This project explores and implements a state-of-the-art approach for automatic image recolorization using Conditional Generative Adversarial Networks (cGANs).
Image colorization is a significant task in computer vision, aiming to add color to grayscale images. This project leverages cGANs for image-to-image translation, particularly for recoloring grayscale images.
The project uses a dataset of grayscale images with their corresponding colored versions. The dataset is split into training and validation sets.
The architecture consists of a Generator and a Discriminator. The Generator uses a U-Net structure, and the Discriminator uses a PatchGAN architecture.
The model is trained using the cGAN framework. The Generator learns to produce realistic color images, while the Discriminator learns to distinguish between real and generated images.
The trained model can recolor grayscale images with high accuracy, preserving details and maintaining visual appeal.