Meta Research ORG

The repository provides code for running inference with the SegmentAnything Model (SAM), links for downloading the trained model checkpoints, and example notebooks that show how to use the model.

A library for efficient similarity search and clustering of dense vectors.

Detectron2 is a platform for object detection, segmentation and other visual recognition tasks.

Facebook AI Research Sequence-to-Sequence Toolkit written in Python.

Library for fast text representation and classification.

FAIR's research platform for object detection research, implementing popular algorithms like Mask R-CNN and RetinaNet.

Audiocraft is a library for audio processing and generation with deep learning. It features the state-of-the-art EnCodec audio compressor / tokenizer, along with MusicGen, a simple and controllable music generation LM with textual and melodic conditioning.

The repository provides code for running inference with the Meta Segment Anything Model 2 (SAM 2), links for downloading the trained model checkpoints, and example notebooks that show how to use the model.

End-to-End Object Detection with Transformers

[CVPR 2025 Best Paper Award] VGGT: Visual Geometry Grounded Transformer

PyTorch code and models for the DINOv2 self-supervised learning method.

Code to accompany "A Method for Animating Children's Drawings of the Human Figure"

Foundational Models for State-of-the-Art Speech and Text Translation

Reference PyTorch implementation and models for DINOv3

The repository provides code for running inference and finetuning with the Meta Segment Anything Model 3 (SAM 3), links for downloading the trained model checkpoints, and example notebooks that show how to use the model.

Hackable and optimized Transformers building blocks, supporting a composable construction.

Hydra is a framework for elegantly configuring complex applications

Code for the paper Hybrid Spectrogram and Waveform Source Separation

Implementation of Nougat Neural Optical Understanding for Academic Documents

PyTorch3D is FAIR's library of reusable components for deep learning with 3D data

High-Resolution 3D Human Digitization from A Single Image.

ImageBind One Embedding Space to Bind Them All

Official PyTorch Implementation of "Scalable Diffusion Models with Transformers"

PyTorch implementation of MAE https//arxiv.org/abs/2111.06377

PyTorch code for Vision Transformers training with the Self-Supervised learning method DINO

SAM 3D Objects

Repo for external large-scale work

Code release for ConvNeXt model

Kats, a kit to analyze time series data, a lightweight, easy-to-use, generalizable, and extendable framework to perform time series analysis, from understanding the key statistics and characteristics, detecting change points and anomalies, to forecasting future trends.

A natural language modeling framework based on PyTorch

A modular framework for vision & language multimodal research from Facebook AI Research (FAIR)

High-resolution models for human tasks.

PyTorch implementation of MoCo: https://arxiv.org/abs/1911.05722

A data augmentations library for audio, image, text, and video.

CoTracker is a model for tracking any point (pixel) on a video.

A PyTorch library for implementing flow matching algorithms, featuring continuous and discrete flow matching implementations. It includes practical examples for both text and image modalities.

A Python toolbox for performing gradient-free optimization

PyTorch code and models for VJEPA2 self-supervised learning from video.

Evolutionary Scale Modeling (esm): Pretrained language models for proteins

Efficient 3D human pose estimation in video using 2D keypoint trajectories

PyTorch code and models for V-JEPA self-supervised learning from video.

State-of-the-art deep learning based audio codec supporting both mono 24 kHz audio and stereo 48 kHz audio.

A flexible, high-performance 3D simulator for Embodied AI research.

MapAnything: Universal Feed-Forward Metric 3D Reconstruction

Official codebase for I-JEPA, the Image-based Joint-Embedding Predictive Architecture. First outlined in the CVPR paper, "Self-supervised learning from images with a joint-embedding predictive architecture."

Code release for "Masked-attention Mask Transformer for Universal Image Segmentation"

The repository provides code for running inference with the SAM 3D Body Model (3DB), links for downloading the trained model checkpoints and datasets, and example notebooks that show how to use the model.

A library for Multilingual Unsupervised or Supervised word Embeddings

[CVPR 2026 Oral] VGGT Omega

A modular high-level library to train embodied AI agents across a variety of tasks and environments.

A Production-ready Reinforcement Learning AI Agent Library brought by the Applied Reinforcement Learning team at Meta.

This repository contains the code to train and evaluate TRIBE v2, a multimodal model for brain response prediction

Code and dataset for photorealistic Codec Avatars driven from audio

Omnilingual ASR Open-Source Multilingual SpeechRecognition for 1600+ Languages

HiPlot makes understanding high dimensional data easy

Self-referential self-improving agents that can optimize for any computable task

Schedule-Free Optimization in PyTorch

State-of-the-art Image & Video CLIP, Multimodal Large Language Models, and More!

