search

TQTQliu / MVSGaussian

[ECCV 2024] MVSGaussian: Fast Generalizable Gaussian Splatting Reconstruction from Multi-View Stereo
https://mvsgaussian.github.io/
MIT License
347 stars 18 forks source link

How to run on Custom data? #10

Closed aniket-gupta1 closed 1 month ago

aniket-gupta1 commented 1 month ago

Hey,

thanks for your amazing work. I was wondering how can I run your method on the output from Colmap?

Basically I have data in the following format: data -- images/ -- sparse -- 0 -- cameras.bin, images.bin, points3D.bin

TQTQliu commented 1 month ago

Hello, we have re-uploaded a friendlier version of the demo, which only requires multi-view images.

henrypearce4D commented 1 month ago

@TQTQliu Our custom dataset is already pre-prepared in the format that @aniket-gupta1 described above for use with the original gaussian splatting code.

Can the colmap model files we already have;

images/
sparse/0/cameras.bin, images.bin, points3D.bin

be used directly with your code, without having to run the colmap process again or is running colmap a requirement to get extra information that your code requires - such as for the depth ranges?

henrypearce4D commented 1 month ago

I ran imgs2poses.py on my custom data set and it did find the existing colmap model files and created a poses_bounds.npy.

I had to do some edits on the mvsgs.py because my camera names are different than the example.

TQTQliu commented 1 month ago

@henrypearce4D Hello, I understand your request. Re-runing colmap is not required, we added some explanations here.