This computer vision project used GAN-based inpainting to reconstruct missing image regions in emoji data. The work centered on pairing a generator and discriminator, balancing adversarial learning with reconstruction loss, and evaluating where the model restored coherent structure versus where high-frequency masks still produced visible artifacts.
Computer Vision / 2025
Image Inpainting with GANs
Implemented a Generative Adversarial Network to reconstruct masked or damaged emoji image regions, combining a U-Net style generator, PatchGAN-style discrimination, adversarial learning, reconstruction loss, and evaluation of visual artifacts.