Pixel-Talk/PEAR

[SIGGRAPH‘2026] PEAR :Pixel-aligned Expressive humAn mesh Recovery

What it solves

PEAR solves the challenge of real-time, expressive 3D human mesh recovery. It allows for the simultaneous prediction of expressive human mesh (EHM-s) parameters from images or video, achieving high speed and detail without sacrificing expressiveness.

How it works

PEAR is a unified framework that uses a pixel-aligned approach to recover 3D human meshes. It is designed for efficiency, enabling the process to occur in real-time at 100 FPS, allowing it to capture the expressive movements of the human body, face, and hands.

Who it’s for

This project is intended for researchers and developers working in 3D human reconstruction, computer vision, and digital human animation, as well as those utilizing models like SMPL, SMPLX, and FLAME.

Highlights

  • Real-time performance: Capable of predicting EHM-s parameters at 100 FPS.
  • Unified framework: A single system for expressive 3D human mesh recovery.
  • Support for multiple models: Compatible with SMPL, SMPLX, and FLAME.
  • Versatile input: Supports both image and video inference.

Related

  • Project
    facebookresearch/MHR

    A high-fidelity parametric 3D human body model that uses identity, pose, and facial expression parameters to generate realistic, differentiable human meshes.

  • Project
    PozzettiAndrea/ComfyUI-SAM3DBody

    A ComfyUI wrapper for Meta's SAM 3D Body that enables full-body 3D human mesh recovery from a single image.

  • Project
    spear-sim/spear

    A Python library for photorealistic embodied AI research that provides deep programmatic control over Unreal Engine applications with high-speed rendering and extensive API access.

  • Project
    phamquiluan/ResidualMaskingNetwork

    A facial expression recognition system using a Residual Masking Network to detect human emotions from images and live video feeds.

  • Dispatch
    BAIR PEVA: Whole-Body Conditioned Egocentric Video Prediction

    BAIR introduces PEVA, a world model for embodied agents that predicts egocentric video frames based on high-dimensional whole-body kinematic pose trajectories.