66

import os
import torch
import torchvision.transforms.functional as TF
from skimage.metrics import peak_signal_noise_ratio as psnr
from skimage.metrics import structural_similarity as ssim
import lpips
from PIL import Image

def calculate_metrics(folder1, folder2):
    psnr_values = []
    ssim_values = []
    lpips_loss = lpips.LPIPS(net='vgg').cuda()
    lpips_loss.eval()
    lpips_values = []

    for file1, file2 in zip(sorted(os.listdir(folder1)), sorted(os.listdir(folder2))):
        image1 = Image.open(os.path.join(folder1, file1))
        image2 = Image.open(os.path.join(folder2, file2))

        # Convert images to tensors
        image1 = TF.to_tensor(image1).unsqueeze(0).cuda()
        image2 = TF.to_tensor(image2).unsqueeze(0).cuda()

        # PSNR
        psnr_val = psnr(image1.squeeze().cpu().numpy(), image2.squeeze().cpu().numpy())
        psnr_values.append(psnr_val)

        # SSIM
        ssim_val = ssim(image1.squeeze().cpu().numpy(), image2.squeeze().cpu().numpy(), multichannel=True, win_size=11)
        ssim_values.append(ssim_val)

        # LPIPS
        lpips_val = lpips_loss(image1, image2).item()
        lpips_values.append(lpips_val)

    avg_psnr = sum(psnr_values) / len(psnr_values)
    avg_ssim = sum(ssim_values) / len(ssim_values)
    avg_lpips = sum(lpips_values) / len(lpips_values)

    return avg_psnr, avg_ssim, avg_lpips

folder1 = 'folder1_path'
folder2 = 'folder2_path'

psnr_val, ssim_val, lpips_val = calculate_metrics(folder1, folder2)
print(f'Average PSNR: {psnr_val}')
print(f'Average SSIM: {ssim_val}')
print(f'Average LPIPS: {lpips_val}')