When I use a self-captured video as colmap dataset format, I have a preferred training visualization result, but the psnr is only 0.28 and ssim is only 0.05. I double checked the calculation of psnr and ssim, it didn't have any questions.
In addition, when I add "--gt_alpha_mask_as_dynamic_mask --gs_with_motion_mask", I receive a good psnr and ssim, but the rendering visualization results are very poor.
When I use a self-captured video as colmap dataset format, I have a preferred training visualization result, but the psnr is only 0.28 and ssim is only 0.05. I double checked the calculation of psnr and ssim, it didn't have any questions.