Video¶
The data product we generate from video is called a splat. This is a visually accurate 3D model useful for human interpretation. Future work is to update this to fuse LiDAR with video, which produces geometrically accurate 3D models that can then be a basis for further analytics.
Setup¶
Install dependencies (Ubuntu instructions)¶
Ensure submodules are checked out:
cd open-goodfire-tools
git submodule update --init
cd open-goodfire-tools/video
sudo ./setup.sh
Gaussian Splat Pipeline¶
Place your video named <dataset_name>.mp4
in the folder video/data/input
. Then run the following:
cd open-goodfire-tools/video
source ../.env/bin/activate
python gsplat.py --dataset <dataset_name> -vv --sfm odm