# Walter0807/MotionBERT

[ICCV 2023] PyTorch Implementation of "MotionBERT: A Unified Perspective on Learning Human Motion Representations"

Repository: https://github.com/Walter0807/MotionBERT
Canonical: https://ross.abutalabs.com/products/motionbert
Language: Python
License: Apache-2.0
License Family: permissive
Topics: 3d-pose-estimation, mesh-recovery, skeleton-based-action-recognition, iccv2023
Last push: 2026-03-14T07:57:31+00:00

## Health v2 (maintenance only)
Score: 65/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 72, release rhythm 35, longevity 100
- inputs: {"age_days": 1584, "days_push": 172, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1439, forks 182 (observed 2026-08-28T04:04:44.034065+00:00)

## What it is
Official PyTorch implementation of MotionBERT (ICCV 2023), a unified pretrained model for learning human motion representations from 2D skeletons. It supports 3D human pose estimation, skeleton-based action recognition, and mesh recovery, including in-the-wild inference on custom videos.

## Use cases
- estimate 3d human pose from 2d keypoints
- recover 3d human mesh from monocular video
- recognize human actions from skeleton sequences
- extract human-centric video representations from motion data
- run 3d pose estimation on my own custom videos
- pretrain a unified human motion representation model

## When to choose
- You need a single pretrained model that handles 3D pose, mesh recovery, and skeleton-based action recognition
- You want a research baseline or reproducible implementation from a published ICCV 2023 paper with checkpoints
- You want to run in-the-wild inference on your own videos with human-centric motion features

## When to avoid
- You need a lightweight real-time pose estimator for production or edge deployment with strict latency budgets
- You work outside PyTorch, e.g. need TensorFlow, ONNX, or mobile-native inference pipelines
- You only need 2D pose detection from raw images, since this focuses on lifting 2D skeletons to 3D

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, deep-learning, computer-vision, transformers, video-processing
- domain: computer-vision, deep-learning, artificial-intelligence, machine-learning
- platform: python
- tags: pytorch, 3d-pose-estimation, human-motion, mesh-recovery, action-recognition, skeleton-sequence, pretrained-models, research-code, iccv-2023, motion-representation, video, gpu

## Member repositories
- Walter0807/MotionBERT (main) score 65

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:44.034065+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 2026-08-30T04:36:35.506213+00:00, confidence not recorded.
  - readme: https://github.com/Walter0807/MotionBERT (fetched 2026-08-28T04:04:44.034065+00:00, sha 04174003e475)
- Data as of 2026-08-30T08:39:29.467469+00:00.