"/fix_after_image.png") break with torch.no_grad(): g_fake_img = model_G(fix_pre_image) #g_fake_img = model_G( fix_after_image ) save_image(tensor=g_fake_img[0], filename=os.path.join(args.results_dir, args.exper_name) + "/fake_image_epoches{}_batch0.png".format(epoch)) save_image(tensor=g_fake_img, filename=os.path.join(args.results_dir, args.exper_name) + "/fake_image_epoches{}_batchAll.png".format(epoch)) fake_images_historys.append(g_fake_img[0].transpose(0, 1).transpose( 1, 2).cpu().clone().numpy()) save_image_historys_gif( fake_images_historys, os.path.join(args.results_dir, args.exper_name) + "/fake_image_epoches{}.gif".format(epoch)) save_checkpoint( model_G, device, os.path.join(args.save_checkpoints_dir, args.exper_name, "G", 'G_final.pth'), iterations) save_checkpoint( model_D, device, os.path.join(args.save_checkpoints_dir, args.exper_name, "D", 'D_final.pth'), iterations) print("Finished Training Loop.")
#save_checkpoint( model_G, device, os.path.join(args.save_checkpoints_dir, args.exper_name, "G", 'step_%08d.pth' % (iterations + 1)), iterations ) save_checkpoint( model_G, device, os.path.join(args.save_checkpoints_dir, args.exper_name, "G", 'G_final.pth'), iterations ) #save_checkpoint( model_D, device, os.path.join(args.save_checkpoints_dir, args.exper_name, "D", 'step_%08d.pth' % (iterations + 1)), iterations ) save_checkpoint( model_D, device, os.path.join(args.save_checkpoints_dir, args.exper_name, "D", 'D_final.pth'), iterations ) print( "saved checkpoints" ) n_print -= 1 #==================================================== # 各 Epoch 終了後の処理 #==================================================== for y_label in range(args.n_classes): eye_tsr = torch.eye( args.n_classes ).to( device ) y_fake_label = torch.full( (args.batch_size,), y_label ).long().to( device ) y_fake_one_hot = eye_tsr[y_fake_label].view( -1, args.n_classes, 1, 1 ).to( device ) # 出力画像の生成&保存 model_G.eval() with torch.no_grad(): G_z = model_G( input_noize_fix_z, y_fake_one_hot ) save_image( tensor = G_z[0], filename = os.path.join(args.results_dir, args.exper_name) + "/fake_image_label{}_epoches{}_batch0.png".format( y_label, epoch ) ) save_image( tensor = G_z, filename = os.path.join(args.results_dir, args.exper_name) + "/fake_image_label{}_epoches{}_batchAll.png".format( y_label, epoch ) ) fake_images_historys.append(G_z[0].transpose(0,1).transpose(1,2).cpu().clone().numpy()) save_image_historys_gif( fake_images_historys, os.path.join(args.results_dir, args.exper_name) + "/fake_image_label{}_epoches{}.gif".format( y_label, epoch, iterations ) ) save_checkpoint( model_G, device, os.path.join(args.save_checkpoints_dir, args.exper_name, "G", 'G_final.pth'), iterations ) save_checkpoint( model_D, device, os.path.join(args.save_checkpoints_dir, args.exper_name, "D", 'D_final.pth'), iterations ) print("Finished Training Loop.")