Beyond Supervised Learning with Zero-Shot Vision

Explore the evolution of zero-shot image classification in computer vision in this 28-minute conference talk from DSC EUROPE 24 in Belgrade. Discover how modern models can classify images of previously unseen objects without specific training examples by leveraging semantic knowledge and the relationships between visual features and textual descriptions. Learn about cutting-edge techniques like CLIP, understand current limitations in the field, and see real-world applications across domains including image retrieval and autonomous vehicles. This presentation by Nemanja Milosevic provides valuable insights into how computer vision is moving beyond traditional supervised learning approaches.