dluvizon/deephar

Deep human action recognition and pose estimation

What it solves

This project provides a deep learning framework for simultaneous 2D and 3D human pose estimation and action recognition. It addresses the challenge of performing these tasks together in real-time, allowing a system to identify both the positions of human joints and the actions being performed by the person in a video or image.

How it works

The software implements multi-task deep learning models that can bet trained on various datasets (such as MPII, Human3.6M, and NTU) to perform multiple computer vision tasks. Depending on the version, it can estimate 2D poses from images, lift 2D poses to 3D, and classify actions based on those poses.

Who it’s for

This is primarily for researchers and developers working in computer vision, specifically those focusing on human motion analysis, skeletal animation, and action recognition.

Highlights

  • Multi-task learning approach for combined pose estimation and action recognition.
  • Supports both 2D and 3D pose estimation.
  • Evaluated on public datasets including MPII, Human3.6M, and PennAction, and NTU.
  • Real-time performance capabilities for 3D human pose estimation.

Related

  • Project
    talmolab/sleap

    A deep-learning framework for multi-animal pose tracking that uses a human-in-the-loop GUI to rapidly label and quantify animal behavior.

  • Project
    NVIDIA-ISAAC-ROS/isaac_ros_pose_estimation

    A collection of NVIDIA-accelerated ROS 2 packages for 3D object pose estimation, enabling robots to locate and track both known and novel objects for navigation and manipulation.

  • Project
    FORTH-ModelBasedTracker/MocapNET

    A real-time 2D-to-3D human pose estimator that converts RGB video feeds into standard BVH motion capture files for 3D animation.

  • Project
    deepinsight/insightface

    An open-source 2D and 3D face analysis toolbox providing state-of-the-art algorithms for face detection, recognition, and alignment.

  • Project
    nghorbani/human_body_prior

    VPoser is a variational human pose prior that provides a low-dimensional representation of valid 3D human poses to improve Inverse Kinematics and body model fitting.