A Metadata Generation System with Semantic Understanding for Video Retrieval in Film Production

In film production, metadata plays an important role in original raw video indexing and classification within the industrial post-production software. Inspired by deep visual-semantic methods, we propose an automated image information extraction process to extend the diversity of metadata entities for massive large-scale raw video searching and retrieval. In this paper, we introduce the proposed system architecture and modules, integrating semantic annotation models and user-demand-oriented information fusion. We conducted experiments to validate the effectiveness of our system on Film Raw Video Semantic Annotation Dataset (Film-RVSAD) and Slate Board Template Dataset (SBTD), two benchmark datasets built for cinematography-related semantic annotation and slate detection. Experimental results show that the proposed system provides an effective strategy to improve the efficiency of metadata generation and transformation, which is necessary and convenient for collaborative work in the filmmaking process.