import logging from collections import OrderedDict import torch import torch.nn as nn from torch.nn.parallel import DataParallel, DistributedDataParallel import models.networks as networks import models.lr_scheduler as lr_scheduler from models.base_model import BaseModel from models.loss import GANLoss, FDPLLoss from apex import amp from data.weight_scheduler import get_scheduler_for_opt from .archs.SPSR_arch import ImageGradient, ImageGradientNoPadding import torch.nn.functional as F import glob import random import torchvision.utils as utils import os logger = logging.getLogger('base') class GaussianBlur(nn.Module): def __init__(self): super(GaussianBlur, self).__init__() # Set these to whatever you want for your gaussian filter kernel_size = 3 sigma = 2 # Create a x, y coordinate grid of shape (kernel_size, kernel_size, 2) x_cord = torch.arange(kernel_size) x_grid = x_cord.repeat(kernel_size).view(kernel_size, kernel_size) y_grid = x_grid.t() xy_grid = torch.stack([x_grid, y_grid], dim=-1) mean = (kernel_size - 1) / 2. variance = sigma ** 2. # Calculate the 2-dimensional gaussian kernel which is # the product of two gaussian distributions for two different # variables (in this case called x and y) gaussian_kernel = (1. / (2. * 3.1415926 * variance)) * \ torch.exp( -torch.sum((xy_grid - mean) ** 2., dim=-1) / \ (2 * variance) ) # Make sure sum of values in gaussian kernel equals 1. gaussian_kernel = gaussian_kernel / torch.sum(gaussian_kernel) # Reshape to 2d depthwise convolutional weight gaussian_kernel = gaussian_kernel.view(1, 1, kernel_size, kernel_size) gaussian_kernel = gaussian_kernel.repeat(3, 1, 1, 1) self.gaussian_filter = nn.Conv2d(in_channels=3, out_channels=3, kernel_size=kernel_size, groups=3, bias=False) self.gaussian_filter.weight.data = gaussian_kernel self.gaussian_filter.weight.requires_grad = False def forward(self, x): return self.gaussian_filter(x) class SRGANModel(BaseModel): def __init__(self, opt): super(SRGANModel, self).__init__(opt) if opt['dist']: self.rank = torch.distributed.get_rank() else: self.rank = -1 # non dist training train_opt = opt['train'] self.spsr_enabled = 'spsr' in opt['model'] # define networks and load pretrained models self.netG = networks.define_G(opt).to(self.device) if self.is_train: self.netD = networks.define_D(opt).to(self.device) if self.spsr_enabled: logger.info("Defining grad net...") self.netD_grad = networks.define_D(opt, wrap=True).to(self.device) # D_grad if 'network_C' in opt.keys(): self.netC = networks.define_G(opt, net_key='network_C').to(self.device) # The corruptor net is fixed. Lock 'her down. self.netC.eval() for p in self.netC.parameters(): p.requires_grad = True else: self.netC = None self.mega_batch_factor = 1 self.disjoint_data = False # define losses, optimizer and scheduler if self.is_train: self.mega_batch_factor = train_opt['mega_batch_factor'] if self.mega_batch_factor is None: self.mega_batch_factor = 1 # G pixel loss if train_opt['pixel_weight'] > 0: l_pix_type = train_opt['pixel_criterion'] if l_pix_type == 'l1': self.cri_pix = nn.L1Loss().to(self.device) elif l_pix_type == 'l2': self.cri_pix = nn.MSELoss().to(self.device) else: raise NotImplementedError('Loss type [{:s}] not recognized.'.format(l_pix_type)) self.l_pix_w = train_opt['pixel_weight'] else: logger.info('Remove pixel loss.') self.cri_pix = None # FDPL loss. if 'fdpl_loss' in train_opt.keys(): fdpl_opt = train_opt['fdpl_loss'] self.fdpl_weight = fdpl_opt['weight'] self.fdpl_enabled = self.fdpl_weight > 0 if self.fdpl_enabled: self.cri_fdpl = FDPLLoss(fdpl_opt['data_mean'], self.device) else: self.fdpl_enabled = False if self.spsr_enabled: spsr_opt = train_opt['spsr'] self.branch_pretrain = spsr_opt['branch_pretrain'] if spsr_opt['branch_pretrain'] else 0 self.branch_init_iters = spsr_opt['branch_init_iters'] if spsr_opt['branch_init_iters'] else 1 if spsr_opt['gradient_pixel_weight'] > 0: self.cri_pix_grad = nn.MSELoss().to(self.device) self.l_pix_grad_w = spsr_opt['gradient_pixel_weight'] else: self.cri_pix_grad = None if spsr_opt['gradient_gan_weight'] > 0: self.cri_grad_gan = GANLoss(train_opt['gan_type'], 1.0, 0.0).to(self.device) self.l_gan_grad_w = spsr_opt['gradient_gan_weight'] else: self.cri_grad_gan = None if spsr_opt['pixel_branch_weight'] > 0: l_pix_type = spsr_opt['pixel_branch_criterion'] if l_pix_type == 'l1': self.cri_pix_branch = nn.L1Loss().to(self.device) elif l_pix_type == 'l2': self.cri_pix_branch = nn.MSELoss().to(self.device) else: raise NotImplementedError('Loss type [{:s}] not recognized.'.format(l_pix_type)) self.l_pix_branch_w = spsr_opt['pixel_branch_weight'] else: logger.info('Remove G_grad pixel loss.') self.cri_pix_branch = None # G feature loss if train_opt['feature_weight'] and train_opt['feature_weight'] > 0: # For backwards compatibility, use a scheduler definition instead. Remove this at some point. l_fea_type = train_opt['feature_criterion'] if l_fea_type == 'l1': self.cri_fea = nn.L1Loss().to(self.device) elif l_fea_type == 'l2': self.cri_fea = nn.MSELoss().to(self.device) else: raise NotImplementedError('Loss type [{:s}] not recognized.'