search

TQTQliu / MVSGaussian

[ECCV 2024] MVSGaussian: Fast Generalizable Gaussian Splatting Reconstruction from Multi-View Stereo
https://mvsgaussian.github.io/
MIT License
416 stars 21 forks source link

Dataset location #16

Closed shaonian123daly closed 3 months ago

shaonian123daly commented 3 months ago

How to use pre-trained models to obtain point clouds of data sets such as TNT and LLFF? Where should these data sets be placed?

TQTQliu commented 3 months ago

Hello, thanks for your interest in our work. For LLFF dataset, you can refer to scripts/mvsgs/llff_ft.sh L2 data_dir represents the location of LLFF dataset. L3 dir_ply represents the save directory of point clouds. L7 python run.py --type evaluate --cfg_file configs/mvsgs/llff_eval.yaml save_ply True dir_ply $dir_ply can generate point clouds for all scenes in LLFF dataset. L9-14 is per-scene optimization by using the above point cloud as initialization.

If you just want to obtain the point cloud, you can simply run:

python run.py --type evaluate --cfg_file configs/mvsgs/llff_eval.yaml save_ply True dir_ply mvsgs_pointcloud test_dataset.data_root <path to LLFF>

dir_ply represents the save directory of point clouds, and test_dataset.data_root represents the location of LLFF dataset. You can also specify the data set directory in the config file.

It is similar for TNT or any other data set.