model_cond.eval() print('Complete !') optimizer_cond = optim.Adam(model_cond.parameters(), lr=args.lr) model_D = MultiscaleDiscriminator(input_nc=3).to(device) model_D = nn.DataParallel(model_D).cuda() print('Loading Model_D...', end='') model_D.load_state_dict(torch.load('/p300/mem/mem_src/SPADE/checkpoint/as_101/vqvae_D_072.pt')) model_D.eval() print('Complete !') optimizer_D = optim.Adam(model_D.parameters(), lr=args.lr) # if args.sched == 'cycle': # scheduler = CycleScheduler( # optimizer, args.lr, n_iter=len(loader) * args.epoch, momentum=None # ) dic_model = {'model_img': model, 'model_cond': model_cond, # 'model_transfer': model_transfer, 'model_D': model_D, 'optimizer_img': optimizer, 'optimizer_cond': optimizer_cond, # 'optimizer_transfer': optimizer_transfer 'optimizer_D': optimizer_D } for i in range(args.epoch): viz.text(f'{DESCRIPTION} ##### Epoch: {i} #####', win='board') train(i, loader_train, dic_model, scheduler, device) val(i, loader_val, dic_model, scheduler, device) torch.save(model.state_dict(), f'checkpoint/{EXPERIMENT_CODE}/vqvae_{str(i + 1).zfill(3)}.pt') torch.save(model_D.state_dict(), f'checkpoint/{EXPERIMENT_CODE}/vqvae_D_{str(i + 1).zfill(3)}.pt')
torch.load('/p300/mem/mem_src/checkpoint/pose_04/vqvae_462.pt')) model_cond.eval() print('Complete !') optimizer_cond = optim.Adam(model_cond.parameters(), lr=args.lr) model_D = MultiscaleDiscriminator(input_nc=3).to(device) model_D = nn.DataParallel(model_D).cuda() optimizer_D = optim.Adam(model_D.parameters(), lr=args.lr) # if args.sched == 'cycle': # scheduler = CycleScheduler( # optimizer, args.lr, n_iter=len(loader) * args.epoch, momentum=None # ) dic_model = { 'model_img': model, 'model_cond': model_cond, # 'model_transfer': model_transfer, 'model_D': model_D, 'optimizer_img': optimizer, 'optimizer_cond': optimizer_cond, # 'optimizer_transfer': optimizer_transfer 'optimizer_D': optimizer_D } for i in range(args.epoch): viz.text(f'{DESCRIPTION} ##### Epoch: {i} #####', win='board') train(i, loader, dic_model, scheduler, device) torch.save( model.state_dict(), f'checkpoint/{EXPERIMENT_CODE}/vqvae_{str(i + 1).zfill(3)}.pt')