| import time
|
| from options.train_options import TrainOptions
|
| from models.networks import ResUnetGenerator, VGGLoss, save_checkpoint, load_checkpoint_parallel
|
| from models.afwm import TVLoss, AFWM
|
| import torch.nn as nn
|
| import torch.nn.functional as F
|
| import os
|
| import numpy as np
|
| import torch
|
| from torch.utils.data import DataLoader
|
| from torch.utils.data.distributed import DistributedSampler
|
| from tensorboardX import SummaryWriter
|
| import datetime
|
| import cv2
|
|
|
| opt = TrainOptions().parse()
|
| path = 'runs/' + opt.name
|
| os.makedirs(path, exist_ok=True)
|
|
|
|
|
| def CreateDataset(opt):
|
| from data.aligned_dataset import AlignedDataset
|
| dataset = AlignedDataset()
|
| print("dataset [%s] was created" % (dataset.name()))
|
| dataset.initialize(opt)
|
| return dataset
|
|
|
|
|
| os.makedirs('sample', exist_ok=True)
|
| opt = TrainOptions().parse()
|
| iter_path = os.path.join(opt.checkpoints_dir, opt.name, 'iter.txt')
|
|
|
| torch.cuda.set_device(opt.local_rank)
|
| torch.distributed.init_process_group(
|
| 'nccl',
|
| init_method='env://'
|
| )
|
| device = torch.device(f'cuda:{opt.local_rank}')
|
|
|
| start_epoch, epoch_iter = 1, 0
|
|
|
| train_data = CreateDataset(opt)
|
| train_sampler = DistributedSampler(train_data)
|
| train_loader = DataLoader(train_data, batch_size=opt.batchSize, shuffle=False,
|
| num_workers=4, pin_memory=True, sampler=train_sampler)
|
| dataset_size = len(train_loader)
|
|
|
| PF_warp_model = AFWM(opt, 3)
|
| print(PF_warp_model)
|
| PF_warp_model.train()
|
| PF_warp_model.cuda()
|
| load_checkpoint_parallel(PF_warp_model, opt.PFAFN_warp_checkpoint)
|
|
|
| PF_gen_model = ResUnetGenerator(7, 4, 5, ngf=64, norm_layer=nn.BatchNorm2d)
|
| print(PF_gen_model)
|
| PF_gen_model.train()
|
| PF_gen_model.cuda()
|
|
|
| PB_warp_model = AFWM(opt, 45)
|
| print(PB_warp_model)
|
| PB_warp_model.eval()
|
| PB_warp_model.cuda()
|
| load_checkpoint_parallel(PB_warp_model, opt.PBAFN_warp_checkpoint)
|
|
|
| PB_gen_model = ResUnetGenerator(8, 4, 5, ngf=64, norm_layer=nn.BatchNorm2d)
|
| print(PB_gen_model)
|
| PB_gen_model.eval()
|
| PB_gen_model.cuda()
|
| load_checkpoint_parallel(PB_gen_model, opt.PBAFN_gen_checkpoint)
|
|
|
| PF_warp_model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(PF_warp_model).to(device)
|
| PF_gen_model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(PF_gen_model).to(device)
|
|
|
| if opt.isTrain and len(opt.gpu_ids):
|
| PF_warp_model = torch.nn.parallel.DistributedDataParallel(PF_warp_model, device_ids=[opt.local_rank])
|
| PF_gen_model = torch.nn.parallel.DistributedDataParallel(PF_gen_model, device_ids=[opt.local_rank])
|
| PB_warp_model = torch.nn.parallel.DistributedDataParallel(PB_warp_model, device_ids=[opt.local_rank])
|
| PB_gen_model = torch.nn.parallel.DistributedDataParallel(PB_gen_model, device_ids=[opt.local_rank])
|
|
|
| criterionL1 = nn.L1Loss()
|
| criterionVGG = VGGLoss()
