Litcius/Paper detail

Interventional Video Relation Detection

Yicong Li, Xun Yang, Xindi Shang, Tat‐Seng Chua

202159 citationsDOIOpen Access PDF

Abstract

Video Visual Relation Detection (VidVRD) aims to semantically describe the dynamic interactions across visual concepts localized in a video in the form of subject, predicate, object. It can help to mitigate the semantic gap between vision and language in video understanding, thus receiving increasing attention in multimedia communities. Existing efforts primarily leverage the multimodal/spatio-temporal feature fusion to augment the representation of object trajectories as well as their interactions and formulate the prediction of predicates as a multi-class classification task. Despite their effectiveness, existing models ignore the severe long-tailed bias in VidVRD datasets. As a result, the models' prediction will be easily biased towards the popular head predicates (e.g., next-to and in-front-of), thus leading to poor generalizability.

Topics & Concepts

Computer scienceGeneralizability theoryLeverage (statistics)Artificial intelligenceRelation (database)Object detectionPredicate (mathematical logic)Video trackingClass (philosophy)Machine learningNatural language processingObject (grammar)Pattern recognition (psychology)Data miningMathematicsStatisticsProgramming languageHuman Pose and Action RecognitionMultimodal Machine Learning ApplicationsVideo Surveillance and Tracking Methods
Interventional Video Relation Detection | Litcius