yangtiming

yangtiming / Fast-SAM-3D-Body

Public

Fast SAM 3D Body: Accelerating SAM 3D Body for Real-Time Full-Body Human Mesh Recovery

yangtiming.github.ioFast-SAM-3D-Body-Page human-mesh-recovery real-time sam-3d-body
87
5
100% credibility
Found Mar 18, 2026 at 87 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Fast SAM 3D Body is a real-time system that reconstructs detailed 3D human body meshes from single RGB images, enabling applications like vision-based humanoid robot teleoperation.

How It Works

1
🔍 Discover Fast Human Tracker

You find this tool that turns everyday photos into instant 3D body models, perfect for making robots mimic human moves.

2
📥 Grab the Starter Kit

Download the ready-to-go package with example photos and one-click setup instructions.

3
🧠 Feed in Your Photo

Pick any picture of a person – your phone snap, webcam shot, or video frame – and watch it analyze the pose.

4
✨ See the 3D Magic

In seconds, it reveals a detailed 3D body model overlaid on your image, showing exact pose and shape.

5
⚡ Make It Super Fast

Hit optimize to speed it up 10x for live video, ready for robot brains or apps.

🤖 Power Your Robot

Your creation now tracks humans in real-time, letting robots copy dances, walks, or gestures perfectly.

Sign up to see the full architecture

4 more

Sign Up Free

Star Growth

See how this repo grew from 87 to 87 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 Fast-SAM-3D-Body?

Fast-SAM-3D-Body accelerates Meta's SAM 3D Body model for real-time full-body human mesh recovery from single RGB images or video streams. Developers feed in webcam footage and get SMPL-compatible 3D meshes, joints, and kinematics at ~65ms per frame on RTX 5090 hardware—perfect for accelerating body tracking without sacrificing accuracy. Built in Python with TensorRT and torch.compile optimizations, it auto-downloads checkpoints via fast GitHub actions for quick setup.

Why is it gaining traction?

It delivers up to 10x end-to-end speedup over the original, swapping slow iterative mesh fitting for a 10,000x faster neural mapping, while matching or beating benchmarks like LSPET. Real-time demos control Unitree G1 humanoids via vision-only teleop, and TensorRT conversion scripts enable fast GitHub downloads of optimized engines. For devs chasing low-latency pose estimation, this turns seconds-per-frame research code into deployable pipelines.

Who should use this?

Robotics engineers building humanoid teleoperation or manipulation policies from monocular cameras. AR/VR developers needing fast body tracking for avatars or interactions. Pose estimation researchers deploying models to edge devices like fast Samsung TV apps or real-time games.

Verdict

Grab it if real-time 3D body recovery is your bottleneck—strong demos and TensorRT tooling make it production-ready despite 87 stars and 1.0% credibility score. Maturity shows in setup scripts and robot integration, but expect tweaks for custom hardware.

(198 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.