|
| criterionL2 = nn.MSELoss('sum')
|
|
|
| params_warp = [p for p in PF_warp_model.parameters()]
|
| params_gen = [p for p in PF_gen_model.parameters()]
|
| optimizer_warp = torch.optim.Adam(params_warp, lr=0.2 * opt.lr, betas=(opt.beta1, 0.999))
|
| optimizer_gen = torch.optim.Adam(params_gen, lr=opt.lr, betas=(opt.beta1, 0.999))
|
|
|
| total_steps = (start_epoch - 1) * dataset_size + epoch_iter
|
|
|
| if opt.local_rank == 0:
|
| writer = SummaryWriter(path)
|
|
|
| step = 0
|
| step_per_batch = dataset_size
|
|
|
| for epoch in range(start_epoch, opt.niter + opt.niter_decay + 1):
|
| epoch_start_time = time.time()
|
| if epoch != start_epoch:
|
| epoch_iter = epoch_iter % dataset_size
|
|
|
| train_sampler.set_epoch(epoch)
|
|
|
| for i, data in enumerate(train_loader):
|
|
|
| iter_start_time = time.time()
|
|
|
| total_steps += 1
|
| epoch_iter += 1
|
| save_fake = True
|
|
|
| t_mask = torch.FloatTensor((data['label'].cpu().numpy() == 7).astype(np.float))
|
| data['label'] = data['label'] * (1 - t_mask) + t_mask * 4
|
| edge = data['edge']
|
| pre_clothes_edge = torch.FloatTensor((edge.detach().numpy() > 0.5).astype(np.int))
|
| clothes = data['color']
|
| clothes = clothes * pre_clothes_edge
|
| edge_un = data['edge_un']
|
| pre_clothes_edge_un = torch.FloatTensor((edge_un.detach().numpy() > 0.5).astype(np.int))
|
| clothes_un = data['color_un']
|
| clothes_un = clothes_un * pre_clothes_edge_un
|
| person_clothes_edge = torch.FloatTensor((data['label'].cpu().numpy() == 4).astype(np.int))
|
| real_image = data['image']
|
| person_clothes = real_image * person_clothes_edge
|
| pose = data['pose']
|
| size = data['label'].size()
|
| oneHot_size1 = (size[0], 25, size[2], size[3])
|
| densepose = torch.cuda.FloatTensor(torch.Size(oneHot_size1)).zero_()
|
| densepose = densepose.scatter_(1, data['densepose'].data.long().cuda(), 1.0)
|
| densepose_fore = data['densepose'] / 24
|
| face_mask = torch.FloatTensor((data['label'].cpu().numpy() == 1).astype(np.int)) + torch.FloatTensor((data['label'].cpu().numpy() == 12).astype(np.int))
|
| other_clothes_mask = torch.FloatTensor((data['label'].cpu().numpy() == 5).astype(np.int)) + torch.FloatTensor((data['label'].cpu().numpy() == 6).astype(np.int)) \
|
| + torch.FloatTensor((data['label'].cpu().numpy() == 8).astype(np.int)) + torch.FloatTensor((data['label'].cpu().numpy() == 9).astype(np.int)) \
|
| + torch.FloatTensor((data['label'].cpu().numpy() == 10).astype(np.int))
|
| face_img = face_mask * real_image
|
| other_clothes_img = other_clothes_mask * real_image
|
| preserve_mask = torch.cat([face_mask, other_clothes_mask], 1)
|
|
|
| concat_un = torch.cat([preserve_mask.cuda(), densepose, pose.cuda()], 1)
|
| flow_out_un = PB_warp_model(concat_un.cuda(), clothes_un.cuda(), pre_clothes_edge_un.cuda())
|
| warped_cloth_un, last_flow_un, cond_un_all, flow_un_all, delta_list_un, x_all_un, x_edge_all_un, delta_x_all_un, delta_y_all_un = flow_out_un
