Partial Image Restoration Project using EBM (Energy-Based Model)

Overview

This project focuses on restoring masked images to their original state using an Energy-Based Model (EBM). EBM utilizes contrastive divergence to effectively handle cases where parts of images are missing or damaged, aiming for accurate and smooth restoration.

Project Goals

• Refactor existing TensorFlow code into PyTorch to implement EBM.
• Train EBM using masked images and evaluate the performance of restored original images.
• Provide an effective image restoration tool, particularly useful for scenarios where parts of images are missing or damaged.

Features

• EBM Training: The model is trained using contrastive divergence. The current implementation of contrastive divergence is undergoing refactoring and is not yet complete.
• Image Restoration: Given masked image inputs, EBM predicts missing parts to reconstruct the original image.

Results from TensorFlow Environment

Here are some example results obtained from the previous TensorFlow implementation:

These images showcase the output generated by the TensorFlow implementation before the transition to PyTorch. They demonstrate the initial capabilities of the image restoration process using the Energy-Based Model (EBM).