.format(l_fea_type)) sched_params = { 'type': 'fixed', 'weight': train_opt['feature_weight'] } self.l_fea_sched = get_scheduler_for_opt(sched_params) elif train_opt['feature_scheduler']: self.l_fea_sched = get_scheduler_for_opt(train_opt['feature_scheduler']) l_fea_type = train_opt['feature_criterion'] if l_fea_type == 'l1': self.cri_fea = nn.L1Loss().to(self.device) elif l_fea_type == 'l2': self.cri_fea = nn.MSELoss().to(self.device) else: raise NotImplementedError('Loss type [{:s}] not recognized.'.format(l_fea_type)) else: logger.info('Remove feature loss.') self.cri_fea = None if self.cri_fea: # load VGG perceptual loss self.use_corrupted_feature_input = train_opt['corrupted_feature_input'] if 'corrupted_feature_input' in train_opt.keys() else False if self.use_corrupted_feature_input: logger.info("Corrupting inputs into the feature network..") self.feature_corruptor = GaussianBlur().to(self.device) self.netF = networks.define_F(use_bn=False).to(self.device) self.lr_netF = None if 'lr_fea_path' in train_opt.keys(): self.lr_netF = networks.define_F(use_bn=False, load_path=train_opt['lr_fea_path']).to(self.device) self.disjoint_data = True if opt['dist']: pass # do not need to use DistributedDataParallel for netF else: self.netF = DataParallel(self.netF) if self.lr_netF: self.lr_netF = DataParallel(self.lr_netF) # You can feed in a list of frozen pre-trained discriminators. These are treated the same as feature losses. self.fixed_disc_nets = [] if 'fixed_discriminators' in opt.keys(): for opt_fdisc in opt['fixed_discriminators'].keys(): netFD = networks.define_fixed_D(opt['fixed_discriminators'][opt_fdisc]).to(self.device) if opt['dist']: pass # do not need to use DistributedDataParallel for netF else: netFD = DataParallel(netFD) self.fixed_disc_nets.append(netFD) # GD gan loss self.cri_gan = GANLoss(train_opt['gan_type'], 1.0, 0.0).to(self.device) self.l_gan_w = train_opt['gan_weight'] # D_update_ratio and D_init_iters self.D_update_ratio = train_opt['D_update_ratio'] if train_opt['D_update_ratio'] else 1 self.D_init_iters = train_opt['D_init_iters'] if train_opt['D_init_iters'] else 0 self.G_warmup = train_opt['G_warmup'] if train_opt['G_warmup'] else -1 self.D_noise_theta = train_opt['D_noise_theta_init'] if train_opt['D_noise_theta_init'] else 0 self.D_noise_final = train_opt['D_noise_final_it'] if train_opt['D_noise_final_it'] else 0 self.D_noise_theta_floor = train_opt['D_noise_theta_floor'] if train_opt['D_noise_theta_floor'] else 0 self.corruptor_swapout_steps = train_opt['corruptor_swapout_steps'] if train_opt['corruptor_swapout_steps'] else 500 self.corruptor_usage_prob = train_opt['corruptor_usage_probability'] if train_opt['corruptor_usage_probability'] else .5 # optimizers # G optimizer wd_G = train_opt['weight_decay_G'] if train_opt['weight_decay_G'] else 0 optim_params = [] if train_opt['lr_scheme'] == 'ProgressiveMultiStepLR': optim_params = self.netG.get_param_groups() else: for k, v in self.netG.named_parameters(): # can optimize for a part of the model if v.requires_grad: optim_params.append(v) else: if self.rank <= 0: logger.warning('Params [{:s}] will not optimize.'.format(k)) self.optimizer_G = torch.optim.Adam(optim_params, lr=train_opt['lr_G'], weight_decay=wd_G, betas=(train_opt['beta1_G'], train_opt['beta2_G'])) self.optimizers.append(self.optimizer_G) # D optimizer optim_params = [] for k, v in self.netD.named_parameters(): # can optimize for a part of the model if v.requires_grad: optim_params.append(v) else: if self.rank <= 0: logger.warning('Params [{:s}] will not optimize.'.format(k)) wd_D = train_opt['weight_decay_D'] if train_opt['weight_decay_D'] else 0 self.optimizer_D = torch.optim.Adam(optim_params, lr=train_opt['lr_D'], weight_decay=wd_D, betas=(train_opt['beta1_D'], train_opt['beta2_D'])) self.optimizers.append(self.optimizer_D) self.disc_optimizers.append(self.optimizer_D) if self.spsr_enabled: # D_grad optimizer optim_params = [] for k, v in self.netD_grad.named_parameters(): # can optimize for a part of the model if v.requires_grad: optim_params.append(v) else: if self.rank <= 0: logger.warning('Params [{:s}] will not optimize.'.format(k)) # D wd_D = train_opt['weight_decay_D'] if train_opt['weight_decay_D'] else 0 self.optimizer_D_grad = torch.optim.Adam(optim_params, lr=train_opt['lr_D'], weight_decay=wd_D, betas=(train_opt['beta1_D'], train_opt['beta2_D'])) self.optimizers.append(self.optimizer_D_grad) self.disc_optimizers.append(self.optimizer_D_grad) if self.spsr_enabled: self.get_grad_nopadding = ImageGradientNoPadding().to(self.device) [self.netG, self.netD, self.netD_grad, self.get_grad_nopadding], \ [self.optimizer_G, self.optimizer_D, self.optimizer_D_grad] = \ amp.initialize([self.netG, self.netD, self.netD_grad, self.get_grad_nopadding], [self.optimizer_G, self.optimizer_D, self.optimizer_D_grad], opt_level=self.amp_level, num_losses=3) else: # AMP [self.netG, self.netD], [self.optimizer_G, self.optimizer_D] = \ amp.initialize([self.netG, self.netD], [self.optimizer_G, self.optimizer_D], opt_level=self.amp_level, num_losses=3) # DataParallel if opt['dist']: self.netG = DistributedDataParallel(self.netG, device_ids=[torch.cuda.current_device()], find_unused_parameters=True) else: self.netG = DataParallel(self.netG) if self.is_train: if opt['dist']: self.netD = DistributedDataParallel(self.netD, device_ids=[torch.cuda.current_device()], find_unused_parameters=True) if self.spsr_enabled: self.netD_grad = DistributedDataParallel(self.netD_grad, device_ids=[torch.cuda.current_device()], find_unused_parameters=True) else: self.netD = DataParallel(self.netD) if self.spsr_enabled: self.netD_grad = DataParallel(self.netD_grad) self.netG.train() self.netD.train() if self.spsr_enabled: self.netD_grad.train() # schedulers if train_opt['lr_scheme'] == 'MultiStepLR': # This is a recent change. assert to make sure any legacy configs dont find their way here. assert 'gen_lr_steps' in train_opt.keys() and 'disc_lr_steps' in train_opt.keys() self.schedulers.append( lr_scheduler.MultiStepLR_Restart(self.optimizer_G, train_opt['gen_lr_steps'], restarts=train_opt['restarts'], weights=train_opt['restart_weights'], gamma=train_opt['lr_gamma'], clear_state=train_opt['clear_state'], force_lr=train_opt['force_lr'])) for o in self.disc_optimizers: self.schedulers.append( lr_scheduler.MultiStepLR_Restart(o, train_opt['disc_lr_steps'], restarts=train_opt['restarts'], weights=train_opt['restart_weights'], gamma=train_opt['lr_gamma'], clear_state=train_opt['clear_state'], force_lr=train_opt['force_lr'])) elif train_opt['lr_scheme'] == 'ProgressiveMultiStepLR': # Only supported when there are two optimizers: G and D. assert len(self.optimizers) == 2 self.schedulers.append(lr_scheduler.ProgressiveMultiStepLR(self.optimizer_G, train_opt['gen_lr_steps'], self.netG.module.get_progressive_starts(), train_opt['lr_gamma'])) for o in self.disc_optimizers: self.schedulers.append(lr_scheduler.ProgressiveMultiStepLR(o, train_opt['disc_lr_steps'], [0], train_opt['lr_gamma'])) elif train_opt['lr_scheme'] == 'CosineAnnealingLR_Restart': for optimizer in self.optimizers: self.schedulers.append( lr_scheduler.CosineAnnealingLR_Restart( optimizer, train_opt['T_period'], eta_min=train_opt['eta_min'], restarts=train_opt['restarts'], weights=train_opt['restart_weights'])) else: raise NotImplementedError('MultiStepLR learning rate scheme is enough.') self.log_dict = OrderedDict() # Swapout params self.swapout_G_freq = train_opt['swapout_G_freq'] if train_opt['swapout_G_freq'] else 0 self.swapout_G_duration = 0 self.swapout_D_freq = train_opt['swapout_D_freq'] if train_opt['swapout_D_freq'] else 0 self.swapout_D_duration = 0 self.swapout_duration = train_opt['swapout_duration'] if train_opt['swapout_duration'] else 0 # GAN LQ image params self.gan_lq_img_use_prob = train_opt['gan_lowres_use_probability'] if train_opt['gan_lowres_use_probability'] else 0 self.img_debug_steps = opt['logger']['img_debug_steps'] if 'img_debug_steps' in opt['logger'].keys() else 50 else: self.netF = networks.define_F(use_bn=False).to(self.device) self.cri_fea = nn.L1Loss().to(self.device) #self.print_network() # print network self.load() # load G and D if needed self.load_random_corruptor() # Setting this to false triggers SRGAN to call the models update_model() function on the first iteration. self.updated = True def feed_data(self, data, need_GT=True): _profile = True if _profile: from time import time _t = time() # Corrupt the data with the given corruptor, if specified. self.fed_LQ = data['LQ'].to(self.device) if self.netC and random.random() < self.corruptor_usage_prob: with torch.no_grad(): corrupted_L = self.netC(self.fed_LQ)[0].detach() else: corrupted_L = self.fed_LQ self.var_L = torch.chunk(corrupted_L, chunks=self.mega_batch_factor, dim=0) if need_GT: self.var_H = [t.to(self.device) for t in torch.chunk(data['GT'], chunks=self.mega_batch_factor, dim=0)] input_ref = data['ref'] if 'ref' in data.keys() else data['GT'] self.var_ref = [t.to(self.device) for t in torch.chunk(input_ref, chunks=self.mega_batch_factor, dim=0)] input_pix = data['PIX'] if 'pix' in data.keys() else data['GT'] self.pix = [t.to(self.device) for t in torch.chunk(input_pix, chunks=self.mega_batch_factor, dim=0)] if 'GAN' in data.keys(): self.gan_img = [t.to(self.device) for t in torch.chunk(data['GAN'], chunks=self.mega_batch_factor, dim=0)] else: # If not provided, use provided LQ for anyplace where the GAN would have been used. self.gan_img = self.var_L if not self.updated: self.netG.module.update_model(self.optimizer_G, self.schedulers[0]) self.updated = True def optimize_parameters(self, step): _profile = False if _profile: from time import time _t = time() # Some generators have variants depending on the current step. if hasattr(self.netG.module, "update_for_step"): self.netG.module.update_for_step(step, os.path.join(self.opt['path']['models'], "..")) if hasattr(self.netD.module, "update_for_step"): self.netD.module.update_for_step(step, os.path.join(self.opt['path']['models'], "..")) # G for p in self.netD.parameters(): p.requires_grad = False if self.spsr_enabled: for p in self.netD_grad.parameters(): p.requires_grad = False self.swapout_D(step) self.swapout_G(step) # Turning off G-grad is required to enable mega-batching and D_update_ratio to work together for some reason. if step % self.D_update_ratio == 0 and step >= self.D_init_iters: if self.spsr_enabled and self.branch_pretrain and step < self.branch_init_iters: for k, v in self.netG.named_parameters(): if v.dtype != torch.int64 and v.dtype != torch.bool: v.requires_grad = '_branch_pretrain' in k else: for p in self.netG.parameters(): if p.dtype != torch.int64 and p.dtype != torch.bool: p.requires_grad = True else: for p in self.netG.parameters(): p.requires_grad = False # Calculate a standard