PSNR problem #228
Description
Why is the 2DGS model I trained negatively optimized
`Number of points at initialisation : 60425 [13/06 08:38:51]
Training progress: 23%|██████████████████████████▊ | 7000/30000 [08:43<30:24, 12.60it/s, Loss=0.01249, distort=0.00000, normal=0.00000, Points=146954]
[ITER 7000] Evaluating test: L1 0.009224211273249239 PSNR 32.3086462020874 [13/06 08:47:34]
[ITER 7000] Evaluating train: L1 0.008699828386306762 PSNR 33.456260681152344 [13/06 08:47:34]
[ITER 7000] Saving Gaussians [13/06 08:47:34]
Training progress: 100%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 30000/30000 [34:54<00:00, 14.33it/s, Loss=0.01252, distort=0.00000, normal=0.00385, Points=155056]
[ITER 30000] Evaluating test: L1 0.010812767141032964 PSNR 28.39763081073761 [13/06 09:13:45]
[ITER 30000] Evaluating train: L1 0.009406386595219375 PSNR 29.69469680786133 [13/06 09:13:45]
[ITER 30000] Saving Gaussians [13/06 09:13:45]`
This result is much worse than the original 3DGS