CReLeRI: Explainable, Concept-centric, Representation, Learning, Reasoning, and Interaction Video Analysis System
Abstract
Existing video analysis models often lack explainability, struggle with long videos, and hallucinate. Commercial solutions are closed-source and costly. We introduce CReLeRI, an open-source system for action detection in untrimmed videos. CReLeRI integrates segmentation, action detection, argument detection, and grounding to improve interpretability and reduce hallucinations, enhancing transparency and trust in AI-driven video analysis. This paper is accompanied by a demonstration video.