cschell/boxrr-23
BOXRR-23: Berkeley Open Extended Reality Recording Dataset 2023 This is a copy of the official Berkeley Open Extended Reality Recording Dataset 2023 (BOXRR-23). Please visit the project website for more information. In users/ you find one tarball for each user (which you can untar with tar xvf <path/to/user.tar>), which includes all replays of that user. Each replay is stored in a dedicated file in the XROR format. Metadata The entire dataset is around 5 TB large… See the full description on the dataset page: https://huggingface.co/datasets/cschell/boxrr-23.
BOXRR-23: Berkeley Open Extended Reality Recording Dataset 2023
This is a copy of the official Berkeley Open Extended Reality Recording Dataset 2023 (BOXRR-23). Please visit the project website for more information.
In users/ you find one tarball for each user (which you can untar with tar xvf <path/to/user.tar>), which includes all replays of that user. Each replay is stored in a dedicated file in the XROR format.
Metadata
The entire dataset is around 5 TB large, which can be inconvenient and you might want to only download users you are interested in. For this we have collected the metadata from each recording and dumped it into a BSON file which you find in metadata/. You can import this as collection into mongodb to quickly select users and replays to compile a list of the specific files you want to download (the field user_id matches the filename in users/).
Notes
- there are BeatSaber and TiltBrush users, which is noted in the field
info.software.appin the metadata - BeatSaber uses Unity's coordinate system. For more details (and why these details matter) have a look at this
Tools
- you'll need XROR to open the individual replays
- for an example script that imports and converts the motion data, you might find the "XR Motion Dataset Conversion Scrips" useful, which include a conversion script for BOXRR-23 (you need to unpack each user first).
- also, there is the Motion Learning Toolbox for further preprocessing of motion data