AmmarkoV

AmmarkoV / SAM3DBody-cpp

Public

Real-time 3D full-body reconstruction from a single camera, C++ runtime, ONNX + ggml, 70-joint skeleton with hands.

13
0
85% credibility
Found May 24, 2026 at 20 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
C
AI Summary

This is a C++ implementation of a 3D body pose estimation system. It takes ordinary photos or videos as input and outputs detailed 3D information about where each person's body parts are positioned in space. The system uses a combination of AI models (including a vision transformer called DINO and a person detector called YOLO) to first locate people, then estimate their 3D pose including hand positions, facial orientation, and full body shape. Results can be viewed live, exported to spreadsheets, or integrated with robotics systems. The project runs entirely on your computer without needing to send data to external services.

How It Works

1
📸 You take a photo or video

You capture a picture or video of yourself or someone else moving around.

2
🧠 The AI spots each person

The system automatically finds where each person is in the image and identifies their key body parts like head, hands, and feet.

3
✨ A 3D skeleton comes to life

The magic happens as the AI builds a complete 3D skeleton for each person, showing exactly how they're posed in real space.

4
Choose how you want to see the results
🖥️
Watch live on your screen

See the 3D body model rendered directly on the video feed with body landmarks highlighted.

📊
Export to CSV

Save all 70 joint positions to a spreadsheet so you can analyze the motion later.

5
🤖 Connect to robotics or research tools

For researchers and robotics developers, the system publishes body pose data directly to ROS topics for real-world applications.

🎯 You get precise 3D body tracking

Whether for animation, health monitoring, or robotics, you end up with accurate 3D body pose data that's ready to use.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 20 to 13 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is SAM3DBody-cpp?

This is a standalone C++ inference engine for real-time 3D full-body reconstruction from a single camera. It takes a BGR image and produces per-person body pose parameters, camera translation, 3D mesh vertices, and 70 body/hand keypoints using ONNX Runtime and ggml. The project offers a zero-Python dependency runtime, making it suitable for embedding in production applications.

Why is it gaining traction?

The ability to reconstruct a full 70-joint skeleton including hands from monocular video is rare in open-source tooling. The hybrid ONNX + ggml approach delivers fast inference while maintaining accuracy. Multiple frontends are provided: a CLI executable, Python bindings via ctypes, and a ROS2 node for robotics integration.

Who should use this?

Game developers building real-time character animation from webcam input will find this useful. Robotics researchers needing body pose estimation for human-robot interaction are another key audience. Developers creating avatar systems for virtual communication or motion capture pipelines could also benefit. The C API with ctypes support makes it accessible across different programming languages.

Verdict

This project shows promise with solid documentation and a practical architecture, though the low star count suggests limited community validation. The 0.8500000238418579% credibility score indicates reasonable technical foundation, but maturity concerns warrant careful evaluation before production deployment.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.