SeaThru-NeRF

SeaThru-NeRF dataset contains four underwater forward-facing scenes.

Web: https://sea-thru-nerf.github.io/
Paper: SeaThru-NeRF: Neural Radiance Fields in Scattering Media
Authors: Deborah Levy, Amit Peleg, Naama Pearl, Dan Rosenbaum, Derya Akkaynak, Tali Treibitz, Simon Korman
Method PSNR SSIM LPIPS Time GPU mem.
Gaussian Splatting 21.31 0.729 0.300 20m 9s 6.49 GB
Curasao 24.15 0.738 0.318 22m 26s 9.30 GB
Panama 23.68 0.749 0.250 22m 41s 8.43 GB
IUI3 18.86 0.677 0.390 18m 36s 4.26 GB
Japanese Gradens 18.54 0.752 0.242 16m 55s 3.96 GB
Mip-Splatting 21.33 0.730 0.304 20m 53s 6.19 GB
Curasao 24.41 0.739 0.316 23m 38s 9.54 GB
Panama 23.99 0.752 0.237 22m 42s 6.85 GB
IUI3 19.18 0.683 0.412 19m 32s 4.26 GB
Japanese Gradens 17.75 0.747 0.253 17m 43s 4.10 GB
SeaThru-NeRF 26.37 0.815 0.245 2h 49m 9s 127.55 GB
Curasao 30.00
Paper's PSNR: 30.48
0.870
Paper's SSIM: 0.87
0.215
Paper's LPIPS: 0.2
2h 48m 47s 127.66 GB
Panama 27.82
Paper's PSNR: 27.89
0.834
Paper's SSIM: 0.83
0.226
Paper's LPIPS: 0.22
2h 54m 57s 127.66 GB
IUI3 25.92 0.787 0.294 2h 46m 50s 127.44 GB
Japanese Gradens 21.73
Paper's PSNR: 21.83
0.768
Paper's SSIM: 0.77
0.246
Paper's LPIPS: 0.25
2h 46m 1s 127.44 GB

PSNR

Peak Signal to Noise Ratio. The higher the better.

Method Curasao Panama IUI3 Japanese Gradens
Gaussian Splatting 24.15 23.68 18.86 18.54
Mip-Splatting 24.41 23.99 19.18 17.75
SeaThru-NeRF 30.00
Paper's PSNR: 30.48
27.82
Paper's PSNR: 27.89
25.92 21.73
Paper's PSNR: 21.83

SSIM

Structural Similarity Index. The higher the better. The implementation matches JAX's SSIM and torchmetrics's SSIM (with default parameters).

Method Curasao Panama IUI3 Japanese Gradens
Gaussian Splatting 0.738 0.749 0.677 0.752
Mip-Splatting 0.739 0.752 0.683 0.747
SeaThru-NeRF 0.870
Paper's SSIM: 0.87
0.834
Paper's SSIM: 0.83
0.787 0.768
Paper's SSIM: 0.77

LPIPS

Learned Perceptual Image Patch Similarity. The lower the better. The implementation uses AlexNet backbone and matches lpips pip package with checkpoint version 0.1

Method Curasao Panama IUI3 Japanese Gradens
Gaussian Splatting 0.318 0.250 0.390 0.242
Mip-Splatting 0.316 0.237 0.412 0.253
SeaThru-NeRF 0.215
Paper's LPIPS: 0.2
0.226
Paper's LPIPS: 0.22
0.294 0.246
Paper's LPIPS: 0.25