📤 Upload & Configure
1 99
ℹ️ How It Works
- Upload any image
- Adjust the masking ratio
- Click Reconstruct
- View the results & metrics
The model randomly masks patches of your image and reconstructs the full image from only the visible parts!
🖼️ Reconstruction Results
📊 Quality Metrics & Analysis
Upload an image and click Reconstruct to see detailed metrics.
🎯 Try These Examples:
- Easy (10-30% masking): Clear reconstruction, tests basic capability
- Medium (40-60% masking): Balanced challenge, realistic scenarios
- Hard (70-85% masking): Significant challenge, impressive results
- Extreme (90-99% masking): Model's absolute limits
🔬 About MAE
Masked Autoencoders (MAE) are self-supervised learning models that learn visual representations by reconstructing masked images. This implementation uses:
- Asymmetric Encoder-Decoder: Efficient processing of visible patches
- ViT Architecture: Transformer-based vision understanding
- High Masking Ratio: Learns robust features from limited information
📄 Paper: Masked Autoencoders Are Scalable Vision Learners (He et al., 2021)