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 can I use my own data to train and test? #6

Closed gwxxx closed 1 month ago

henrypearce4D commented 1 month ago

Hi great work, I would love to test on our own datasets.

Could you help with instructions for running training on colmap model files datasets?

\images \sparse

TQTQliu commented 1 month ago

Hi, thanks for your interest in our work. When using your own data, the important thing is to obtain the correct format for both intrinsic and extrinsic camera parameters. We've added a simple demo, hoping that it will be helpful to you. Additionally, you can refer to our dataset files(dtu, llff, NeRF Synthetic, and Tanks and Temples).

chenqi13814529300 commented 1 month ago

Can you provide a detailed explanation? When reproducing other papers, I used the same solution: Python train.py - s xxxx - m xxxx to obtain the results

TQTQliu commented 1 month ago

@chenqi13814529300 Sorry for the late reply, we will review the README and add as many details as possible. We also welcome your specific suggestions or any questions or problems you encounter when using our code.

chenqi13814529300 commented 1 month ago

We hope that your code has strong extensibility, as most of us only have data such as/parse and/images, which come from Colmap.

TQTQliu commented 1 month ago

@chenqi13814529300 Thanks for your suggestion, we will provide the code for the colmap data format later.

TQTQliu commented 1 month ago

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