A hybrid approach for video memorability prediction
Viola, Alexander J.
MetadataShow full item record
In this working note, we present our approach and investigation on the Predicting Media Memorability Task at MediaEval 2019. We used original video frames and caption data from the provided dataset, but extracted features ourselves using selected pretrained networks. Several combinations of pretrained models and recurrent networks were attempted to conduct a comparative study. Official results, as well as our investigation on the task data are provided. The best combination yielded a Spearman’s correlation score of 0.455 on short-term task and 0.218 on long-term task in the test set.
Viola, A., & Yoon, S. (2019). A hybrid approach for video memorability prediction. CEUR Workshop Proceedings, 2670 (Working Notes Proceedings of the MediaEval 2019 Workshop).
Department of Computer Science