|
| warped_prod_edge_un = F.grid_sample(pre_clothes_edge_un.cuda(), last_flow_un.permute(0, 2, 3, 1),
|
| mode='bilinear', padding_mode='zeros')
|
|
|
| flow_out_sup = PB_warp_model(concat_un.cuda(), clothes.cuda(), pre_clothes_edge.cuda())
|
| warped_cloth_sup, last_flow_sup, cond_sup_all, flow_sup_all, delta_list_sup, x_all_sup, x_edge_all_sup, delta_x_all_sup, delta_y_all_sup = flow_out_sup
|
|
|
| arm_mask = torch.FloatTensor((data['label'].cpu().numpy() == 11).astype(np.float)) + torch.FloatTensor((data['label'].cpu().numpy() == 13).astype(np.float))
|
| hand_mask = torch.FloatTensor((data['densepose'].cpu().numpy() == 3).astype(np.int)) + torch.FloatTensor((data['densepose'].cpu().numpy() == 4).astype(np.int))
|
| dense_preserve_mask = torch.FloatTensor((data['densepose'].cpu().numpy() == 15).astype(np.int)) + torch.FloatTensor((data['densepose'].cpu().numpy() == 16).astype(np.int)) \
|
| + torch.FloatTensor((data['densepose'].cpu().numpy() == 17).astype(np.int)) + torch.FloatTensor((data['densepose'].cpu().numpy() == 18).astype(np.int)) \
|
| + torch.FloatTensor((data['densepose'].cpu().numpy() == 19).astype(np.int)) + torch.FloatTensor((data['densepose'].cpu().numpy() == 20).astype(np.int)) \
|
| + torch.FloatTensor((data['densepose'].cpu().numpy() == 21).astype(np.int)) + torch.FloatTensor((data['densepose'].cpu().numpy() == 22))
|
| hand_img = (arm_mask * hand_mask) * real_image
|
| dense_preserve_mask = dense_preserve_mask.cuda() * (1 - warped_prod_edge_un)
|
| preserve_region = face_img + other_clothes_img + hand_img
|
|
|
| gen_inputs_un = torch.cat([preserve_region.cuda(), warped_cloth_un, warped_prod_edge_un, dense_preserve_mask], 1)
|
| gen_outputs_un = PB_gen_model(gen_inputs_un)
|
| p_rendered_un, m_composite_un = torch.split(gen_outputs_un, [3, 1], 1)
|
| p_rendered_un = torch.tanh(p_rendered_un)
|
| m_composite_un = torch.sigmoid(m_composite_un)
|
| m_composite_un = m_composite_un * warped_prod_edge_un
|
| p_tryon_un = warped_cloth_un * m_composite_un + p_rendered_un * (1 - m_composite_un)
|
|
|
| flow_out = PF_warp_model(p_tryon_un.detach(), clothes.cuda(), pre_clothes_edge.cuda())
|
| warped_cloth, last_flow, cond_all, flow_all, delta_list, x_all, x_edge_all, delta_x_all, delta_y_all = flow_out
|
| warped_prod_edge = x_edge_all[4]
|
|
|
| epsilon = 0.001
|
| loss_smooth = sum([TVLoss(x) for x in delta_list])
|
| loss_warp = 0
|
| loss_fea_sup_all = 0
|
| loss_flow_sup_all = 0
|
|
|
| l1_loss_batch = torch.abs(warped_cloth_sup.detach() - person_clothes.cuda())
|
| l1_loss_batch = l1_loss_batch.reshape(opt.batchSize, 3 * 256 * 192)
|
| l1_loss_batch = l1_loss_batch.sum(dim=1) / (3 * 256 * 192)
|
| l1_loss_batch_pred = torch.abs(warped_cloth.detach() - person_clothes.cuda())
|
| l1_loss_batch_pred = l1_loss_batch_pred.reshape(opt.batchSize, 3 * 256 * 192)
|
| l1_loss_batch_pred = l1_loss_batch_pred.sum(dim=1) / (3 * 256 * 192)
|
| weight = (l1_loss_batch < l1_loss_batch_pred).float()