deviation for the gaussian noise to be applied to the discriminator, termed noise-theta. if self.D_noise_final == 0: noise_theta = 0 else: noise_theta = (self.D_noise_theta - self.D_noise_theta_floor) * (self.D_noise_final - min(step, self.D_noise_final)) / self.D_noise_final + self.D_noise_theta_floor if _profile: print("Misc setup %f" % (time() - _t,)) _t = time() self.optimizer_G.zero_grad() self.fake_GenOut = [] self.fea_GenOut = [] self.fake_H = [] self.spsr_grad_GenOut = [] var_ref_skips = [] for var_L, var_LGAN, var_H, var_ref, pix in zip(self.var_L, self.gan_img, self.var_H, self.var_ref, self.pix): if self.spsr_enabled: using_gan_img = False # SPSR models have outputs from three different branches. fake_H_branch, fake_GenOut, grad_LR = self.netG(var_L) fea_GenOut = fake_GenOut self.spsr_grad_GenOut.append(fake_H_branch) # Get image gradients for later use. fake_H_grad = self.get_grad_nopadding(fake_GenOut) else: if random.random() > self.gan_lq_img_use_prob: fea_GenOut, fake_GenOut = self.netG(var_L) using_gan_img = False else: fea_GenOut, fake_GenOut = self.netG(var_LGAN) using_gan_img = True if _profile: print("Gen forward %f" % (time() - _t,)) _t = time() self.fake_GenOut.append(fake_GenOut.detach()) self.fea_GenOut.append(fea_GenOut.detach()) l_g_total = 0 if step % self.D_update_ratio == 0 and step >= self.D_init_iters: fea_w = self.l_fea_sched.get_weight_for_step(step) l_g_pix_log = None l_g_fea_log = None l_g_fdpl = None l_g_fea_log = None if self.cri_pix and not using_gan_img: # pixel loss l_g_pix = self.l_pix_w * self.cri_pix(fea_GenOut, pix) l_g_pix_log = l_g_pix / self.l_pix_w l_g_total += l_g_pix if self.spsr_enabled and self.cri_pix_grad: # gradient pixel loss if self.disjoint_data: grad_truth = self.get_grad_nopadding(var_L) grad_pred = F.interpolate(fake_H_grad, size=grad_truth.shape[2:], mode="nearest") else: grad_truth = self.get_grad_nopadding(var_H) grad_pred = fake_H_grad l_g_pix_grad = self.l_pix_grad_w * self.cri_pix_grad(grad_pred, grad_truth) l_g_total += l_g_pix_grad if self.spsr_enabled and self.cri_pix_branch: # branch pixel loss if self.disjoint_data: grad_truth = self.get_grad_nopadding(var_L) grad_pred = F.interpolate(fake_H_branch, size=grad_truth.shape[2:], mode="nearest") else: grad_truth = self.get_grad_nopadding(var_H) grad_pred = fake_H_branch l_g_pix_grad_branch = self.l_pix_branch_w * self.cri_pix_branch(grad_pred, grad_truth) l_g_total += l_g_pix_grad_branch if self.fdpl_enabled and not using_gan_img: l_g_fdpl = self.cri_fdpl(fea_GenOut, pix) l_g_total += l_g_fdpl * self.fdpl_weight if self.cri_fea and not using_gan_img and fea_w > 0: # feature loss if self.lr_netF is not None: real_fea = self.lr_netF(var_L, interpolate_factor=self.opt['scale']) elif self.use_corrupted_feature_input: cor_Pix = F.interpolate(self.feature_corruptor(pix), size=var_L.shape[2:]) real_fea = self.netF(cor_Pix).detach() else: real_fea = self.netF(pix).detach() if self.use_corrupted_feature_input: fake_fea = self.netF(F.interpolate(self.feature_corruptor(fea_GenOut), size=var_L.shape[2:])) else: fake_fea = self.netF(fea_GenOut) l_g_fea = fea_w * self.cri_fea(fake_fea, real_fea) l_g_fea_log = l_g_fea / fea_w l_g_total += l_g_fea if _profile: print("Fea forward %f" % (time() - _t,)) _t = time() # Note to future self: The BCELoss(0, 1) and BCELoss(0, 0) = .6931 # Effectively this means that the generator has only completely "won" when l_d_real and l_d_fake is # equal to this value. If I ever come up with an algorithm that tunes fea/gan weights automatically, # it should target this l_g_fix_disc = torch.zeros(1, requires_grad=False, device=self.device).squeeze() for fixed_disc in self.fixed_disc_nets: weight = fixed_disc.module.fdisc_weight real_fea = fixed_disc(pix).detach() fake_fea = fixed_disc(fea_GenOut) l_g_fix_disc = l_g_fix_disc + weight * self.cri_fea(fake_fea, real_fea) l_g_total += l_g_fix_disc if self.l_gan_w > 0: if self.opt['train']['gan_type'] in ['gan', 'pixgan', 'pixgan_fea', 'crossgan']: if self.opt['train']['gan_type'] == 'crossgan': pred_g_fake = self.netD(fake_GenOut, var_L) else: pred_g_fake = self.netD(fake_GenOut) l_g_gan = self.l_gan_w * self.cri_gan(pred_g_fake, True) elif self.opt['train']['gan_type'] == 'ragan': pred_d_real = self.netD(var_ref).detach() pred_g_fake = self.netD(fake_GenOut) l_g_gan = self.l_gan_w * ( self.cri_gan(pred_d_real - torch.mean(pred_g_fake), False) + self.cri_gan(pred_g_fake - torch.mean(pred_d_real), True)) / 2 l_g_gan_log = l_g_gan / self.l_gan_w l_g_total += l_g_gan if self.spsr_enabled and self.cri_grad_gan: if self.opt['train']['gan_type'] == 'crossgan': pred_g_fake_grad = self.netD_grad(fake_H_grad, var_L) pred_g_fake_grad_branch = self.netD_grad(fake_H_branch, var_L) else: pred_g_fake_grad = self.netD_grad(fake_H_grad) pred_g_fake_grad_branch = self.netD_grad(fake_H_branch) if self.opt['train']['gan_type'] in ['gan', 'pixgan', 'pixgan_fea', 'crossgan']: l_g_gan_grad = self.l_gan_grad_w * self.cri_grad_gan(pred_g_fake_grad, True) l_g_gan_grad_branch = self.l_gan_grad_w * self.cri_grad_gan(pred_g_fake_grad_branch, True) elif self.opt['train']['gan_type'] == 'ragan': pred_g_real_grad = self.netD_grad(self.get_grad_nopadding(var_ref)).detach() l_g_gan_grad = self.l_gan_grad_w * ( self.cri_gan(pred_g_real_grad - torch.mean(pred_g_fake_grad), False) + self.cri_gan(pred_g_fake_grad - torch.mean(pred_g_real_grad), True)) / 2 l_g_gan_grad_branch = self.l_gan_grad_w * ( self.cri_gan(pred_g_real_grad - torch.mean(pred_g_fake_grad_branch), False) + self.cri_gan(pred_g_fake_grad_branch - torch.mean(pred_g_real_grad), True)) / 2 l_g_total += l_g_gan_grad + l_g_gan_grad_branch # Scale the loss down by the batch factor. l_g_total_log = l_g_total l_g_total = l_g_total / self.mega_batch_factor with amp.scale_loss(l_g_total, self.optimizer_G, loss_id=0) as l_g_total_scaled: l_g_total_scaled.backward() if _profile: print("Gen backward %f" % (time() - _t,)) _t = time() self.optimizer_G.step() if _profile: print("Gen step %f" % (time() - _t,)) _t = time() # D if self.l_gan_w > 0 and step >= self.G_warmup: for p in self.netD.parameters(): if p.dtype != torch.int64 and p.dtype != torch.bool: p.requires_grad = True noise = torch.randn_like(var_ref) * noise_theta noise.to(self.device) real_disc_images = [] fake_disc_images = [] for fake_GenOut, var_LGAN, var_L, var_H, var_ref, pix in zip(self.fake_GenOut, self.gan_img, self.var_L, self.var_H, self.var_ref, self.pix): if random.random() > self.gan_lq_img_use_prob: fake_H = fake_GenOut.clone().detach().requires_grad_(False) else: # Re-compute generator outputs with the GAN inputs. with torch.no_grad(): if self.spsr_enabled: _, fake_H, _ = self.netG(var_LGAN) else: _, fake_H = self.netG(var_LGAN) fake_H = fake_H.detach() if _profile: print("Gen forward for disc %f" % (time() - _t,)) _t = time() # Apply noise to the inputs to slow discriminator convergence. var_ref = var_ref + noise fake_H = fake_H + noise l_d_fea_real = 0 l_d_fea_fake = 0 self.optimizer_D.zero_grad() if self.opt['train']['gan_type'] == 'pixgan_fea': # Compute a feature loss which is added to the GAN loss computed later to guide the discriminator better. disc_fea_scale = .1 _, fea_real = self.netD(var_ref, output_feature_vector=True) actual_fea = self.netF(var_ref) l_d_fea_real = self.cri_fea(fea_real, actual_fea) * disc_fea_scale / self.mega_batch_factor _, fea_fake = self.netD(fake_H, output_feature_vector=True) actual_fea = self.netF(fake_H) l_d_fea_fake = self.cri_fea(fea_fake, actual_fea) * disc_fea_scale / self.mega_batch_factor if self.opt['train']['gan_type'] == 'crossgan': # need to forward and backward separately, since batch norm statistics differ # real pred_d_real = self.netD(var_ref, var_L) l_d_real = self.cri_gan(pred_d_real, True) l_d_real_log = l_d_real # fake pred_d_fake = self.netD(fake_H, var_L) l_d_fake = self.cri_gan(pred_d_fake, False) l_d_fake_log = l_d_fake # mismatched mismatched_L = torch.roll(var_L, shifts=1, dims=0) pred_d_real_mismatched = self.netD(var_ref, mismatched_L) pred_d_fake_mismatched = self.netD(fake_H, mismatched_L) l_d_mismatched = (self.cri_gan(pred_d_real_mismatched, False) + self.cri_gan(pred_d_fake_mismatched, False)) / 2 l_d_total = (l_d_real + l_d_fake + l_d_mismatched) / 3 l_d_total = l_d_total / self.mega_batch_factor with amp.scale_loss(l_d_total, self.optimizer_D, loss_id=1) as l_d_total_scaled: l_d_total_scaled.backward() elif self.opt['train']['gan_type'] == 'gan': # real pred_d_real = self.netD(var_ref) l_d_real = self.cri_gan(pred_d_real, True) / self.mega_batch_factor l_d_real_log = l_d_real * self.mega_batch_factor # fake pred_d_fake = self.netD(fake_H) l_d_fake = self.cri_gan(pred_d_fake, False) / self.mega_batch_factor l_d_fake_log = l_d_fake * self.mega_batch_factor l_d_total = (l_d_real + l_d_fake) / 2 with amp.scale_loss(l_d_total, self.optimizer_D, loss_id=1) as l_d_total_scaled: l_d_total_scaled.backward() elif 'pixgan' in self.opt['train']['gan_type']: pixdisc_channels, pixdisc_output_reduction = self.netD.module.pixgan_parameters() disc_output_shape = (var_ref.shape[0], pixdisc_channels, var_ref.shape[2] // pixdisc_output_reduction, var_ref.shape[3] // pixdisc_output_reduction) b, _, w, h = var_ref.shape real = torch.ones((b, pixdisc_channels, w, h), device=var_ref.device) fake = torch.zeros((b, pixdisc_channels, w, h), device=var_ref.device) if not self.disjoint_data: # randomly determine portions of the image to swap to keep the discriminator honest. SWAP_MAX_DIM = w // 4 SWAP_MIN_DIM = 16 assert SWAP_MAX_DIM > 0 if random.random() > .5: # Make this only happen half the time. Earlier experiments had it happen # more often and the model was "cheating" by using the presence of # easily discriminated fake swaps to count the entire generated image # as fake. random_swap_count = random.randint(0, 4) for i in range(random_swap_count): # Make the swap across fake_H and var_ref swap_x, swap_y = random.randint(0, w - SWAP_MIN_DIM), random.randint(0, h - SWAP_MIN_DIM) swap_w, swap_h = random.randint(SWAP_MIN_DIM, SWAP_MAX_DIM), random.randint(SWAP_MIN_DIM, SWAP_MAX_DIM) if swap_x + swap_w > w: swap_w = w - swap_x if swap_y + swap_h > h: swap_h = h - swap_y t = fake_H[:, :, swap_x:(swap_x+swap_w), swap_y:(swap_y+swap_h)].clone() fake_H[:, :, swap_x:(swap_x+swap_w), swap_y:(swap_y+swap_h)] = var_ref[:, :, swap_x:(swap_x+swap_w), swap_y:(swap_y+swap_h)] var_ref[:, :, swap_x:(swap_x+swap_w), swap_y:(swap_y+swap_h)] = t real[:, :, swap_x:(swap_x+swap_w), swap_y:(swap_y+swap_h)] = 0.0 fake[:, :, swap_x:(swap_x+swap_w), swap_y:(swap_y+swap_h)] = 1.0 # Interpolate down to the dimensionality that the