Decompose the Sounds and Pixels, Recompose the Events

TL;DR

This paper introduces EDRNet, a novel architecture for audio-visual event localization that models event progress checkpoints and relationships, improving performance over previous methods.

Contribution

The paper presents EDRNet, a new framework that models event checkpoints, introduces a state machine video fusion, a novel loss function, and a label correction method for better AVE localization.

Findings

Outperforms state-of-the-art on AVE dataset

Effectively models event progress checkpoints and relationships

Improves weak supervision stability

Abstract

In this paper, we propose a framework centering around a novel architecture called the Event Decomposition Recomposition Network (EDRNet) to tackle the Audio-Visual Event (AVE) localization problem in the supervised and weakly supervised settings. AVEs in the real world exhibit common unravelling patterns (termed as Event Progress Checkpoints (EPC)), which humans can perceive through the cooperation of their auditory and visual senses. Unlike earlier methods which attempt to recognize entire event sequences, the EDRNet models EPCs and inter-EPC relationships using stacked temporal convolutions. Based on the postulation that EPC representations are theoretically consistent for an event category, we introduce the State Machine Based Video Fusion, a novel augmentation technique that blends source videos using different EPC template sequences. Additionally, we design a new loss function called the Land-Shore-Sea loss to compactify continuous foreground and background representations. Lastly, to alleviate the issue of confusing events during weak supervision, we propose a prediction stabilization method called Bag to Instance Label Correction. Experiments on the AVE dataset show that our collective framework outperforms the state-of-the-art by a sizable margin.