| import numpy as np | |
| from skimage.metrics import structural_similarity as ssim | |
| def compute_ssim(rendering_gt, rendering_reconstructed): | |
| # The images passed to the function should already be normalized | |
| ssim_value = ssim(rendering_gt, rendering_reconstructed, data_range=rendering_gt.max() - rendering_gt.min(), multichannel=True, channel_axis=0) | |
| return ssim_value | |
| def compute_psnr(gt_image, rendered_image): | |
| # The images passed to the function should already be normalized | |
| gt_image = gt_image.astype(np.float32) | |
| rendered_image = rendered_image.astype(np.float32) | |
| mse = np.mean((gt_image - rendered_image) ** 2) | |
| psnr = 20 * np.log10(1.0 / np.sqrt(mse)) | |
| return psnr |