Open beibeirory opened 7 months ago
Hi,
We test our method on a single NVIDIA RTX3090 GPU.
The comparison of rendering speed on the DNeRF dataset with 800*800 resolution is as the following table:
Method | 3D-GS | 4D-GS | Baseline | Ours |
---|---|---|---|---|
FPS | 298.5 | 145.8 | 113.3 | 295.4 |
Baseline is our method without sparse control points and with per-Gaussian motion prediction. Notice that our method uses cached NN information and does not need KNN queries. The cost of MLP prediction for each sparse control point is very little.
Thanks for your reply. The work is helpful.
Thanks for your work. What’s the inference rendering speed? And GPU?