forked from mrq/DL-Art-School
226 lines
12 KiB
Python
226 lines
12 KiB
Python
import math
|
|
import functools
|
|
from models.archs.arch_util import MultiConvBlock, ConvGnLelu, ConvGnSilu, ReferenceJoinBlock
|
|
from models.archs.SwitchedResidualGenerator_arch import ConfigurableSwitchComputer
|
|
from models.archs.SPSR_arch import ImageGradientNoPadding
|
|
from torch import nn
|
|
import torch
|
|
import torch.nn.functional as F
|
|
from switched_conv_util import save_attention_to_image_rgb
|
|
from switched_conv import compute_attention_specificity
|
|
import os
|
|
import torchvision
|
|
|
|
|
|
# VGG-style layer with Conv(stride2)->BN->Activation->Conv->BN->Activation
|
|
# Doubles the input filter count.
|
|
class HalvingProcessingBlock(nn.Module):
|
|
def __init__(self, filters):
|
|
super(HalvingProcessingBlock, self).__init__()
|
|
self.bnconv1 = ConvGnSilu(filters, filters * 2, kernel_size=1, stride=2, norm=False, bias=False)
|
|
self.bnconv2 = ConvGnSilu(filters * 2, filters * 2, norm=True, bias=False)
|
|
|
|
def forward(self, x):
|
|
x = self.bnconv1(x)
|
|
return self.bnconv2(x)
|
|
|
|
|
|
class ExpansionBlock2(nn.Module):
|
|
def __init__(self, filters_in, filters_out=None, block=ConvGnSilu):
|
|
super(ExpansionBlock2, self).__init__()
|
|
if filters_out is None:
|
|
filters_out = filters_in // 2
|
|
self.decimate = block(filters_in, filters_out, kernel_size=1, bias=False, activation=True, norm=False)
|
|
self.process_passthrough = block(filters_out, filters_out, kernel_size=3, bias=True, activation=True, norm=False)
|
|
self.conjoin = block(filters_out*2, filters_out*2, kernel_size=1, bias=False, activation=True, norm=False)
|
|
self.reduce = block(filters_out*2, filters_out, kernel_size=1, bias=False, activation=False, norm=True)
|
|
|
|
# input is the feature signal with shape (b, f, w, h)
|
|
# passthrough is the structure signal with shape (b, f/2, w*2, h*2)
|
|
# output is conjoined upsample with shape (b, f/2, w*2, h*2)
|
|
def forward(self, input, passthrough):
|
|
x = F.interpolate(input, scale_factor=2, mode="nearest")
|
|
x = self.decimate(x)
|
|
p = self.process_passthrough(passthrough)
|
|
x = self.conjoin(torch.cat([x, p], dim=1))
|
|
return self.reduce(x)
|
|
|
|
|
|
# Basic convolutional upsampling block that uses interpolate.
|
|
class UpconvBlock(nn.Module):
|
|
def __init__(self, filters_in, filters_out=None, block=ConvGnSilu, norm=True, activation=True, bias=False):
|
|
super(UpconvBlock, self).__init__()
|
|
self.reduce = block(filters_in, filters_out, kernel_size=1, bias=False, activation=False, norm=False)
|
|
self.process = block(filters_out, filters_out, kernel_size=3, bias=bias, activation=activation, norm=norm)
|
|
|
|
def forward(self, x):
|
|
x = self.reduce(x)
|
|
x = F.interpolate(x, scale_factor=2, mode="nearest")
|
|
return self.process(x)
|
|
|
|
|
|
class SSGMultiplexer(nn.Module):
|
|
def __init__(self, nf, multiplexer_channels, reductions=2):
|
|
super(SSGMultiplexer, self).__init__()
|
|
|
|
# Blocks used to create the query
|
|
self.input_process = ConvGnSilu(nf, nf, activation=True, norm=False, bias=True)
|
|
self.embedding_process = ConvGnSilu(256, 256, activation=True, norm=False, bias=True)
|
|
self.reduction_blocks = nn.ModuleList([HalvingProcessingBlock(nf * 2 ** i) for i in range(reductions)])
|
|
reduction_filters = nf * 2 ** reductions
|
|
self.processing_blocks = nn.Sequential(
|
|
ConvGnSilu(reduction_filters + 256, reduction_filters + 128, kernel_size=1, activation=True, norm=False, bias=True),
|
|
ConvGnSilu(reduction_filters + 128, reduction_filters, kernel_size=3, activation=True, norm=True, bias=False))
|
|
self.expansion_blocks = nn.ModuleList([ExpansionBlock2(reduction_filters // (2 ** i)) for i in range(reductions)])
|
|
|
|
# Blocks used to create the key
|
|
self.key_process = ConvGnSilu(nf, nf, kernel_size=1, activation=True, norm=False, bias=True)
