UniCon+: ICTCAS-UCAS Submission to the AVA-ActiveSpeaker Task at ActivityNet Challenge 2022

TL;DR

This paper introduces UniCon+, a state-of-the-art active speaker detection model that builds on previous architectures with a GRU-based module, achieving top performance with 94.47% mAP at ActivityNet Challenge 2022.

Contribution

UniCon+ extends the Unified Context Network with a GRU module for better identity flow, setting new state-of-the-art results in active speaker detection.

Findings

Achieved 94.47% mAP on AVA-ActiveSpeaker test set.

Ranked first on the ActivityNet Challenge 2022 leaderboard.

Significantly improved over previous models.

Abstract

This report presents a brief description of our winning solution to the AVA Active Speaker Detection (ASD) task at ActivityNet Challenge 2022. Our underlying model UniCon+ continues to build on our previous work, the Unified Context Network (UniCon) and Extended UniCon which are designed for robust scene-level ASD. We augment the architecture with a simple GRU-based module that allows information of recurring identities to flow across scenes through read and update operations. We report a best result of 94.47% mAP on the AVA-ActiveSpeaker test set, which continues to rank first on this year's challenge leaderboard and significantly pushes the state-of-the-art.