_max, filters))
|
|
init_channels = filters[0]
|
|
filters = [init_channels, *filters]
|
|
|
|
in_out_pairs = zip(filters[:-1], filters[1:])
|
|
self.no_const = no_const
|
|
|
|
if no_const:
|
|
self.to_initial_block = nn.ConvTranspose2d(latent_dim, init_channels, 4, 1, 0, bias=False)
|
|
else:
|
|
self.initial_block = nn.Parameter(torch.randn((1, init_channels, 4, 4)))
|
|
|
|
self.initial_conv = nn.Conv2d(filters[0], filters[0], 3, padding=1)
|
|
self.blocks = nn.ModuleList([])
|
|
self.attns = nn.ModuleList([])
|
|
|
|
for ind, (in_chan, out_chan) in enumerate(in_out_pairs):
|
|
not_first = ind != 0
|
|
not_last = ind != (self.num_layers - 1)
|
|
num_layer = self.num_layers - ind
|
|
|
|
attn_fn = attn_and_ff(in_chan) if num_layer in attn_layers else None
|
|
|
|
self.attns.append(attn_fn)
|
|
|
|
if structure_input:
|
|
block_fn = GeneratorBlockWithStructure
|
|
else:
|
|
block_fn = GeneratorBlock
|
|
|
|
block = block_fn(
|
|
latent_dim,
|
|
in_chan,
|
|
out_chan,
|
|
upsample=not_first,
|
|
upsample_rgb=not_last,
|
|
rgba=transparent
|
|
)
|
|
self.blocks.append(block)
|
|
|
|
def forward(self, styles, input_noise, structure_input=None, starting_shape=None):
|
|
batch_size = styles.shape[0]
|
|
image_size = self.image_size
|
|
|
|
if self.no_const:
|
|
avg_style = styles.mean(dim=1)[:, :, None, None]
|
|
x = self.to_initial_block(avg_style)
|
|
else:
|
|
x = self.initial_block.expand(batch_size, -1, -1, -1)
|
|
if starting_shape is not None:
|
|
x = F.interpolate(x, size=starting_shape, mode="bilinear")
|
|
|
|
rgb = None
|
|
styles = styles.transpose(0, 1)
|
|
x = self.initial_conv(x)
|
|
|
|
if structure_input is not None:
|
|
s = torch.nn.functional.interpolate(structure_input, size=x.shape[2:], mode="nearest")
|
|
for style, block, attn in zip(styles, self.blocks, self.attns):
|
|
if exists(attn):
|
|
x = checkpoint(attn, x)
|
|
if structure_input is not None:
|
|
if exists(block.upsample):
|
|
# In this case, the structural guidance is given by the extra information over the previous layer.
|
|
twoX = (x.shape[2]*2, x.shape[3]*2)
|
|
sn = torch.nn.functional.interpolate(structure_input, size=twoX, mode="nearest")
|
|
s_int = torch.nn.functional.interpolate(s, size=twoX, mode="bilinear")
|
|
s_diff = sn - s_int
|
|
else:
|
|
# This is the initial case - just feed in the base structure.
|
|
s_diff = s
|
|
else:
|
|
s_diff = None
|
|
x, rgb = checkpoint(block, x, rgb, style, input_noise, s_diff)
|
|
|
|
return rgb
|
|
|
|
|
|
# Wrapper that combines style vectorizer with the actual generator.
|
|
class StyleGan2GeneratorWithLatent(nn.Module):
|
|
def __init__(self, image_size, latent_dim=512, style_depth=8, lr_mlp=.1, network_capacity=16, transparent=False,
|
|
attn_layers=[], no_const=False, fmap_max=512, structure_input=False):
|
|
super().__init__()
|
|
self.vectorizer = StyleVectorizer(latent_dim, style_depth, lr_mul=lr_mlp)
|
|
self.gen = Generator(image_size, latent_dim, network_capacity, transparent, attn_layers, no_const, fmap_max,
|
|
structure_input=structure_input)
|
|
self.mixed_prob = .9
|
|
self._init_weights()
|
|
|
|
def noise(self, n, latent_dim, device):
|
|
return torch.randn(n, latent_dim).cuda(device)
|
|
|
|
def noise_list(self, n, layers, latent_dim, device):
|
|
return [(self.noise(n, latent_dim, device), layers)]
|
|
|
|
def mixed_list(self, n, layers, latent_dim, device):
|
|
tt = int(torch.rand(()).numpy() * layers)
|
|
return self.noise_list(n, tt, latent_dim, device) + self.noise_list(n, layers - tt, latent_dim, device)
|
|
|
|
def latent_to_w(self, style_vectorizer, latent_descr):
|
|
return [(style_vectorizer(z), num_layers) for z, num_layers in latent_descr]
|
|
|
|
def styles_def_to_tensor(self, styles_def):
|
|
return torch.cat([t[:, None, :].expand(-1, n, -1) for t, n in styles_def], dim=1)
