Advancing Image Understanding through Image Translation, Anomaly Detection, and Concept Retrieval

Image understanding in computer vision aims to extract structured, semantically meaningful information from raw visual data, enabling robust operation in complex environments. High-level visual semantics are essential for applications such as autonomous driving, medical diagnostics, and security systems. In this work, we focus on three key challenges—image translation, Anomaly detection and concept retrieval—all unified by a central theme: identifying and manipulating abstract concepts within the neural network embedding space. Image translation involves converting images across domains while preserving core semantics. Anomaly detection targets the identification of deviation from expected conceptual norms. Concept retrieval seeks to extract and match high-level visual concepts. By framing these diverse tasks through the lens of concept identification in learned representations, we highlight new opportunities for advancing image understanding and developing more robust, generalizable vision systems.

Ph.D. student Under the supervision of Prof. Ayellet Tal.