|
|
|
|
# Postprocessing blocks.
|
|
self.query_key_combine = ConvGnSilu(nf*2, nf, kernel_size=1, activation=True, norm=False, bias=False)
|
|
self.cbl1 = ConvGnSilu(nf, nf // 4, kernel_size=1, activation=True, norm=True, bias=False, num_groups=4)
|
|
self.cbl2 = ConvGnSilu(nf // 4, 1, kernel_size=1, activation=False, norm=False, bias=False)
|
|
|
|
def forward(self, x, embedding, transformations):
|
|
q = self.input_process(x)
|
|
embedding = self.embedding_process(embedding)
|
|
reduction_identities = []
|
|
for b in self.reduction_blocks:
|
|
reduction_identities.append(q)
|
|
q = b(q)
|
|
q = self.processing_blocks(torch.cat([q, embedding], dim=1))
|
|
for i, b in enumerate(self.expansion_blocks):
|
|
q = b(q, reduction_identities[-i - 1])
|
|
|
|
b, t, f, h, w = transformations.shape
|
|
k = transformations.view(b * t, f, h, w)
|
|
k = self.key_process(k)
|
|
|
|
q = q.view(b, 1, f, h, w).repeat(1, t, 1, 1, 1).view(b * t, f, h, w)
|
|
v = self.query_key_combine(torch.cat([q, k], dim=1))
|
|
|
|
v = self.cbl1(v)
|
|
v = self.cbl2(v)
|
|
|
|
return v.view(b, t, h, w)
|
|
|
|
class SSGr1(nn.Module):
|
|
def __init__(self, in_nc, out_nc, nf, xforms=8, upscale=4, init_temperature=10):
|
|
super(SSGr1, self).__init__()
|
|
n_upscale = int(math.log(upscale, 2))
|
|
|
|
# switch options
|
|
transformation_filters = nf
|
|
self.transformation_counts = xforms
|
|
multiplx_fn = functools.partial(SSGMultiplexer, transformation_filters)
|
|
transform_fn = functools.partial(MultiConvBlock, transformation_filters, int(transformation_filters * 1.25),
|
|
transformation_filters, kernel_size=3, depth=3,
|
|
weight_init_factor=.1)
|
|
|
|
# Feature branch
|
|
self.model_fea_conv = ConvGnLelu(in_nc, nf, kernel_size=3, norm=False, activation=False)
|
|
self.noise_ref_join = ReferenceJoinBlock(nf, residual_weight_init_factor=.1, kernel_size=1, depth=2)
|
|
self.sw1 = ConfigurableSwitchComputer(transformation_filters, multiplx_fn,
|
|
pre_transform_block=None, transform_block=transform_fn,
|
|
attention_norm=True,
|
|
transform_count=self.transformation_counts, init_temp=init_temperature,
|
|
add_scalable_noise_to_transforms=False, feed_transforms_into_multiplexer=True)
|
|
self.feature_lr_conv = ConvGnLelu(nf, nf, kernel_size=3, norm=True, activation=False)
|
|
self.feature_lr_conv2 = ConvGnLelu(nf, nf, kernel_size=3, norm=False, activation=False, bias=False)
|
|
|
|
# Grad branch. Note - groupnorm on this branch is REALLY bad. Avoid it like the plague.
|
|
self.get_g_nopadding = ImageGradientNoPadding()
|
|
self.grad_conv = ConvGnLelu(in_nc, nf, kernel_size=3, norm=False, activation=False, bias=False)
|
|
self.noise_ref_join_grad = ReferenceJoinBlock(nf, residual_weight_init_factor=.1, kernel_size=1, depth=2)
|
|
self.grad_ref_join = ReferenceJoinBlock(nf, residual_weight_init_factor=.3, final_norm=False, kernel_size=1, depth=2)
|
|
self.sw_grad = ConfigurableSwitchComputer(transformation_filters, multiplx_fn,
|
|
pre_transform_block=None, transform_block=transform_fn,
|
|
attention_norm=True,
|
|
transform_count=self.transformation_counts // 2, init_temp=init_temperature,
|
|
add_scalable_noise_to_transforms=False, feed_transforms_into_multiplexer=True)
|
|
self.grad_lr_conv = ConvGnLelu(nf, nf, kernel_size=3, norm=False, activation=True, bias=True)
|
|
self.upsample_grad = nn.Sequential(*[UpconvBlock(nf, nf // 2, block=ConvGnLelu, norm=False, activation=True, bias=False) for _ in range(n_upscale)])
|
|
self.grad_branch_output_conv = ConvGnLelu(nf // 2, out_nc, kernel_size=1, norm=False, activation=False, bias=True)
|
|
|
|
# Join branch (grad+fea)
|
|
self.noise_ref_join_conjoin = ReferenceJoinBlock(nf, residual_weight_init_factor=.1, kernel_size=1, depth=2)
|
|
self.conjoin_ref_join = ReferenceJoinBlock(nf, residual_weight_init_factor=.3, kernel_size=1, depth=2)
|
|
self.conjoin_sw = ConfigurableSwitchComputer(transformation_filters, multiplx_fn,
|
|
pre_transform_block=None, transform_block=transform_fn,
|
|
attention_norm=True,
|
|
transform_count=self.transformation_counts, init_temp=init_temperature,
|
|
add_scalable_noise_to_transforms=False, feed_transforms_into_multiplexer=True)
|
|
self.final_lr_conv = ConvGnLelu(nf, nf, kernel_size=3, norm=False, activation=True, bias=True)
|
|
self.upsample = nn.Sequential(*[UpconvBlock(nf, 64, block=ConvGnLelu, norm=False, activation=True, bias=True) for _ in range(n_upscale)])
|
|
self.final_hr_conv2 = ConvGnLelu(64, out_nc, kernel_size=3, norm=False, activation=False, bias=False)
|
|
self.switches = [self.sw1, self.sw_grad, self.conjoin_sw]
|
|
self.attentions = None
|
|
self.lr = None
|
|
self.init_temperature = init_temperature
|
|
self.final_temperature_step = 10000
|
|
|
|
def forward(self, x, embedding):
|
|
noise_stds = []
