PSNR are not good.

[AutoSenseTech]

I noticed that when I run SplatAM on the Freiburg1_desk of the TUM dataset, the PSNR for the first frame drops as the Gaussians are added to the map, and the final average PSNR is not good. Is there any way to improve the PSNR, and would changing the mapping_window_size parameter in the config help?

[Nik-V9]

Hi, the PSNR performance on TUM can be improved by improving the depth quality or mapping iterations.

Running Gaussian Splatting with the TUM poses might give you a sense of the upper ceiling in terms of PSNR.