discriminator uses. real = F.interpolate(real, size=disc_output_shape[2:], mode="bilinear", align_corners=False) fake = F.interpolate(fake, size=disc_output_shape[2:], mode="bilinear", align_corners=False) # We're also assuming that this is exactly how the flattened discriminator output is generated. real = real.view(-1, 1) fake = fake.view(-1, 1) # real pred_d_real = self.netD(var_ref) l_d_real = self.cri_gan(pred_d_real, real) / self.mega_batch_factor l_d_real_log = l_d_real * self.mega_batch_factor l_d_real += l_d_fea_real # fake pred_d_fake = self.netD(fake_H) l_d_fake = self.cri_gan(pred_d_fake, fake) / self.mega_batch_factor l_d_fake_log = l_d_fake * self.mega_batch_factor l_d_fake += l_d_fea_fake l_d_total = (l_d_real + l_d_fake) / 2 with amp.scale_loss(l_d_total, self.optimizer_D, loss_id=1) as l_d_total_scaled: l_d_total_scaled.backward() pdr = pred_d_real.detach() + torch.abs(torch.min(pred_d_real)) pdr = pdr / torch.max(pdr) real_disc_images.append(pdr.view(disc_output_shape)) pdf = pred_d_fake.detach() + torch.abs(torch.min(pred_d_fake)) pdf = pdf / torch.max(pdf) fake_disc_images.append(pdf.view(disc_output_shape)) elif self.opt['train']['gan_type'] == 'ragan': pred_d_fake = self.netD(fake_H) pred_d_real = self.netD(var_ref) l_d_real = self.cri_gan(pred_d_real - torch.mean(pred_d_fake), True) l_d_real_log = l_d_real l_d_fake = self.cri_gan(pred_d_fake - torch.mean(pred_d_real), False) l_d_fake_log = l_d_fake l_d_total = (l_d_real + l_d_fake) / 2 l_d_total /= self.mega_batch_factor with amp.scale_loss(l_d_total, self.optimizer_D, loss_id=1) as l_d_total_scaled: l_d_total_scaled.backward() var_ref_skips.append(var_ref.detach()) self.fake_H.append(fake_H.detach()) self.optimizer_D.step() if _profile: print("Disc step %f" % (time() - _t,)) _t = time() # D_grad. if self.spsr_enabled and self.cri_grad_gan and step >= self.G_warmup: for p in self.netD_grad.parameters(): p.requires_grad = True self.optimizer_D_grad.zero_grad() for var_L, var_ref, fake_H, fake_H_grad_branch in zip(self.var_L, var_ref_skips, self.fake_H, self.spsr_grad_GenOut): fake_H_grad = self.get_grad_nopadding(fake_H).detach() var_ref_grad = self.get_grad_nopadding(var_ref) fake_H_grad_branch = fake_H_grad_branch.detach() + noise if self.opt['train']['gan_type'] == 'crossgan': pred_d_real_grad = self.netD_grad(var_ref_grad, var_L) pred_d_fake_grad = self.netD_grad(fake_H_grad, var_L) # Tensor already detached above. # var_ref and fake_H already has noise added to it. We **must** add noise to fake_H_grad_branch too. pred_d_fake_grad_branch = self.netD_grad(fake_H_grad_branch, var_L) else: pred_d_real_grad = self.netD_grad(var_ref_grad) pred_d_fake_grad = self.netD_grad(fake_H_grad) # Tensor already detached above. # var_ref and fake_H already has noise added to it. We **must** add noise to fake_H_grad_branch too. pred_d_fake_grad_branch = self.netD_grad(fake_H_grad_branch) if self.opt['train']['gan_type'] == 'gan' or self.opt['train']['gan_type'] == 'crossgan': l_d_real_grad = self.cri_gan(pred_d_real_grad, True) l_d_fake_grad = (self.cri_gan(pred_d_fake_grad, False) + self.cri_gan(pred_d_fake_grad_branch, False)) / 2 elif self.opt['train']['gan_type'] == 'pixgan': real = torch.ones_like(pred_d_real_grad) fake = torch.zeros_like(pred_d_fake_grad) l_d_real_grad = self.cri_grad_gan(pred_d_real_grad, real) l_d_fake_grad = (self.cri_grad_gan(pred_d_fake_grad, fake) + \ self.cri_grad_gan(pred_d_fake_grad_branch, fake)) / 2 elif self.opt['train']['gan_type'] == 'ragan': l_d_real_grad = self.cri_grad_gan(pred_d_real_grad - torch.mean(pred_d_fake_grad), True) l_d_fake_grad = (self.cri_grad_gan(pred_d_fake_grad - torch.mean(pred_d_real_grad), False) + \ self.cri_grad_gan(pred_d_fake_grad_branch - torch.mean(pred_d_real_grad), False)) / 2 l_d_total_grad = (l_d_real_grad + l_d_fake_grad) / 2 l_d_total_grad /= self.mega_batch_factor with amp.scale_loss(l_d_total_grad, self.optimizer_D_grad, loss_id=2) as l_d_total_grad_scaled: l_d_total_grad_scaled.backward() self.optimizer_D_grad.step() # Log sample images from first microbatch. if step % self.img_debug_steps == 0 and self.rank <= 0: sample_save_path = os.path.join(self.opt['path']['models'], "..", "temp") os.makedirs(os.path.join(sample_save_path, "hr"), exist_ok=True) os.makedirs(os.path.join(sample_save_path, "lr"), exist_ok=True) os.makedirs(os.path.join(sample_save_path, "gen_fea"), exist_ok=True) os.makedirs(os.path.join(sample_save_path, "gen"), exist_ok=True) os.makedirs(os.path.join(sample_save_path, "disc_fake"), exist_ok=True) os.makedirs(os.path.join(sample_save_path, "pix"), exist_ok=True) os.makedirs(os.path.join(sample_save_path, "disc"), exist_ok=True) os.makedirs(os.path.join(sample_save_path, "ref"), exist_ok=True) if self.spsr_enabled: os.makedirs(os.path.join(sample_save_path, "gen_grad"), exist_ok=True) # fed_LQ is not chunked. for i in range(self.mega_batch_factor): utils.save_image(self.var_H[i].cpu(), os.path.join(sample_save_path, "hr", "%05i_%02i.png" % (step, i))) utils.save_image(self.var_L[i].cpu(), os.path.join(sample_save_path, "lr", "%05i_%02i.png" % (step, i))) utils.save_image(self.pix[i].cpu(), os.path.join(sample_save_path, "pix", "%05i_%02i.png" % (step, i))) utils.save_image(self.fake_GenOut[i].cpu(), os.path.join(sample_save_path, "gen", "%05i_%02i.png" % (step, i))) utils.save_image(self.fea_GenOut[i].cpu(), os.path.join(sample_save_path, "gen_fea", "%05i_%02i.png" % (step, i))) if