Taming 3DGS

Taming 3DGS improves the densification process to make the primitive count deterministic and implements several low-level optimizations for fast convergence

Web: https://humansensinglab.github.io/taming-3dgs/
Paper: Taming 3DGS: High-Quality Radiance Fields with Limited Resources
Authors: Saswat Subhajyoti Mallick, Rahul Goel, Bernhard Kerbl, Francisco Vicente Carrasco, Markus Steinberger, Fernando De La Torre

Mip-NeRF 360

Mip-NeRF 360 is a collection of four indoor and five outdoor object-centric scenes. The camera trajectory is an orbit around the object with fixed elevation and radius. The test set takes each n-th frame of the trajectory as test views.

Scene PSNR SSIM LPIPS (VGG) Time GPU mem.
garden 27.42
Paper's PSNR: 27.45

Experiments use the default 'ours' configuration. There is also 'big' configuration which uses more resources and achieves better results.

 0.858 0.145 8m 20s 8.10 GB
bicycle 24.75
Paper's PSNR: 24.97

Experiments use the default 'ours' configuration. There is also 'big' configuration which uses more resources and achieves better results.

 0.712 0.336 5m 16s 9.15 GB
flowers 21.10
Paper's PSNR: 21.18

Experiments use the default 'ours' configuration. There is also 'big' configuration which uses more resources and achieves better results.

 0.551 0.438 4m 51s 6.18 GB
treehill 23.02
Paper's PSNR: 23.04

Experiments use the default 'ours' configuration. There is also 'big' configuration which uses more resources and achieves better results.

 0.623 0.443 5m 18s 6.58 GB
stump 25.95
Paper's PSNR: 26.04

Experiments use the default 'ours' configuration. There is also 'big' configuration which uses more resources and achieves better results.

 0.734 0.336 4m 12s 5.35 GB
kitchen 30.92
Paper's PSNR: 31.14

Experiments use the default 'ours' configuration. There is also 'big' configuration which uses more resources and achieves better results.

 0.921 0.172 7m 30s 10.91 GB
bonsai 31.86
Paper's PSNR: 31.8

Experiments use the default 'ours' configuration. There is also 'big' configuration which uses more resources and achieves better results.

 0.935 0.271 6m 7s 10.80 GB
counter 28.53
Paper's PSNR: 28.59

Experiments use the default 'ours' configuration. There is also 'big' configuration which uses more resources and achieves better results.

 0.897 0.283 6m 37s 9.88 GB
room 31.41
Paper's PSNR: 31.39

Experiments use the default 'ours' configuration. There is also 'big' configuration which uses more resources and achieves better results.

 0.906 0.321 5m 40s 11.28 GB
Average 27.22
Paper's PSNR: 27.29

Experiments use the default 'ours' configuration. There is also 'big' configuration which uses more resources and achieves better results.

 0.793 0.305 5m 59s 8.69 GB