|
|
# The attention_maps debugger outputs <x>. Save that here.
|
|
self.lr = x.detach().cpu()
|
|
|
|
x_grad = self.get_g_nopadding(x)
|
|
|
|
x = self.model_fea_conv(x)
|
|
x1 = x
|
|
x1, a1 = self.sw1(x1, True, identity=x, att_in=(x1, embedding))
|
|
|
|
x_grad = self.grad_conv(x_grad)
|
|
x_grad_identity = x_grad
|
|
x_grad, nstd = self.noise_ref_join_grad(x_grad, torch.randn_like(x_grad))
|
|
x_grad, grad_fea_std = self.grad_ref_join(x_grad, x1)
|
|
x_grad, a3 = self.sw_grad(x_grad, True, identity=x_grad_identity, att_in=(x_grad, embedding))
|
|
x_grad = self.grad_lr_conv(x_grad)
|
|
x_grad_out = self.upsample_grad(x_grad)
|
|
x_grad_out = self.grad_branch_output_conv(x_grad_out)
|
|
noise_stds.append(nstd)
|
|
|
|
x_out = x1
|
|
x_out, nstd = self.noise_ref_join_conjoin(x_out, torch.randn_like(x_out))
|
|
x_out, fea_grad_std = self.conjoin_ref_join(x_out, x_grad)
|
|
x_out, a4 = self.conjoin_sw(x_out, True, identity=x1, att_in=(x_out, embedding))
|
|
x_out = self.final_lr_conv(x_out)
|
|
x_out = self.upsample(x_out)
|
|
x_out = self.final_hr_conv2(x_out)
|
|
noise_stds.append(nstd)
|
|
|
|
self.attentions = [a1, a3, a4]
|
|
self.noise_stds = torch.stack(noise_stds).mean().detach().cpu()
|
|
self.grad_fea_std = grad_fea_std.detach().cpu()
|
|
self.fea_grad_std = fea_grad_std.detach().cpu()
|
|
return x_grad_out, x_out, x_grad
|
|
|
|
def set_temperature(self, temp):
|
|
[sw.set_temperature(temp) for sw in self.switches]
|
|
|
|
def update_for_step(self, step, experiments_path='.'):
|
|
if self.attentions:
|
|
temp = max(1, 1 + self.init_temperature *
|
|
(self.final_temperature_step - step) / self.final_temperature_step)
|
|
self.set_temperature(temp)
|
|
if step % 200 == 0:
|
|
output_path = os.path.join(experiments_path, "attention_maps")
|
|
prefix = "amap_%i_a%i_%%i.png"
|
|
[save_attention_to_image_rgb(output_path, self.attentions[i], self.transformation_counts, prefix % (step, i), step, output_mag=False) for i in range(len(self.attentions))]
|
|
torchvision.utils.save_image(self.lr, os.path.join(experiments_path, "attention_maps", "amap_%i_base_image.png" % (step,)))
|
|
|
|
|
|
def get_debug_values(self, step):
|
|
temp = self.switches[0].switch.temperature
|
|
mean_hists = [compute_attention_specificity(att, 2) for att in self.attentions]
|
|
means = [i[0] for i in mean_hists]
|
|
hists = [i[1].clone().detach().cpu().flatten() for i in mean_hists]
|
|
val = {"switch_temperature": temp,
|
|
"noise_branch_std_dev": self.noise_stds,
|
|
"grad_branch_feat_intg_std_dev": self.grad_fea_std,
|
|
"conjoin_branch_grad_intg_std_dev": self.fea_grad_std}
|
|
for i in range(len(means)):
|
|
val["switch_%i_specificity" % (i,)] = means[i]
|
|
val["switch_%i_histogram" % (i,)] = hists[i]
|
|
return val
|