Spaces:
Sleeping
Sleeping
A newer version of the Gradio SDK is available:
5.49.1
metadata
title: FlowIID Intrinsic Image Decomposition
emoji: π¨
colorFrom: blue
colorTo: purple
sdk: gradio
sdk_version: 4.0.0
app_file: app.py
pinned: false
license: mit
FlowIID: Single-Step Intrinsic Image Decomposition
This space demonstrates FlowIID, a novel approach for intrinsic image decomposition that separates images into albedo (reflectance) and shading components using latent flow matching.
Features
- Single-step inference: Results in just one forward pass
- Parameter efficient: Only 52M parameters
- Real-time processing: Fast decomposition for interactive use
How to Use
- Upload any image (JPG, PNG)
- Click "Submit" to process
- Get albedo and shading components instantly!
Model Details
- Architecture: VAE + UNet with Flow Matching
- Training: Hypersim, InteriorVerse, MID datasets
- Performance: State-of-the-art results on MIT Intrinsic and ARAP benchmarks
Links
- π Paper
- π» GitHub Repository