Collection of common code that's shared among different research projects in FAIR computer vision team.

FAIR Chemistry's library of machine learning methods for chemistry

Submanifold sparse convolutional networks

A python tool for evaluating the quality of sentence embeddings.

Code release for ConvNeXt V2 model

Code for BLT research paper

Non-local Neural Networks for Video Classification

A Python framework for AI-driven character animation using neural networks.

The official pytorch implementation of our paper "Is Space-Time Attention All You Need for Video Understanding?"

Deep Hough Voting for 3D Object Detection in Point Clouds

Public release of the TransCoder research project https://arxiv.org/pdf/2006.03511.pdf

Training Large Language Model to Reason in a Continuous Latent Space

DomainBed is a suite to test domain generalization algorithms

Learning Continuous Signed Distance Functions for Shape Representation

Per-Pixel Classification is Not All You Need for Semantic Segmentation (NeurIPS 2021, spotlight)

Code for Cicero, an AI agent that plays the game of Diplomacy with open-domain natural language negotiation.

VGGSfM: Visual Geometry Grounded Deep Structure From Motion

The Replica Dataset v1 as published in https://arxiv.org/abs/1906.05797 .

A method to increase the speed and lower the memory footprint of existing vision transformers.

a Realistic and Rich 3D Environment

mixup: Beyond Empirical Risk Minimization

Fast Differentiable Tensor Library in JavaScript and TypeScript with Bun + Flashlight

Official implementation of the paper "Watermark Anything with Localized Messages"

PyTorch implementation of Barlow Twins.

[CVPR 2025] Official PyTorch implementation of "EdgeTAM: On-Device Track Anything Model"

Mobile vision models and code

1K resolution vision transformers pretrained on 1B human images.

projectaria_tools is an C++/Python open-source toolkit to interact with Project Aria data.

Code for the ShapeR research paper

Ocean is the in-house framework for Computer Vision (CV) and Augmented Reality (AR) applications at Meta. It is platform independent and is mainly implemented in C/C++.

Supervision-by-Registration: An Unsupervised Approach to Improve the Precision of Facial Landmark Detectors

The first behavioral foundation model to control a virtual physics-based humanoid agent for a wide range of whole-body tasks.

Can Language Models Rebuild Programs From Scratch?

[CVPR'25 Highlight] Official repository of Sonata: Self-Supervised Learning of Reliable Point Representations

The balance python package offers a simple workflow and methods for dealing with biased data samples when looking to infer from them to some target population of interest.

Localized watermarking for AI-generated speech audios, with SOTA on robustness and very fast detector

Unified automatic quality assessment for speech, music, and sound.

Momentum Human Rig is an anatomically-inspired parametric full-body digital human model developed at Meta. It includes: A parametric body skeletal model; A realistic 3D mesh skinned to the skeleton with levels of detail;A body blendshape and pose corrective model; A facial blendshape model.Its design is friendly for both CG and CV communities.

Official implementation of Tuna-2: Pixel Embeddings Beat Vision Encoders for Unified Understanding and Generation

An open source library designed to provide community examples of Joint Embedding Predictive Architectures (JEPAs). It contains code and examples for learning representations from images, video, and action-conditioned video, as well as planning using JEPA-based models.

An algorithm that generalizes the paradigm of self-play reinforcement learning and search to imperfect-information games.

Open and efficient video and image watermarking

This repo hosts the code and models of "Masked Autoencoders that Listen".

Efficient Universal Perception Encoder: a single on-device vision encoder with versatile representations that match or exceed specialized experts across multiple task domains.

Multiview Compressive Coding for 3D Reconstruction

Pippo: High-Resolution Multi-View Humans from a Single Image

Official code for the CVPR 2025 paper "Navigation World Models".

BenchMARL is a library for benchmarking Multi-Agent Reinforcement Learning (MARL). BenchMARL allows to quickly compare different MARL algorithms, tasks, and models while being systematically grounded in its two core tenets: reproducibility and standardization.

Ego4d dataset repository. Download the dataset, visualize, extract features & example usage of the dataset

MLGym A New Framework and Benchmark for Advancing AI Research Agents

For optimization algorithm research and development.

Code for the Boxer research paper

VICReg official code base

Source Code for Paper "OrienterNet Visual Localization in 2D Public Maps with Neural Matching"

Meta Agents Research Environments is a comprehensive platform designed to evaluate AI agents in dynamic, realistic scenarios. Unlike static benchmarks, this platform introduces evolving environments where agents must adapt their strategies as new information becomes available, mirroring real-world challenges.

