深度学习-训练网络后直接进行验证保存最优模型

     best_psnr, best_ssim, best_epoch_ssim, best_epoch_psnr = 0., 0., 0, 0
    if epoch % 1 == 0:
        model.eval()
        PSNRs, SSIMs = [], []
        pbar = tqdm(val_loader)
        for ii, data_val in enumerate(pbar, 0):
            input_ = data_val[0].cuda()
            target_SR = data_val[1].cuda()

            model.forward(input_, input_, training=False)
            restored_SR = model.sample(testing=True)

            with torch.no_grad():
                for res, tar in zip(restored_SR, target_SR):
                    print(res.shape, tar.shape)
                    temp_psnr = torchPSNR(res, tar)
                    temp_ssim = torchSSIM(restored_SR, target_SR)
                    PSNRs.append(temp_psnr)
                    SSIMs.append(temp_ssim)
                    pbar.set_description("[Epoch] {} [MODE] VALID [PSNR] {:.4f} [SSIM] {:.4f}".format(
                        epoch,
                        torchPSNR(restored_SR, target_SR),
                        torchSSIM(restored_SR, target_SR))
                    )

        PSNRs = torch.stack(PSNRs).mean().item()
        SSIMs = torch.stack(SSIMs).mean().item()

        save_image(torch.cat((pad_img(input_[0], (256, 256)),
                              restored_SR[0], target_SR[0]), -1),
                   os.path.join(opt.output, str(epoch) + '.png'))  # save image

        # Save the best PSNR model of validation
        if PSNRs > best_psnr:
            best_psnr = PSNRs
            best_epoch_psnr = epoch
            model_out_path = opt.save_folder + "epoch_best_psnr.pth"
            torch.save(model.state_dict(), model_out_path)
            print("Checkpoint saved to {}".format(model_out_path))
        print("[PSNR] {:.4f} [Best_PSNR] {:.4f} (epoch {})".format(PSNRs, best_psnr, best_epoch_psnr))

        # Save the best SSIM model of validation
        if SSIMs > best_ssim:
            best_ssim = SSIMs
            best_epoch_ssim = epoch
            model_out_path = opt.save_folder + "epoch_best_ssim.pth"
            torch.save(model.state_dict(), model_out_path)
            print("Checkpoint saved to {}".format(model_out_path))
        print("[SSIM] {:.4f}  [Best_SSIM] {:.4f} (epoch {})".format(SSIMs, best_ssim, best_epoch_ssim))