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

Confusion about the dataset setting of the paper #17

Open lzstudent opened 10 months ago

lzstudent commented 10 months ago

Thanks for the cool work. However i am a little bit confused about the experiment setting. In fig 1, it says given a set of monocular multi-view images and camera poses. In section 3.2, it says "this can be practical for an MVS capture at a certain moment but not for a monocular capture in a temporal sequence". It means your method is not a Multi-view capture. My question is what's the real setting of your dataset and experiment? I am not really familiar with dynamic scene reconstruction so my question might be stupid. Thanks again.

ingra14m commented 10 months ago

Hi, thanks for your interest in our work.

I have updated a new version on arxiv. What we want to express is that the method proposed by Dynamic Gaussian can only be used to represent multi-view scenes, while our method can be applied to monocular scenes. Therefore, in the paper, we expanded our method from the application of monocular scenes. Our dataset in the experiment is definitely monocular (except for the ablation figure, which is a multi-view scene from Neu3D).