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
779 stars 40 forks source link

About the camera model #43

Open BerenChou opened 3 months ago

BerenChou commented 3 months ago

Hi! Thank you for sharing your work! However, I have one question about the colmap camera model. In the readme of Gaussian-Splatting, they say the camera model must be a SIMPLE_PINHOLE or PINHOLE camera. But the images in NeRF-DS dataset have tangential_distortion and radial_distortion, which is not a SIMPLE_PINHOLE or PINHOLE camera. The authors of Gaussian-Splatting, they provide a convert.py to tackle this problem. But it seems that you directly use data (poses, points.npy, etc) from NeRF-DS without using convert.py. Will there be a problem?

Thank you in advance for your help!

ingra14m commented 3 months ago

Hi, Thanks for pointing that out. I indeed hadn't considered this aspect. I adopted the reading method consistent with HyperNeRF and Nerfies. After trying using convert.py, I found that there is no fundamental difference between these two methods in the results on NeRF-DS.

But I think it is meaningful. Welcome to pull requests!