UniSH: Unifying Scene and Human Reconstruction in a Feed-Forward Pass
🛠️ Configuration
0 10
0 10
💡 Use Tips:
1. Contain only a single person.
2. No occlusion (self-occlusion is fine).
3. Keep the full body mostly visible.
4. Subject should not be too small.
🎥 Examples
Click to try:
▶️ Interactive Results
Upload a video and click Start to view result.
⚡ Performance Note:
Inference (feed-forward) is very fast, but generating visualization assets (meshing & rendering) takes up most of the processing time.
Inference (feed-forward) is very fast, but generating visualization assets (meshing & rendering) takes up most of the processing time.
👁️ Visual Quality:
The displayed results are downsampled for better web rendering performance.
The displayed results are downsampled for better web rendering performance.