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
884 stars 49 forks source link

How to train on custom datasets #60

Open RuiqingTang opened 4 months ago

ingra14m commented 3 months ago

Hi, sorry for the late reply.

You can refer to SC-GS for the pre-processing.

DingChunQ commented 2 months ago

Hi, sorry for the late reply.

You can refer to SC-GS for the pre-processing.

What's the difference between SC-GS and NeRF-Texture you mentioned in #40?
Which one is better for videos taken from a simulation platform (such as maniskill)?

ingra14m commented 2 months ago

Hi @wangyancongAC , thanks for your interest.

tbh, I'm not familiar with simulation platform. But as for the difference between SC-GS and NeRF-Texture, they are actually the same when considering the pre-processing of the dataset. They all use COLMAP to obtain the camera poses and other useful information such as intrinsics and extrinsics of camera.

So I believe that COLMAP can also be employed to get essential components for GS-splatting in the simulation platform. And I recommend to use SC-GS since it has been tested lately.