The Omnidata annotator is a tool to parametrically resample 3D scans and output a vision dataset. It works through a simple command-line interface (CLI) and to simplify installation we provide a public Dockerized implementation. The source code is available on github.
The demo annotator folder shows how to generate a dataset from a single building (in the Habitat-Matterport 3D dataset).
Head over to the annotator folder to try it out!
Once you've generated some data, you'll probably want to distribute it and/or use it for training your own models. We've released some generic PyTorch dataloaders that should make it easy to use the generated data for training.
We also provide some CLI utilities via the
omnitools command, to help with efficiently moving and releasing data. In particular,
omnitools.upload is for compressing & uploading the data while
omnitools.download does downloadling, validation, and decompression. More info on the omnitools CLI can be found here