|
|
|
|
# To use per the stylegan paper, input should be uniform noise. This gen takes it in as a normal "image" format:
|
|
# b,f,h,w.
|
|
def forward(self, x, structure_input=None, fit_starting_shape_to_structure=False):
|
|
b, f, h, w = x.shape
|
|
|
|
full_random_latents = True
|
|
if full_random_latents:
|
|
style = self.noise(b*2, self.gen.latent_dim, x.device)
|
|
w = self.vectorizer(style)
|
|
# Randomly distribute styles across layers
|
|
w_styles = w[:,None,:].expand(-1, self.gen.num_layers, -1).clone()
|
|
for j in range(b):
|
|
cutoff = int(torch.rand(()).numpy() * self.gen.num_layers)
|
|
if cutoff == self.gen.num_layers or random() > self.mixed_prob:
|
|
w_styles[j] = w_styles[j*2]
|
|
else:
|
|
w_styles[j, :cutoff] = w_styles[j*2, :cutoff]
|
|
w_styles[j, cutoff:] = w_styles[j*2+1, cutoff:]
|
|
w_styles = w_styles[:b]
|
|
else:
|
|
get_latents_fn = self.mixed_list if random() < self.mixed_prob else self.noise_list
|
|
style = get_latents_fn(b, self.gen.num_layers, self.gen.latent_dim, device=x.device)
|
|
w_space = self.latent_to_w(self.vectorizer, style)
|
|
w_styles = self.styles_def_to_tensor(w_space)
|
|
|
|
starting_shape = None
|
|
if fit_starting_shape_to_structure:
|
|
starting_shape = (x.shape[2] // 32, x.shape[3] // 32)
|
|
# The underlying model expects the noise as b,h,w,1. Make it so.
|
|
return self.gen(w_styles, x[:,0,:,:].unsqueeze(dim=3), structure_input, starting_shape), w_styles
|
|
|
|
def _init_weights(self):
|
|
for m in self.modules():
|
|
if type(m) in {nn.Conv2d, nn.Linear} and hasattr(m, 'weight'):
|
|
nn.init.kaiming_normal_(m.weight, a=0, mode='fan_in', nonlinearity='leaky_relu')
|
|
|
|
for block in self.gen.blocks:
|
|
nn.init.zeros_(block.to_noise1.weight)
|
|
nn.init.zeros_(block.to_noise2.weight)
|
|
nn.init.zeros_(block.to_noise1.bias)
|
|
nn.init.zeros_(block.to_noise2.bias)
|
|
|
|
|
|
class DiscriminatorBlock(nn.Module):
|
|
def __init__(self, input_channels, filters, downsample=True):
|
|
super().__init__()
|
|
self.conv_res = nn.Conv2d(input_channels, filters, 1, stride=(2 if downsample else 1))
|
|
|
|
self.net = nn.Sequential(
|
|
nn.Conv2d(input_channels, filters, 3, padding=1),
|
|
leaky_relu(),
|
|
nn.Conv2d(filters, filters, 3, padding=1),
|
|
leaky_relu()
|
|
)
|
|
|
|
self.downsample = nn.Sequential(
|
|
Blur(),
|
|
nn.Conv2d(filters, filters, 3, padding=1, stride=2)
|
|
) if downsample else None
|
|
|
|
def forward(self, x):
|
|
res = self.conv_res(x)
|
|
x = self.net(x)
|
|
if exists(self.downsample):
|
|
x = self.downsample(x)
|
|
x = (x + res) * (1 / math.sqrt(2))
|
|
return x
|
|
|
|
|
|
class StyleGan2Discriminator(nn.Module):
|
|
def __init__(self, image_size, network_capacity=16, fq_layers=[], fq_dict_size=256, attn_layers=[],
|
|
transparent=False, fmap_max=512, input_filters=3):
|
|
super().__init__()
|
|
num_layers = int(log2(image_size) - 1)
|
|
|
|
blocks = []
|
|
filters = [input_filters] + [(64) * (2 ** i) for i in range(num_layers + 1)]
|
|
|
|
set_fmap_max = partial(min, fmap_max)
|
|
filters = list(map(set_fmap_max, filters))
|
|
chan_in_out = list(zip(filters[:-1], filters[1:]))
|
|
|
|
blocks = []
|
|
attn_blocks = []
|
|
quantize_blocks = []
|
|
|
|
for ind, (in_chan, out_chan) in enumerate(chan_in_out):
|
|
num_layer = ind + 1
|
|
is_not_last = ind != (len(chan_in_out) - 1)
|
|
|
|
block = DiscriminatorBlock(in_chan, out_chan, downsample=is_not_last)
|
|
blocks.append(block)
|
|
|
|
attn_fn = attn_and_ff(out_chan) if num_layer in attn_layers else None
|
|
|
|
attn_blocks.append(attn_fn)
|
|
|
|
quantize_fn = PermuteToFrom(VectorQuantize(out_chan, fq_dict_size)) if num_layer in fq_layers else None
|
|
quantize_blocks.append(quantize_fn)
|
|
|
|
self.blocks = nn.ModuleList(blocks)
|
|
self.attn_blocks = nn.ModuleList(attn_blocks)
|
|
