Blind source separation
Separating mixed MNIST images with U-Net and autoencoder-decoder architectures. U-Net reached an MSE of 0.00086.
- Trained U-Net and autoencoder-decoder models to separate mixed MNIST images
- U-Net reached an MSE of 0.00086, against 0.0048 for the autoencoder-decoder
- Date
- Feb 2025
- Domain
- Deep learning & NLP
- Stack
- TensorFlow, Keras, U-Net