Skip to main content

A data generation pipeline for creating semi-realistic synthetic multi-object videos with rich annotations such as instance segmentation, depth maps, and optical flow.

Project description

Kubric

Blender Kubruntu Test Coverage Docs

A data generation pipeline for creating semi-realistic synthetic multi-object videos with rich annotations such as instance segmentation masks, depth maps, and optical flow.

:warning: This project is pre-alpha work in progress and subject to extensive change.

Motivation and design

We need better data for training and evaluating machine learning systems, especially in the context of unsupervised multi-object video understanding. Current systems succeed on toy datasets, but fail on real-world data. Progress could be greatly accelerated if we had the ability to create suitable datasets of varying complexity on demand. Kubric is mainly built on-top of pybullet (for physics simulation) and Blender (for rendering); however, the code is kept modular to potentially support different rendering backends.

Getting started

For instructions, please refer to https://kubric.readthedocs.io

Assuming you have docker installed, to generate the data above simply execute:

git clone https://github.com/google-research/kubric.git
cd kubric
docker pull kubricdockerhub/kubruntu
docker run --rm --interactive \
           --user $(id -u):$(id -g) \
           --volume "$(pwd):/kubric" \
           kubricdockerhub/kubruntu \
           /usr/bin/python3 examples/helloworld.py
ls output

Requirements

  • A pipeline for conveniently generating video data.
  • Physics simulation for automatically generating physical interactions between multiple objects.
  • Good control over the complexity of the generated data, so that we can evaluate individual aspects such as variability of objects and textures.
  • Realism: Ideally, the ability to span the entire complexity range from CLEVR all the way to real-world video such as YouTube8. This is clearly not feasible, but we would like to get as close as possible.
  • Access to rich ground truth information about the objects in a scene for the purpose of evaluation (eg. object segmentations and properties)
  • Control the train/test split to evaluate compositionality and systematic generalization (for example on held-out combinations of features or objects)

Contributors

Klaus Greff (Google), Andrea Tagliasacchi (Google and University of Toronto), Derek Liu (University of Toronto), Issam Laradji (McGill and MILA)

Disclaimer

This is not an official Google Product

Project details


Release history Release notifications | RSS feed

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distributions

No source distribution files available for this release.See tutorial on generating distribution archives.

Built Distribution

kubric_nightly-2021.11.23-py3-none-any.whl (86.1 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page