AI-based Video Content Understanding for Automatic and Interactive Multimedia Retrieval

We present diveXplore, a distributed system for AI-based video content understanding and retrieval, which will be used in the interactive task of the IViSE 2025 workshop. The system combines state-of-the-art deep learning components for shot segmentation, text and speech recognition, and multimodal embeddings with a scalable architecture designed for efficient storage, querying, and user interaction. A key feature of the frontend is an intuitive web-based GUI that supports free-text and semantic search, video summarization, and temporal query composition. We evaluate the performance of a newly developed keyframe scrubbing feature and conduct a qualitative user experiment based on all IViSE 2025 KIS tasks. The results demonstrate the system’s effectiveness in interactive video retrieval and inform a set of improvements for future versions.