|
| num_all = len(np.where(weight.cpu().numpy() > 0)[0])
|
| if num_all == 0:
|
| num_all = 1
|
|
|
| for num in range(5):
|
| cur_person_clothes = F.interpolate(person_clothes, scale_factor=0.5 ** (4 - num), mode='bilinear')
|
| cur_person_clothes_edge = F.interpolate(person_clothes_edge, scale_factor=0.5 ** (4 - num), mode='bilinear')
|
| loss_l1 = criterionL1(x_all[num], cur_person_clothes.cuda())
|
| loss_vgg = criterionVGG(x_all[num], cur_person_clothes.cuda())
|
| loss_edge = criterionL1(x_edge_all[num], cur_person_clothes_edge.cuda())
|
| b, c, h, w = delta_x_all[num].shape
|
| loss_flow_x = (delta_x_all[num].pow(2) + epsilon * epsilon).pow(0.45)
|
| loss_flow_x = torch.sum(loss_flow_x) / (b * c * h * w)
|
| loss_flow_y = (delta_y_all[num].pow(2) + epsilon * epsilon).pow(0.45)
|
| loss_flow_y = torch.sum(loss_flow_y) / (b * c * h * w)
|
| loss_second_smooth = loss_flow_x + loss_flow_y
|
| b1, c1, h1, w1 = cond_all[num].shape
|
| weight_all = weight.reshape(-1, 1, 1, 1).repeat(1, 256, h1, w1)
|
| cond_sup_loss = ((cond_sup_all[num].detach() - cond_all[num]) ** 2 * weight_all).sum() / (256 * h1 * w1 * num_all)
|
| loss_fea_sup_all = loss_fea_sup_all + (5 - num) * 0.04 * cond_sup_loss
|
| loss_warp = loss_warp + (num + 1) * loss_l1 + (num + 1) * 0.2 * loss_vgg + (num + 1) * 2 * loss_edge + (num + 1) * 6 * loss_second_smooth + (5 - num) * 0.04 * cond_sup_loss
|
| if num >= 2:
|
| b1, c1, h1, w1 = flow_all[num].shape
|
| weight_all = weight.reshape(-1, 1, 1).repeat(1, h1, w1)
|
| flow_sup_loss = (torch.norm(flow_sup_all[num].detach() - flow_all[num], p=2, dim=1) * weight_all).sum() / (h1 * w1 * num_all)
|
| loss_flow_sup_all = loss_flow_sup_all + (num + 1) * 1 * flow_sup_loss
|
| loss_warp = loss_warp + (num + 1) * 1 * flow_sup_loss
|
|
|
| loss_warp = 0.01 * loss_smooth + loss_warp
|
|
|
| if opt.local_rank == 0:
|
| writer.add_scalar('loss_warp', loss_warp, step)
|
| writer.add_scalar('loss_fea_sup_all', loss_fea_sup_all, step)
|
| writer.add_scalar('loss_flow_sup_all', loss_flow_sup_all, step)
|
|
|
| skin_mask = warped_prod_edge_un.detach() * (1 - person_clothes_edge.cuda())
|
| gen_inputs = torch.cat([p_tryon_un.detach(), warped_cloth, warped_prod_edge], 1)
|
| gen_outputs = PF_gen_model(gen_inputs)
|
| p_rendered, m_composite = torch.split(gen_outputs, [3, 1], 1)
|
| p_rendered = torch.tanh(p_rendered)
|
| m_composite = torch.sigmoid(m_composite)
|
| m_composite1 = m_composite * warped_prod_edge
|
| m_composite = person_clothes_edge.cuda() * m_composite1
|
| p_tryon = warped_cloth * m_composite + p_rendered * (1 - m_composite)
|
|
|
| loss_mask_l1 = torch.mean(torch.abs(1 - m_composite))
|
| loss_l1_skin = criterionL1(p_rendered * skin_mask, skin_mask * real_image.cuda())
|
| loss_vgg_skin = criterionVGG(p_rendered * skin_mask, skin_mask * real_image.cuda())
|
| loss_l1 = criterionL1(p_tryon, real_image.cuda())
|
| loss_vgg = criterionVGG(p_tryon, real_image.cuda())
