Fusionista: Fusion of 3-D Information of Video in Retrieval System
摘要
This paper presents Fusionista, an interactive video retrieval system developed for the Visual Browser Showdown 2025. Utilizing CLIP-based feature extraction, Fusionista captures comprehensive visual representations to boost search performance. It also integrates multiple modalities, including color information, textual information, image similarity, and speech information. Additionally, we introduce fusion search, which divides and combines various search functions for efficiency, and temporal search, which identifies the best frame sequences. The user interface is also optimized for fast, intuitive navigation, making it user-friendly.