Autoencoder for docs/books, etc. with 4 decode and 4 encode layers and ReLU-activations. You can use this to remove noise from photos or document scans. Model output is an image of 640x640 image size. Model class in AutoEncoder_model.py, it weights in weights/improved_model_weights. Model was fitted by shabby-docs dataframe with 981/290/186 images in train/valid/test on 35 epochs. It was written on PyTorch 2.1.
This service is implemented on Streamlit