MiniHack the Planet: A Sandbox for Open-Ended Reinforcement Learning Research

The repository for the largest and most comprehensive empirical study of visual foundation models for Embodied AI (EAI).

Training and evaluation pipeline for MEG and EEG brain signal encoding and decoding using deep learning. Code for our paper "Decoding speech perception from non-invasive brain recordings" published in Nature Machine Intelligence, 2023.

A general physic-based retargeting framework.

A Large-scale Video Action Dataset

A first-of-its-kind acoustic simulation platform for audio-visual embodied AI research. It supports training and evaluating multiple tasks and applications.

[NeurIPS 2025 (Spotlight)] The implementation for the paper "4DGT Learning a 4D Gaussian Transformer Using Real-World Monocular Videos"

PHYRE is a benchmark for physical reasoning.

Open source repo for Locate 3D Model, 3D-JEPA and Locate 3D Dataset

DocAgent is a system designed to generate high-quality, context-aware code documentation for Python codebases using a multi-agent approach and hierarchical processing.

VRS is a file format optimized to record & playback streams of sensor data, such as images, audio samples, and any other discrete sensors (IMU, temperature, etc), stored in per-device streams of timestamped records.

Large dataset of hand-object contact, hand- and object-pose, and 2.9 M RGB-D grasp images.

A Rust library for lattice-based additive homomorphic encryption.

Code, data and weights for the paper **What drives success in physical planning with Joint-Embedding Predictive World Models?**

Open source implementation of "A Self-Supervised Descriptor for Image Copy Detection" (SSCD).

Code repo for the paper "SpinQuant LLM quantization with learned rotations"

GPU implementation of a fast generalized ANS (asymmetric numeral system) entropy encoder and decoder, with extensions for lossless compression of numerical and other data types in HPC/ML applications.

Foundation Models and Data for Human-Human and Human-AI interactions.

Scalable and Performant Data Loading

Official code for "LagerNVS Latent Geometry for Fully Neural Real-time Novel View Synthesis" (CVPR 2026)

🎬ActionMesh: A fast video to animated mesh model with unprecedented quality. Generate animated mesh seamlessly importable into any 3D software in less than a minute.

ViewDiff generates high-quality, multi-view consistent images of a real-world 3D object in authentic surroundings. (CVPR2024).

A library for human kinematic motion and numerical optimization solvers to apply human motion

Official codebase for the paper "Beyond A* Better Planning with Transformers via Search Dynamics Bootstrapping".

Code for "LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding", ACL 2024

A framework to study AI models in Reasoning, Alignment, and use of Memory (RAM).

Goliath Dataset and Official PyTorch Implementation of RelightableHands, Relightable Gaussian Codec Avatars, and Driving-Signal Aware Full-Body Avatars.

Official implementation of paper "VLM³: Vision Language Models Are Native 3D Learners".

Comprehensive benchmark for RAG

RLMeta is a light-weight flexible framework for Distributed Reinforcement Learning Research.

Matrix (Multi-Agent daTa geneRation Infra and eXperimentation framework) is a versatile engine for multi-agent conversational data generation.

Repository for the CVPR 2026 paper MeshFlow Efficient Artistic Mesh Generation via MeshVAE and Flow-based Diffusion Transformer by Weiyu Li, Antoine Toisoul, Tom Monnier, Roman Shapovalov, Rakesh Ranjan, Ping Tan and Andrea Vedaldi.

Digit 360 is a modular platform that unlocks new capabilities, and enables future research on the nature of touch.

Python suite for neuroscience research across all modalities.

Mixture-of-Transformers: A Sparse and Scalable Architecture for Multi-Modal Foundation Models. TMLR 2025.

ATLAS Autoformalized Textbook Library At Scale

This repository contains a command-line interface(CLI) that can detect and blur out faces and license plates(PII) from images and videos. The CLI takes an image or video file as input, runs an anonymization algorithm on it, and writes the blurred output to a specified path.

Code for exploring surface electromyography (sEMG) data and training models associated with Reality Labs' paper

DiffQ performs differentiable quantization using pseudo quantization noise. It can automatically tune the number of bits used per weight or group of weights, in order to achieve a given trade-off between model size and accuracy.

VICRegL official code base

GPU Cluster Monitoring (GCM): Large-Scale AI Research Cluster Monitoring

Train universal codec avatars

Nymeria: a massive collection of multimodal egocentric daily motion in the wild

Global Reasoning module for visual recognition

Repository to track versions of concrete strength data, models, and active learning proposals.

