Lumina: Real-Time Mobile Neural Rendering by Exploiting Computational Redundancy

Anonymous

Video Comparisons

The following shows a comparison of our method (Left), 3DGS [Kerbl 2023] (Middle), and the Ground Truth (Right). All tests were conducted with a Radiance Record Length of 5, and we also annotate the Cache Hit Rate of our method.


Our Cache Hit Rate: 48.62%


Our Cache Hit Rate: 45.90%


Our Cache Hit Rate: 46.40%


Our Cache Hit Rate: 65.30%


Image Comparisons

The comparisons between our method and 3DGS [Kerbl 2023] are shown below. We also annotate the Cache Hit Rate (CHR) of our method for each scene.

Ours
(CHR=48.62%)
3DGS
[Kerbl 2023]
Ours
(CHR=51.40%)
3DGS
[Kerbl 2023]
Ours
(CHR=65.30%)
3DGS
[Kerbl 2023]
Ours
(CHR=51.03%)
3DGS
[Kerbl 2023]
Ours
(CHR=45.90%)
3DGS
[Kerbl 2023]
Ours
(CHR=54.20%)
3DGS
[Kerbl 2023]
Ours
(CHR=46.40%)
3DGS
[Kerbl 2023]
Ours
(CHR=76.84%)
3DGS
[Kerbl 2023]