|
| bg_loss_l1 = criterionL1(p_rendered, real_image.cuda())
|
| bg_loss_vgg = criterionVGG(p_rendered, real_image.cuda())
|
|
|
| if epoch < opt.niter:
|
| loss_gen = (loss_l1 * 5 + loss_l1_skin * 30 + loss_vgg + loss_vgg_skin * 2 + bg_loss_l1 * 5 + bg_loss_vgg + 1 * loss_mask_l1)
|
| else:
|
| loss_gen = (loss_l1 * 5 + loss_l1_skin * 60 + loss_vgg + loss_vgg_skin * 4 + bg_loss_l1 * 5 + bg_loss_vgg + 1 * loss_mask_l1)
|
|
|
| loss_all = 0.25 * loss_warp + loss_gen
|
|
|
| if opt.local_rank == 0:
|
| writer.add_scalar('loss_gen', loss_gen, step)
|
|
|
| optimizer_warp.zero_grad()
|
| optimizer_gen.zero_grad()
|
| loss_all.backward()
|
| optimizer_warp.step()
|
| optimizer_gen.step()
|
|
|
|
|
| path = 'sample/' + opt.name
|
| os.makedirs(path, exist_ok=True)
|
|
|
| if step % 1000 == 0:
|
| if opt.local_rank == 0:
|
| a = real_image.float().cuda()
|
| b = p_tryon_un.detach()
|
| c = clothes.cuda()
|
| d = person_clothes.cuda()
|
| e = torch.cat([skin_mask.cuda(), skin_mask.cuda(), skin_mask.cuda()], 1)
|
| f = warped_cloth
|
| g = p_rendered
|
| h = torch.cat([m_composite1, m_composite1, m_composite1], 1)
|
| i = p_tryon
|
| combine = torch.cat([a[0], b[0], c[0], d[0], e[0], f[0], g[0], h[0], i[0]], 2).squeeze()
|
| cv_img = (combine.permute(1, 2, 0).detach().cpu().numpy() + 1) / 2
|
| writer.add_image('combine', (combine.data + 1) / 2.0, step)
|
| rgb = (cv_img * 255).astype(np.uint8)
|
| bgr = cv2.cvtColor(rgb, cv2.COLOR_RGB2BGR)
|
| cv2.imwrite('sample/' + opt.name + '/' + str(step) + '.jpg', bgr)
|
|
|
| step += 1
|
| iter_end_time = time.time()
|
| iter_delta_time = iter_end_time - iter_start_time
|
| step_delta = (step_per_batch - step % step_per_batch) + step_per_batch * (opt.niter + opt.niter_decay - epoch)
|
| eta = iter_delta_time * step_delta
|
| eta = str(datetime.timedelta(seconds=int(eta)))
|
| time_stamp = datetime.datetime.now()
|
| now = time_stamp.strftime('%Y.%m.%d-%H:%M:%S')
|
|
|
| if step % 100 == 0:
|
| if opt.local_rank == 0:
|
| print('{}:{}:[step-{}]--[loss-{:.6f}]--[loss-{:.6f}]--[ETA-{}]'.format(now, epoch_iter, step, loss_gen, loss_warp, eta))
|
|
|
| if epoch_iter >= dataset_size:
|
| break
|
|
|
|
|
| iter_end_time = time.time()
|
| if opt.local_rank == 0:
|
| print('End of epoch %d / %d \t Time Taken: %d sec' %
|
| (epoch, opt.niter + opt.niter_decay, time.time() - epoch_start_time))
|
|
|
| if epoch % opt.save_epoch_freq == 0:
|
| if opt.local_rank == 0:
|
| print('saving the model at the end of epoch %d, iters %d' % (epoch, total_steps))
|
| save_checkpoint(PF_warp_model.module,
|
| os.path.join(opt.checkpoints_dir, opt.name, 'PFAFN_warp_epoch_%03d.pth' % (epoch + 1)))
|
| save_checkpoint(PF_gen_model.module,
|
| os.path.join(opt.checkpoints_dir, opt.name, 'PFAFN_gen_epoch_%03d.pth' % (epoch + 1)))
|
|
|
| if epoch > opt.niter:
|
| PF_warp_model.module.update_learning_rate_warp(optimizer_warp)
|
| PF_warp_model.module.update_learning_rate(optimizer_gen)
|
|
|
