Generative and Advanced Methods in Vision

This session explores the rapidly advancing domain of generative models and their applications in computer vision analytics.

We review generative adversarial networks (GANs) such as DCGAN and StyleGAN, as well as diffusion-based methods like Stable Diffusion and DALL·E. These models enable the creation of realistic synthetic images and open new possibilities for data augmentation in scenarios where labeled data is limited—a common challenge in analytics.

Application areas such as synthetic data generation, image inpainting, super-resolution, and denoising will be discussed with practical demonstrations. Additionally, students will analyze the ethical implications and potential bias introduced by generative AI when applied to analytics tasks. Real-world case studies drawn from medical imaging, industrial inspection, and retail product analytics will illustrate the transformative impact and limitations of recent advances.

Hands-on components will involve generating synthetic images with pre-trained models, augmenting datasets, and evaluating how these approaches improve the performance of previously studied detection or classification tasks. Students will conclude this session with a clear understanding of how generative and advanced vision methods are shaping the future of data-driven image analytics.

Required Reading and Listening

TBA