Skip to content

About

No description, website, or topics provided.

Resources

Stars

7 stars

Watchers

4 watching

Forks

Repository files navigation

Bayesian Fields

Note: Do not run the Clio datasets with undistorted images, run them with the images at the original size

We include several bash scripts to make running the code automated and easier. Before launching any bash scripts for the first time, remember to run

chmod +x <bash_script>

Installation

Third-party code (the Semantic Gaussians rasterizers, simple-knn, segment-anything and FastSAM) is pulled in as git submodules under their own licenses. Clone with --recursive, or in an existing clone run:

git submodule update --init --recursive

To install required packages and set up repo run:

pip install -r requirements.txt

pip install -e .

Running on provided dataset

We will give an example of running on the Clio cubicle dataset.

Set all paths in config/config_cubicle.yaml and in scripts/run_cubicle.bash

Run ./scripts/run_cubicle.bash which will run the entire Bayesian Fields pipeline and evaluations given a trained Gaussian Splat

To visualize, after the above finishes running, run python view_viewer.py --config config/config_cubicle.yaml

Making a custom dataset

If you are not trying to create ground truth labels and instead just want to perform qualitative experiments, the setup should be quick since we do not need to put metric scale into the scene. Create a video of a scene with a cellphone camera. Download the video and convert it to a folder of images with:

mkdir -p <desired_location>/custom_dataset/images

ffmpeg -i /home/dominic/Downloads/IMG_3390.MOV -vf "fps=10" <desired_location>/custom_dataset/images/frame_%04d.jpg

Run COLMAP to get poses. Inside scripts/run_colmap.bash change DATASET_PATH. Then run:

./scripts/run_colmap.bash

Run Gaussian Splatting to get a reconstruction of the scene. Note we are not currently running the most recent version of Gaussian Splatting which is much faster. Change the fields in offical_train.yaml marked with CHANGE THIS FOR NEW DATASET then run:

python train.py

About

No description, website, or topics provided.

Resources

Stars

7 stars

Watchers

4 watching

Forks

Releases

Packages

Contributors

Languages