awesome-repositories.com
Blog
MCP
awesome-repositories.com

Discover the best open-source repositories with AI-powered search.

ExploreCurated searchesOpen-source alternativesSelf-hosted softwareBlogSitemap
ProjectMCP serverAboutHow we rankPress
LegalPrivacyTerms
© 2026 Bringes Technology SRL·VAT RO45896025·hello@awesome-repositories.com
facebookresearch avatar

facebookresearch/vjepa2

0
View on GitHub↗

Vjepa2

vjepa2 is a joint-embedding predictive architecture and video self-supervised learning framework. It functions as a visual representation learner and a robotic manipulation model designed to learn representations by predicting future latent states without reconstructing pixels.

The system enables the pretraining of video encoders that learn temporally consistent features through masked-token prediction and multi-modal tokenization. It further maps these latent embeddings to specific physical movements via action-conditioned post-training to plan and execute robot arm grasping and picking tasks.

The framework covers visual motion anticipation and the evaluation of visual understanding using attentive feature probing on frozen latent representations.

AI search

Explore more awesome repositories

Describe what you need in plain English — the AI ranks thousands of curated open-source projects by relevance.

Start searching with AI

Features

  • Joint-Embedding Predictive Architectures - A neural network design that learns representations by predicting future latent states without reconstructing pixels.
  • Joint Embedding Spaces - Implements a joint-embedding architecture that maps different views of video data into a common coordinate system.
  • Multi-Modal Tokenizers - Converts raw visual video data into discrete numerical sequences for latent space processing.
  • Self-Supervised Vision Representation Trainers - Trains models to recognize patterns in dense visual data without the need for manually labeled datasets.
  • Robotic Manipulation Models - Provides a visual-action system for planning robot arm movements and executing grasping tasks based on image goals.
  • Temporally Consistent Embeddings - Ensures visual features remain stable and coherent across consecutive video frames through a shared latent space.
  • Latent Masked Token Prediction - Learns representations by predicting missing video segments within a hidden embedding space.
  • Video Encoder Pretraining - Provides a framework to pretrain video encoders that learn consistent visual features from natural videos.
  • Video Pre-training - Provides a framework for pretraining video encoders using self-supervised techniques to learn temporal representations.
  • Latent Action Prediction - Implements predictive systems that map visual states into latent spaces to generate robot control signals.
  • Discretized Visual Representations - Extracts dense, discretized visual features used for the downstream evaluation of motion and action anticipation.
  • Latent Attention Analysis - Provides tools to interrogate compressed latent spaces using attention layers to analyze motion and action anticipation.
  • Visual Motion Anticipation - Evaluates how well a model predicts future movements by analyzing frozen latent features of a video.
  • Grasping and Manipulation - Enables robot arm grasping and picking tasks by planning movements from image goals.
  • Video Understanding - Evaluates visual understanding of motion and action anticipation using attentive probes on latent features.
  • Embodied Foundation Models - Self-supervised video models for understanding and planning.
3,021 stars·341 forks·Python·mit·22 views

Star history

Star history chart for facebookresearch/vjepa2Star history chart for facebookresearch/vjepa2

How this analysis was created: This summary and feature list are AI-generated from collected project material and can contain mistakes. Stars, license and language are imported from GitHub. Inclusion does not mean that we have tested or audited this project. Check the source documentation for any feature you depend on. Learn more on our About page.

Frequently asked questions

What does facebookresearch/vjepa2 do?

vjepa2 is a joint-embedding predictive architecture and video self-supervised learning framework. It functions as a visual representation learner and a robotic manipulation model designed to learn representations by predicting future latent states without reconstructing pixels.

What are the main features of facebookresearch/vjepa2?

The main features of facebookresearch/vjepa2 are: Joint-Embedding Predictive Architectures, Joint Embedding Spaces, Multi-Modal Tokenizers, Self-Supervised Vision Representation Trainers, Robotic Manipulation Models, Temporally Consistent Embeddings, Latent Masked Token Prediction, Video Encoder Pretraining.

Which projects share features with facebookresearch/vjepa2?

Projects with overlapping indexed features include: facebookresearch/jepa — This is a PyTorch self-supervised learning framework designed to train models that learn visual representations from… deepseek-ai/janus — Janus is a multimodal large language model and unified framework that integrates visual understanding and image… facebookresearch/slowfast — SlowFast is a PyTorch video understanding framework and spatiotemporal neural network library. It serves as a toolset… open-gigaai/giga-brain-0 — giga-brain-0 is a robot action model framework designed to train and deploy neural networks that map multi-modal… facebookresearch/dino — This project is a PyTorch vision transformer framework designed for self-supervised learning. It implements a model… facebookresearch/dinov3 — This project is a self-supervised vision foundation model based on a vision transformer architecture. It is designed…

Projects sharing features with Vjepa2

These projects share indexed features with Vjepa2. Shared tags can include platform or build tooling; verify the primary use case before treating a result as a replacement.
  • facebookresearch/jepafacebookresearch avatar

    facebookresearch/jepa

    3,986View on GitHub↗

    This is a PyTorch self-supervised learning framework designed to train models that learn visual representations from video. It implements a joint-embedding predictive architecture that extracts spatio-temporal features by predicting missing regions of a signal within a latent representation space rather than reconstructing raw pixels. The project includes a latent space visualization tool that uses a conditional diffusion model to decode feature-space predictions back into pixels. This allows for the verification of learned representations by transforming abstract predictions into interpretab

    Python
    View on GitHub↗3,986
  • deepseek-ai/janusdeepseek-ai avatar

    deepseek-ai/Janus

    17,746View on GitHub↗

    Janus is a multimodal large language model and unified framework that integrates visual understanding and image generation within a single neural network. It functions as both a visual understanding model for analyzing images and a text-to-image generator. The system uses a unified transformer backbone and a multimodal latent space to bridge the gap between text and visual data. This architecture employs decoupled visual encoding and cross-modal tokenization to separate the paths for discriminative understanding and generative tasks, representing images as grids of discrete codes. The projec

    Pythonany-to-anyfoundation-modelsllm
    View on GitHub↗17,746
  • facebookresearch/slowfastfacebookresearch avatar

    facebookresearch/SlowFast

    7,377View on GitHub↗

    SlowFast is a PyTorch video understanding framework and spatiotemporal neural network library. It serves as a toolset for video action recognition, enabling the training and evaluation of models designed to classify complex activities and objects within video sequences. The framework is distinguished by its use of dual-pathway spatiotemporal sampling to capture both slow and fast motions. It supports self-supervised video learning for pre-training models on unlabeled data and employs multigrid spatiotemporal training to optimize learning across multiple spatial and temporal resolutions. The

    Python
    View on GitHub↗7,377
  • open-gigaai/giga-brain-0open-gigaai avatar

    open-gigaai/giga-brain-0

    2,542View on GitHub↗

    giga-brain-0 is a robot action model framework designed to train and deploy neural networks that map multi-modal sensor data to physical robot control signals. It functions as a robot manipulation controller that processes high-dimensional observations to execute dexterous, long-horizon physical tasks. The project provides a multi-modal robot inference server using a client-server architecture to stream real-time vision and language observations for instant action prediction. It includes an embodiment fine-tuning pipeline to adapt pre-trained base models to specific robot hardware configurati

    Python
    View on GitHub↗2,542
Compare all 30 related projects→