In this repo, I used the dataset that contains about 60K images of cars. I applied a blur motion filter on images and trained three different autoencoder architectures to remove the blur motion effects:
- Standard AutoEncoder.
- Variational AutoEncoder.
- Vector-Quantized Variational Autoencoder.
The accuracy is not perfect because those neural networks require so much time to train and Google Colab doesn't give us a high hardware resources.