zju3dv/easymocap

★ 4,602PythonAudience · researcherComplexity · 4/5Setup · hard

Mindmap

mindmap
  root((repo))
    What it does
      Human motion capture
      3D pose estimation
    Recording setups
      Multi-camera
      Single video
      Mirror mode
      Multi-person
    Features
      Novel view synthesis
      Camera calibration
      Real-time 3D view
    Tech stack
      Python
      Computer vision

mindmap root((repo)) What it does Human motion capture 3D pose estimation Recording setups Multi-camera Single video Mirror mode Multi-person Features Novel view synthesis Camera calibration Real-time 3D view Tech stack Python Computer vision

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Capture 3D human body motion from multiple video cameras for animation or academic research

USE CASE 2

Estimate body pose from a single internet video without special motion-capture equipment

USE CASE 3

Use a mirror in your recording to get an extra camera viewpoint without buying another camera

USE CASE 4

Synthesize what a scene would look like from a new camera angle never present in the original footage

Tech stack

Python

Getting it running

Difficulty · hard Time to first run · 1day+

Requires camera calibration before use and specific Python dependencies, multi-camera setups need synchronized hardware.

In plain English

EasyMocap is a Python toolbox from Zhejiang University that makes it easier to capture how a human body moves, using only regular video cameras instead of the special reflective-marker suits you see in film studios. Given video footage of a person, the software figures out the 3D position and pose of their body, hands, and face, and produces a digital model of that motion. The project supports several different recording setups. You can feed it footage from multiple cameras pointing at the same person, or give it a single internet video from somewhere like YouTube. There is also a mode for videos that include a mirror in the shot, since the mirror reflection provides an extra viewpoint without needing an extra camera. Another mode handles footage of multiple people moving at the same time, tracked across several camera angles simultaneously. Beyond capturing motion, EasyMocap can also generate new views of a scene that were never filmed. Given sparse camera coverage, it can synthesize what the scene would look like from a completely different angle. This technique has been used in published research for things like rendering human interactions from novel viewpoints. The toolkit ships with helper tools for calibrating cameras (getting their positions and lens characteristics right before a capture session) and for annotating video frames with bounding boxes and body keypoints. There is also a real-time 3D visualization component for watching motion data as it is processed. The team behind this project also released a large dataset of human motion recordings called ZJU-MoCap, captured in a professional LightStage facility. Researchers can request access to that dataset by signing an agreement and emailing the authors. Several other published research projects have built on top of both the EasyMocap codebase and the ZJU-MoCap dataset.

Copy-paste prompts

Prompt 1

I have footage from 4 synchronized cameras of a person moving. Using zju3dv/easymocap, walk me through how to reconstruct their 3D body pose from this multi-camera setup.

Prompt 2

Using easymocap, I want to extract 3D motion data from a single YouTube video. What command do I run and what output format does it produce?

Prompt 3

Help me calibrate my 3-camera setup for zju3dv/easymocap before a capture session, what does the calibration step require and what does it output?

Prompt 4

I want to visualize motion capture data in real time using zju3dv/easymocap. Walk me through running the 3D visualization component while processing video.

Open on GitHub → Explain another repo

← zju3dv on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.