New Arquitectures and features

Autoencoder Project - Release 2.0.0

Release Date: October 13, 2023.

Highlights:

• Expanded model architectures, introducing Convolutional Autoencoder and Variational Autoencoder, including a combination of both.
• Implementation of checkpointing functionality, providing an advanced and seamless way to save and continue model training.
• Enhanced evaluation and visualization mechanisms to cater to different autoencoder architectures.

Features:

• Diverse Model Architectures: Users now have the flexibility to train not just a simple Autoencoder but also a Convolutional Autoencoder, a Variational Autoencoder, and a Convolutional Variational Autoencoder.
• Checkpointing: Advanced training control with checkpointing, allowing users to save intermediate states of training and resume from them whenever required.
• Enhanced Visualization: With the introduction of new models, visualization capabilities have been expanded to provide a clearer understanding of how different architectures perform.
• General Code Improvements: Refactoring for cleaner code, optimized imports, and better modularization.

Enhancements:

• Extended run.py to detect and handle different autoencoder architectures seamlessly.
• Updated trainer.py to handle the training nuances of the newly introduced autoencoder architectures.
• Modular design ensures easy extensibility to accommodate more sophisticated models in the future.

Usage:

1. Clone the repository and navigate to the project directory.
2. Install the necessary dependencies using pip install -r requirements.txt.
3. Adjust data paths and settings in settings/settings.py based on your dataset.
4. Decide on the autoencoder type and adjust the main method in run.py.
5. Run the main script with python run.py.
6. Post-training, visualize the reconstructed results displayed, which will now cater to the chosen autoencoder type.
7. Utilize the checkpointing feature to save intermediate training states and resume from them whenever required.