whatisgithub

What is jepa?

facebookresearch/jepa — explained in plain English

Analysis updated 2026-07-03 · repo last pushed 2025-02-27

★ 3,994PythonAudience · researcherComplexity · 4/5StaleLicenseSetup · hard

In one sentence

Meta's V-JEPA trains AI to understand video by predicting hidden parts of clips, producing reusable representations for action recognition and image tasks without any labeled data.

Mindmap

mindmap
  root((jepa))
    What it does
      Masked video prediction
      Feature learning
      No labels needed
    Outputs
      Video representations
      Action recognition
      Image classification
    How it works
      Hide video regions
      Predict features
      Freeze and probe
    Tech Stack
      Python
      PyTorch
    Audience
      AI researchers
      Video ML engineers
Click or tap to explore — scroll the page freely

Code map

Detail Auto

An interactive map of this repo's files and how they connect — its source is parsed live in your browser. Click Visualize to build it.

filefunction / class

What do people build with it?

USE CASE 1

Download a pretrained V-JEPA model and attach a small probe to classify actions in your own video dataset without labels.

USE CASE 2

Use V-JEPA features as a strong starting point for a video understanding task to avoid collecting expensive annotations.

USE CASE 3

Evaluate how well video-only self-supervised learning transfers to image recognition tasks like ImageNet.

USE CASE 4

Train a V-JEPA model on your own video dataset to build domain-specific video representations.

What is it built with?

PythonPyTorch

How does it compare?

facebookresearch/jepakarpathy/makemorefacebookresearch/videopose3d
Stars3,9944,0104,036
LanguagePythonPythonPython
Last pushed2025-02-272024-06-042022-12-10
MaintenanceStaleDormantDormant
Setup difficultyhardeasymoderate
Complexity4/52/53/5
Audienceresearcherresearcherdeveloper

Figures from each repo's GitHub metadata at analysis time.

How do you get it running?

Difficulty · hard Time to first run · 1h+

Requires GPU, training uses large video datasets and the VideoMix2M dataset is not publicly hosted.

Apache 2.0, free for any use including commercial, keep the license notice.

Copy-paste prompts

Prompt 1
Using facebookresearch/jepa, write Python code to load a pretrained V-JEPA model and extract video features for a local video clip.
Prompt 2
Show me how to attach a linear probe on top of a frozen V-JEPA model and fine-tune it for video action classification.
Prompt 3
How do I run the evaluation scripts in facebookresearch/jepa to benchmark a pretrained model on Kinetics-400?
Prompt 4
Explain how V-JEPA's masked feature prediction differs from masked pixel prediction and why it produces better representations.

Frequently asked questions

What is jepa?

Meta's V-JEPA trains AI to understand video by predicting hidden parts of clips, producing reusable representations for action recognition and image tasks without any labeled data.

What language is jepa written in?

Mainly Python. The stack also includes Python, PyTorch.

Is jepa actively maintained?

Stale — no commits in 1-2 years (last push 2025-02-27).

What license does jepa use?

Apache 2.0, free for any use including commercial, keep the license notice.

How hard is jepa to set up?

Setup difficulty is rated hard, with roughly 1h+ to a first successful run.

Who is jepa for?

Mainly researcher.

Open on GitHub → Ask about another repo

This repo across BitVibe Labs

Verify against the repo before relying on details.

What are similar repos?