self.quantize_blocks = nn.ModuleList(quantize_blocks)
|
|
|
|
chan_last = filters[-1]
|
|
latent_dim = 2 * 2 * chan_last
|
|
|
|
self.final_conv = nn.Conv2d(chan_last, chan_last, 3, padding=1)
|
|
self.flatten = Flatten()
|
|
self.to_logit = nn.Linear(latent_dim, 1)
|
|
|
|
self._init_weights()
|
|
|
|
def forward(self, x):
|
|
b, *_ = x.shape
|
|
|
|
quantize_loss = torch.zeros(1).to(x)
|
|
|
|
for (block, attn_block, q_block) in zip(self.blocks, self.attn_blocks, self.quantize_blocks):
|
|
x = block(x)
|
|
|
|
if exists(attn_block):
|
|
x = attn_block(x)
|
|
|
|
if exists(q_block):
|
|
x, _, loss = q_block(x)
|
|
quantize_loss += loss
|
|
|
|
x = self.final_conv(x)
|
|
x = self.flatten(x)
|
|
x = self.to_logit(x)
|
|
if exists(q_block):
|
|
return x.squeeze(), quantize_loss
|
|
else:
|
|
return x.squeeze()
|
|
|
|
def _init_weights(self):
|
|
for m in self.modules():
|
|
if type(m) in {nn.Conv2d, nn.Linear}:
|
|
nn.init.kaiming_normal_(m.weight, a=0, mode='fan_in', nonlinearity='leaky_relu')
|
|
|
|
|
|
class StyleGan2DivergenceLoss(L.ConfigurableLoss):
|
|
def __init__(self, opt, env):
|
|
super().__init__(opt, env)
|
|
self.real = opt['real']
|
|
self.fake = opt['fake']
|
|
self.discriminator = opt['discriminator']
|
|
self.for_gen = opt['gen_loss']
|
|
self.gp_frequency = opt['gradient_penalty_frequency']
|
|
self.noise = opt['noise'] if 'noise' in opt.keys() else 0
|
|
|
|
def forward(self, net, state):
|
|
real_input = state[self.real]
|
|
fake_input = state[self.fake]
|
|
if self.noise != 0:
|
|
fake_input = fake_input + torch.rand_like(fake_input) * self.noise
|
|
real_input = real_input + torch.rand_like(real_input) * self.noise
|
|
|
|
D = self.env['discriminators'][self.discriminator]
|
|
fake = D(fake_input)
|
|
if self.for_gen:
|
|
return fake.mean()
|
|
else:
|
|
real_input.requires_grad_() # <-- Needed to compute gradients on the input.
|
|
real = D(real_input)
|
|
divergence_loss = (F.relu(1 + real) + F.relu(1 - fake)).mean()
|
|
|
|
# Apply gradient penalty. TODO: migrate this elsewhere.
|
|
if self.env['step'] % self.gp_frequency == 0:
|
|
from models.archs.stylegan.stylegan2_lucidrains import gradient_penalty
|
|
gp = gradient_penalty(real_input, real)
|
|
self.metrics.append(("gradient_penalty", gp.clone().detach()))
|
|
divergence_loss = divergence_loss + gp
|
|
|
|
real_input.requires_grad_(requires_grad=False)
|
|
return divergence_loss
|
|
|
|
|
|
class StyleGan2PathLengthLoss(L.ConfigurableLoss):
|
|
def __init__(self, opt, env):
|
|
super().__init__(opt, env)
|
|
self.w_styles = opt['w_styles']
|
|
self.gen = opt['gen']
|
|
self.pl_mean = None
|
|
from models.archs.stylegan.stylegan2_lucidrains import EMA
|
|
self.pl_length_ma = EMA(.99)
|
|
|
|
def forward(self, net, state):
|
|
w_styles = state[self.w_styles]
|
|
gen = state[self.gen]
|
|
from models.archs.stylegan.stylegan2_lucidrains import calc_pl_lengths
|
|
pl_lengths = calc_pl_lengths(w_styles, gen)
|
|
avg_pl_length = np.mean(pl_lengths.detach().cpu().numpy())
|
|
|
|
from models.archs.stylegan.stylegan2_lucidrains import is_empty
|
|
if not is_empty(self.pl_mean):
|
|
pl_loss = ((pl_lengths - self.pl_mean) ** 2).mean()
|
|
if not torch.isnan(pl_loss):
|
|
return pl_loss
|
|
else:
|
|
print("Path length loss returned NaN!")
|
|
|
|
self.pl_mean = self.pl_length_ma.update_average(self.pl_mean, avg_pl_length)
|
|
return 0
|
|
|
|
|
|
@register_model
|
|
def register_stylegan2_lucidrains(opt_net, opt):
|
|
is_structured = opt_net['structured'] if 'structured' in opt_net.keys() else False
|
|
attn = opt_net['attn_layers'] if 'attn_layers' in opt_net.keys() else []
|
|
return StyleGan2GeneratorWithLatent(image_size=opt_net['image_size'], latent_dim=opt_net['latent_dim'],
|
|
style_depth=opt_net['style_depth'], structure_input=is_structured,
|
|
attn_layers=attn)
|