omnidata-tools 0.0.23

Tooling for Omnidata: A Scalable Pipeline for Making Multi-Task Mid-Level Vision Datasets from 3D Scans

Omnidata Docs

Quick links to docs: [ Pretrained Models ] [ Starter Dataset ] [ Annotator Demo ]

This site is intended to be a wiki/documentation site for everything that we open-sourced from the paper. There are three main folders: the annotator, utilities (dataloaders, download tools, pretrained models, etc), and a code dump of stuff from the paper that is just for reference.

(Check out the main site for an overview of 'steerable datastes' and the 3D → 2D rendering pipeline).

Download the code

If you want to see and edit the code, then you can clone the github and install with:

git clone https://github.com/EPFL-VILAB/omnidata-tools
cd omnidata-tools
pip install -e .    # this will install the python requirements (and also install the CLI)

This is probably the best option for you if you want to use the pretrained models, dataloaders, etc in other work.

Install just CLI tools (omnitools)

If you are only interested in using the CLI tools, you can install them with: pip install omnidata-tools. This might be preferable if you only want to quickly download the starter data, or if you just want a simple way to manipulate the vision datasets output by the annotator.

Note: The annotator can also be used with a docker-based CLI, but you don't need to use the annotator to use the starter dataset, pretrained models, or training code.

...were you looking for the research paper or project website?