self.spsr_enabled: utils.save_image(self.spsr_grad_GenOut[i].cpu(), os.path.join(sample_save_path, "gen_grad", "%05i_%02i.png" % (step, i))) if self.l_gan_w > 0 and step >= self.G_warmup and 'pixgan' in self.opt['train']['gan_type']: utils.save_image(var_ref_skips[i].cpu(), os.path.join(sample_save_path, "ref", "%05i_%02i.png" % (step, i))) utils.save_image(self.fake_H[i], os.path.join(sample_save_path, "disc_fake", "fake%05i_%02i.png" % (step, i))) utils.save_image(F.interpolate(fake_disc_images[i], scale_factor=4), os.path.join(sample_save_path, "disc", "fake%05i_%02i.png" % (step, i))) utils.save_image(F.interpolate(real_disc_images[i], scale_factor=4), os.path.join(sample_save_path, "disc", "real%05i_%02i.png" % (step, i))) # Log metrics if step % self.D_update_ratio == 0 and step >= self.D_init_iters: if self.cri_pix and l_g_pix_log is not None: self.add_log_entry('l_g_pix', l_g_pix_log.detach().item()) if self.fdpl_enabled and l_g_fdpl is not None: self.add_log_entry('l_g_fdpl', l_g_fdpl.detach().item()) if self.cri_fea and l_g_fea_log is not None: self.add_log_entry('feature_weight', fea_w) self.add_log_entry('l_g_fea', l_g_fea_log.detach().item()) self.add_log_entry('l_g_fix_disc', l_g_fix_disc.detach().item()) if self.l_gan_w > 0: self.add_log_entry('l_g_gan', l_g_gan_log.detach().item()) self.add_log_entry('l_g_total', l_g_total_log.detach().item()) if self.opt['train']['gan_type'] == 'pixgan_fea': self.add_log_entry('l_d_fea_fake', l_d_fea_fake.detach().item() * self.mega_batch_factor) self.add_log_entry('l_d_fea_real', l_d_fea_real.detach().item() * self.mega_batch_factor) self.add_log_entry('l_d_fake_total', l_d_fake.detach().item() * self.mega_batch_factor) self.add_log_entry('l_d_real_total', l_d_real.detach().item() * self.mega_batch_factor) if self.opt['train']['gan_type'] == 'crossgan': self.add_log_entry('l_d_mismatched', l_d_mismatched.detach().item()) if self.spsr_enabled: if self.cri_pix_grad: self.add_log_entry('l_g_pix_grad_branch', l_g_pix_grad.detach().item()) if self.cri_pix_branch: self.add_log_entry('l_g_pix_grad_branch', l_g_pix_grad_branch.detach().item()) if self.cri_grad_gan: self.add_log_entry('l_g_gan_grad', l_g_gan_grad.detach().item() / self.l_gan_grad_w) self.add_log_entry('l_g_gan_grad_branch', l_g_gan_grad_branch.detach().item() / self.l_gan_grad_w) if self.l_gan_w > 0 and step >= self.G_warmup: self.add_log_entry('l_d_real', l_d_real_log.detach().item()) self.add_log_entry('l_d_fake', l_d_fake_log.detach().item()) self.add_log_entry('D_fake', torch.mean(pred_d_fake.detach())) self.add_log_entry('D_diff', torch.mean(pred_d_fake.detach()) - torch.mean(pred_d_real.detach())) if self.spsr_enabled and self.cri_grad_gan: self.add_log_entry('l_d_real_grad', l_d_real_grad.detach().item()) self.add_log_entry('l_d_fake_grad', l_d_fake_grad.detach().item()) self.add_log_entry('D_fake_grad', torch.mean(pred_d_fake_grad.detach())) self.add_log_entry('D_diff_grad', torch.mean(pred_d_fake_grad.detach()) - torch.mean(pred_d_real_grad.detach())) # Log learning rates. for i, pg in enumerate(self.optimizer_G.param_groups): self.add_log_entry('gen_lr_%i' % (i,), pg['lr']) for i, pg in enumerate(self.optimizer_D.param_groups): self.add_log_entry('disc_lr_%i' % (i,), pg['lr']) if step % self.corruptor_swapout_steps == 0 and step > 0: self.load_random_corruptor() # Allows the log to serve as an easy-to-use rotating buffer. def add_log_entry(self, key, value): key_it = "%s_it" % (key,) log_rotating_buffer_size = 50 if key not in self.log_dict.keys(): self.log_dict[key] = [] self.log_dict[key_it] = 0 if len(self.log_dict[key]) < log_rotating_buffer_size: self.log_dict[key].append(value) else: self.log_dict[key][self.log_dict[key_it] % log_rotating_buffer_size] = value self.log_dict[key_it] += 1 def pick_rand_prev_model(self, model_suffix): previous_models = glob.glob(os.path.join(self.opt['path']['models'], "*_%s.pth" % (model_suffix,))) if len(previous_models) <= 1: return None # Just a note: this intentionally includes the swap model in the list of possibilities. return previous_models[random.randint(0, len(previous_models)-1)] def compute_fea_loss(self, real, fake): with torch.no_grad(): real = real.unsqueeze(dim=0).to(self.device) fake = fake.unsqueeze(dim=0).to(self.device) real_fea = self.netF(real).detach() fake_fea = self.netF(fake) return self.cri_fea(fake_fea, real_fea).item() # Called before verification/checkpoint to ensure we're using the real models and not a swapout variant. def force_restore_swapout(self): if self.swapout_D_duration > 0: logger.info("Swapping back to current D model: %s" % (self.stashed_D,)) self.load_network(self.stashed_D, self.netD, self.opt['path']['strict_load']) self.stashed_D = None self.swapout_D_duration = 0 if self.swapout_G_duration > 0: logger.info("Swapping back to current G model: %s" % (self.stashed_G,)) self.load_network(self.stashed_G, self.netG, self.opt['path']['strict_load']) self.stashed_G = None self.swapout_G_duration = 0 def swapout_D(self, step): if self.swapout_D_duration > 0: self.swapout_D_duration -= 1 if self.swapout_D_duration == 0: # Swap back. logger.info("Swapping back to current D model: %s" % (self.stashed_D,)) self.load_network(self.stashed_D, self.netD, self.opt['path']['strict_load']) self.stashed_D = None elif self.swapout_D_freq != 0 and step % self.swapout_D_freq == 0: swapped_model = self.pick_rand_prev_model('D') if swapped_model is not None: logger.info("Swapping to previous D model: %s" % (swapped_model,)) self.stashed_D = self.save_network(self.netD, 'D', 'swap_model') self.load_network(swapped_model, self.netD, self.opt['path']['strict_load']) self.swapout_D_duration = self.swapout_duration def swapout_G(self, step): if self.swapout_G_duration > 0: self.swapout_G_duration -= 1 if self.swapout_G_duration == 0: # Swap back. logger.info("Swapping back to current G model: %s" % (self.stashed_G,)) self.load_network(self.stashed_G, self.netG, self.opt['path']['strict_load']) self.stashed_G = None elif self.swapout_G_freq != 0 and step % self.swapout_G_freq == 0: swapped_model = self.pick_rand_prev_model('G') if swapped_model is not None: logger.info("Swapping to previous G model: %s" % (swapped_model,)) self.stashed_G = self.save_network(self.netG, 'G', 'swap_model') self.load_network(swapped_model, self.netG, self.opt['path']['strict_load']) self.swapout_G_duration = self.swapout_duration def test(self): self.netG.eval() with torch.no_grad(): if self.spsr_enabled: self.fake_H_branch = [] self.fake_GenOut = [] self.grad_LR = [] fake_H_branch, fake_GenOut, grad_LR = self.netG(self.var_L[0]) self.fake_H_branch.append(fake_H_branch) self.fake_GenOut.append(fake_GenOut) self.grad_LR.append(grad_LR) else: self.fake_GenOut = [self.netG(self.var_L[0])] self.netG.train() # Fetches a summary of the log. def get_current_log(self, step): return_log = {} for k in self.log_dict.keys(): if not isinstance(self.log_dict[k], list): continue return_log[k] = sum(self.log_dict[k]) / len(self.log_dict[k]) # Some generators can do their own metric logging. if hasattr(self.netG.module, "get_debug_values"): return_log.update(self.netG.module.get_debug_values(step)) if hasattr(self.netD.module, "get_debug_values"): return_log.update(self.netD.module.get_debug_values(step)) return return_log def get_current_visuals(self, need_GT=True): out_dict = OrderedDict() out_dict['LQ'] = self.var_L[0].detach().float().cpu() gen_batch = self.fake_GenOut[0] if isinstance(gen_batch, tuple): gen_batch = gen_batch[0] out_dict['rlt'] = gen_batch.detach().float().cpu() if need_GT: out_dict['GT'] = self.var_H[0].detach().float().cpu() if self.spsr_enabled: out_dict['SR_branch'] = self.fake_H_branch[0].float().cpu() out_dict['LR_grad'] = self.grad_LR[0].float().cpu() return out_dict def print_network(self): # Generator s, n = self.get_network_description(self.netG) if isinstance(self.netG, nn.DataParallel) or isinstance(self.netG, DistributedDataParallel): net_struc_str = '{} - {}'.format(self.netG.__class__.__name__, self.netG.module.__class__.__name__) else: net_struc_str = '{}'.format(self.netG.__class__.__name__) if self.rank <= 0: logger.info('Network G structure: {}, with parameters: {:,d}'.format(net_struc_str, n)) logger.info(s) if self.is_train: # Discriminator s, n = self.get_network_description(self.netD) if isinstance(self.netD, nn.DataParallel) or isinstance(self.netD, DistributedDataParallel): net_struc_str = '{} - {}'.format(self.netD.__class__.__name__, self.netD.module.__class__.__name__) else: net_struc_str = '{}'.format(self.netD.__class__.__name__) if self.rank <= 0: logger.info('Network D structure: {}, with parameters: {:,d}'.format( net_struc_str, n)) logger.info(s) if self.cri_fea: # F, Perceptual Network s, n = self.get_network_description(self.netF) if isinstance(self.netF, nn.DataParallel) or isinstance( self.netF, DistributedDataParallel): net_struc_str = '{} - {}'.format(self.netF.__class__.__name__, self.netF.module.__class__.__name__) else: net_struc_str = '{}'.format(self.netF.__class__.__name__) if self.rank <= 0: logger.info('Network F structure: {}, with parameters: {:,d}'.format( net_struc_str, n)) logger.info(s) def load(self): load_path_G = self.opt['path']['pretrain_model_G'] if load_path_G is not None: logger.info('Loading model for G [{:s}] ...'.format(load_path_G)) self.load_network(load_path_G, self.netG, self.opt['path']['strict_load']) load_path_D = self.opt['path']['pretrain_model_D'] if self.opt['is_train'] and load_path_D is not None: logger.info('Loading model for D [{:s}] ...'.format(load_path_D)) self.load_network(load_path_D, self.netD, self.opt['path']['strict_load']) if self.spsr_enabled: load_path_D_grad = self.opt['path']['pretrain_model_D_grad'] if self.opt['is_train'] and load_path_D_grad is not None: logger.info('Loading pretrained model for D_grad [{:s}] ...'.format(load_path_D_grad)) self.load_network(load_path_D_grad, self.netD_grad) def load_random_corruptor(self): if self.netC is None: return corruptor_files = glob.glob(os.path.join(self.opt['path']['pretrained_corruptors_dir'], "*.pth")) corruptor_to_load = corruptor_files[random.randint(0, len(corruptor_files)-1)] logger.info('Swapping corruptor to: %s' % (corruptor_to_load,)) self.load_network(corruptor_to_load, self.netC, self.opt['path']['strict_load']) def save(self, iter_step): self.save_network(self.netG, 'G', iter_step) self.save_network(self.netD, 'D', iter_step) if self.spsr_enabled: self.save_network(self.netD_grad, 'D_grad', iter_step)