# google-research/kubric

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.

Repository: https://github.com/google-research/kubric
Canonical: https://ross.abutalabs.com/products/kubric
Language: Jupyter Notebook
License: Apache-2.0
License Family: permissive
Last push: 2026-05-21T14:58:48+00:00

## Health v2 (maintenance only)
Score: 60/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 83, release rhythm 8, longevity 100
- inputs: {"age_days": 2233, "days_push": 104, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2808, forks 281 (observed 2026-08-28T04:07:22.918231+00:00)

## What it is
Kubric is a data generation pipeline from Google Research for creating semi-realistic synthetic multi-object videos with rich annotations like instance segmentation masks, depth maps, and optical flow. It is built on top of pybullet for physics simulation and Blender for rendering, with a modular design that can support other rendering backends.

## Use cases
- generate synthetic video datasets for training machine learning models
- create annotated videos with instance segmentation masks and depth maps
- benchmark multi-object video understanding systems
- generate physics-based object interaction scenes for research
- produce controlled-complexity datasets for evaluating vision models
- create synthetic training data for optical flow estimation

## When to choose
- you need labeled synthetic video data for training or evaluating computer vision models
- you want controllable dataset complexity from toy scenes toward realistic video
- you need automatic ground-truth annotations like segmentation, depth, and flow that are hard to obtain from real footage

## When to avoid
- you need fully photorealistic real-world video data
- you want a simple one-off renderer rather than a dataset generation pipeline
- you cannot run Docker or Blender-based rendering infrastructure

## Facets
- artifact type: library
- maturity: active
- function: data-generation, simulation, image-processing, machine-learning
- domain: machine-learning, computer-vision, data-science, simulation
- platform: python
- tags: synthetic-data, blender, pybullet, video-datasets, segmentation-masks, optical-flow, depth-maps, scene-rendering, docker, linux, gpu

## Member repositories
- google-research/kubric (main) score 60

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:22.918231+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T08:14:26.267349+00:00, confidence not recorded.
  - readme: https://github.com/google-research/kubric (fetched 2026-08-28T04:07:22.918231+00:00, sha 940f02319c4f)
  - registry_pypi: https://pypi.org/pypi/kubric/json (fetched 2026-08-29T09:54:16.513100+00:00, sha 4d8b1b3d893a)
- Data as of 2026-08-30T08:39:29.467469+00:00.