THAU-UPM at MediaEval 2021: from video semantics to memorability using pretrained transformers
- This paper reports on our experience after participating at the MediaEval 2021: Predicting Media Memorability challenge. The memorability of a video is defined as the proportion of people that successfully remembered having watched a video on a second viewing during a memory game. Given this setup, teams were requested to provide systems able to predict the degree of memorability for individual videos from two different datasets: TRECVid and Memento10k. Our proposal builds upon previous work in which we find that non-adapted features extracted from Transformer architectures can be closely tied to semantic differences between samples, which in turn point to the overall memorability degree within different semantic units, or topics. We feed these precomputed features to linear regressors, showing that even without adapting the input representation competitive prediction rates can be achieved.
Download full text files
Statistics
Print On Demand
Additional Services
| Author: | Ricardo Kleinlein, Cristina Luna-JiménezORCiDGND, Fernando Fernández Martínez |
|---|---|
| URN: | urn:nbn:de:bvb:384-opus4-1226893 |
| Frontdoor URL | https://opus.bibliothek.uni-augsburg.de/opus4/122689 |
| URL: | https://nbn-resolving.org/urn:nbn:de:0074-3181-0 |
| ISSN: | 1613-0073OPAC |
| Parent Title (English): | MediaEval 2021 - Multimedia Benchmark 2021: Working Notes Proceedings of the MediaEval 2021 Workshop, online, 13-15 December 2021 |
| Publisher: | CEUR-WS |
| Place of publication: | Aachen |
| Editor: | Steven Hicks, Konstantin Pogorelov, Andreas Lommatzsch, Alba García Seco De Herrera, Pierre-Etienne Martin, Syed Zohaib Hassan, Alastair Porter, Asem Kasem, Stelios Andreadis, Mathias Lux, Marc Gallofré Ocaña, Alex Liu, Martha Larson |
| Type: | Conference Proceeding |
| Language: | English |
| Year of first Publication: | 2021 |
| Publishing Institution: | Universität Augsburg |
| Release Date: | 2025/06/05 |
| First Page: | 42 |
| Series: | CEUR Workshop Proceedings ; 3181 |
| Institutes: | Fakultät für Angewandte Informatik |
| Fakultät für Angewandte Informatik / Institut für Informatik | |
| Fakultät für Angewandte Informatik / Institut für Informatik / Lehrstuhl für Menschzentrierte Künstliche Intelligenz | |
| Dewey Decimal Classification: | 0 Informatik, Informationswissenschaft, allgemeine Werke / 00 Informatik, Wissen, Systeme / 004 Datenverarbeitung; Informatik |
| Licence (German): |  CC-BY 4.0: Creative Commons: Namensnennung (mit Print on Demand) |
