Skip to content

README.md

Latest commit

 

History

History
96 lines (72 loc) · 2.46 KB

README.md

File metadata and controls

96 lines (72 loc) · 2.46 KB

Real-time 3D Motion Generation Demo

Quick Start

cd web_demo
./server.sh start

Visit: http://localhost:5000
Refresh: Ctrl+Shift+R (clear cache)

Features

  • 🎨 Text-driven: Input motion descriptions to generate 3D animations in real-time
  • 🎭 Streaming Rendering: Smooth 20 FPS playback
  • 🎪 Colorful Stick Figure: Different colors for different body parts
  • 📹 Smart Camera Follow: Auto-follow after 3 seconds of inactivity
  • 🌐 Infinite Scene: 1000x1000 unit ground with grid

Usage

  1. Enter motion description (e.g., "walk forward", "jump", "dance")
  2. Click "Start"
  3. Watch real-time generated 3D animation
  4. Use "Update Text" to change motion anytime
  5. Mouse drag to rotate view, scroll to zoom
  6. Camera auto-follows character after 3 seconds of inactivity

Architecture

Text Input → Model(1 token) → VAE(4 frames) →
StreamJointRecovery(22 joints) → Buffer → 20fps Rendering

File Structure

web_demo/
├── app.py                # Flask server
├── model_manager.py      # Model and buffer management
├── server.sh            # Server management script
├── start.sh             # Legacy startup script
├── static/
│   ├── css/style.css    # Clean white UI
│   └── js/
│       ├── main.js      # Main logic (20fps + camera follow)
│       └── skeleton.js  # Stick figure rendering (colorful)
└── templates/
    └── index.html       # Web interface

Performance

  • Model Generation: 52 FPS on 5090 (76ms/step)
  • Rendering: 20 FPS (matches model output)
  • Buffer: 4 frames minimum

Technical Details

StreamJointRecovery263

  • Process 263-dim motion data frame by frame
  • Output 22 joint 3D coordinates
  • Test verified: Identical to batch processing

HumanML3D Skeleton

[0,2,5,8,11]      # Spine (orange)
[0,1,4,7,10]      # Left leg (cyan)
[0,3,6,9,12,15]   # Right leg (blue)
[9,14,17,19,21]   # Left arm (amber)
[9,13,16,18,20]   # Right arm (aquamarine)

Server Management

Use server.sh for reliable process management:

./server.sh start    # Start server
./server.sh stop     # Stop server
./server.sh restart  # Restart server
./server.sh status   # Check status

Tiny Model

You can use different model configurations by specifying a config file:

./server.sh start configs/stream_tiny.yaml
./server.sh restart configs/stream_tiny.yaml