search

ingra14m / Deformable-3D-Gaussians

[CVPR 2024] Official implementation of "Deformable 3D Gaussians for High-Fidelity Monocular Dynamic Scene Reconstruction"
https://ingra14m.github.io/Deformable-Gaussians/
MIT License
868 stars 49 forks source link

Regarding the reproduction results #74

Open Jiang-HB opened 1 week ago

Jiang-HB commented 1 week ago

Dear authors,

Thank you for open-sourcing your codes; this is a very interesting work. When I attempted to reproduce your results, I found that my performance on the D-NeRF dataset was not very good. For example, on "jumpingjacks," the PSNR is only 32.7251129. I believe one possible reason is that the number of points is quite low ("Loss=0.0074367, #Point=1738"). Could you please suggest any potential solutions?

Bests

ingra14m commented 1 week ago

I think that you may have modified the code. The number of point is too little.

You could refer to the Lightweight-Deformable-GS to compare your results with the normal results.