Geometric Retargeting A Principled, Ultrafast Neural Hand Retargeting Algorithm

Code for “FlowMM Generating Materials with Riemannian Flow Matching” and "FlowLLM: Flow Matching for Material Generation with Large Language Models as Base Distributions"

Fashion++: Minimal Edits for Outfit Improvement

MTM Masked Trajectory Models for Prediction, Representation, and Control.

Data and sample evaluation codes for Multimodal Rewardbench 2

AssemblyHands Toolkit is a Python package that provides data loader, visualization, and evaluation tools for the AssemblyHands dataset (CVPR 2023).

This repository contains the training code of ParetoQ introduced in our work "ParetoQ Scaling Laws in Extremely Low-bit LLM Quantization"

Direct solver for sparse SPD matrices for nonlinear optimization. Implements supernodal Cholesky decomposition algorithm, and supports GPU (CUDA).

An open source environment for digital agents.

GoTrack: Generic 6DoF Object Pose Refinement and Tracking, CV4MR 2025

Reinforcement Learning Assembly

Official implementation of the WASP web agent security benchmark

Superintelligent Retrieval Agent (SIRA)

The code for creating the iGSM datasets in papers "Physics of Language Models Part 2.1, Grade-School Math and the Hidden Reasoning Process" (arxiv 2407.20311) and "Physics of Language Models Part 2.2, How to Learn From Mistakes on Grade-School Math Problems" (arxiv 2408.16293)

MobileLLM-R1

Official code and data from DexWM ("World Models Can Leverage Human Videos for Dexterous Manipulation").

Autoform Bot

[CVPR 2026 Highlight] Leveraging latent world model's physics understanding to improve the physics plausibility of video generation

Memory Mosaics are networks of associative memories working in concert to achieve a prediction task.

Python interface for https//github.com/facebookresearch/vrs.

Efficient image to 3D geometry foundation models from Meta Reality Labs for monocular depth, point maps, and surface normals. Featuring HyDen (ICLR 2026).

A research workbench for developing and testing attacks against large language models, with a focus on prompt injection vulnerabilities and defenses.

A framework for training, evaluating, and comparing state-of-the-art zero-shot reinforcement learning methods. It allows reproducing the experiments of the paper "TD-JEPA Latent-predictive Representations for Zero-Shot Reinforcement Learning"

We introduce DiffH2O, a diffusion-based framework to synthesize dexterous hand-object interactions. DiffH2O generates realistic hand-object motion from natural language, generalizes to unseen objects at test time and enables fine-grained control over the motion with detailed textual descriptions.

[CVPR 26'] Code for the LAMP research paper.

code to reproduce the empirical results in the research paper

swebench repro script for running confucius-code-agent (CCA)

Aria Training and Evaluation Kit

Code for the paper "Verifying Chain-of-Thought Reasoning via its Computational Graph".

Dance is an end-to-end framework that detects and classifies events in EEG signals. In a single forward pass, it extracts a set of events directly from the raw, unaligned recording.

Code for the ICML 2025 paper "SelfCite Self-Supervised Alignment for Context Attribution in Large Language Models"

This repository provides a comprehensive benchmark for evaluating the performance of neural watermarking techniques. The benchmark includes a variety of datasets, evaluation metrics, and tools for training and testing neural networks for watermarking.

[ICLR 2026 Oral] Official Implementation of the paper "MetaEmbed: Scaling Multimodal Retrieval at Test-Time with Flexible Late Interactions"

Body motion estimation from monocular videos via two stage diffusion (CVPR 2026).

LSRM is a SOTA, feed-forward 3D reconstruction model that generates high-fidelity, relightable 3D digital twins from sparse 2D views.

Code repository for Group-MATES Group-Level Data Selection for Efficient Pretraining

PyTorch original implementation of "Making Hard Problems Easier with Custom Data Distributions and Loss Regularization A Case Study in Modular Arithmetic" (ICML 2025).

Repository for the EgoBabyVLM Challenge

The repository contains raw data results and code for scaling laws fitting and visualization used in "Compute Optimal Tokenization" paper.

The repository provides code for running STyMo Fast and Controllable Few-Shot Motion Style Transfer.

Efficient Calculation of Shapley Values in Critical Path Method (CPM) Networks for Concurrent Delay Analysis

HST-Bench evaluation dataset contains 753 agentic tasks along with the time taken by human annotators to solve each task. This dataset was collected as part of our ICML 2026 paper on Scaling Small Agents Through Strategy Auctions https//arxiv.org/pdf/2602.02751