Temporal and cross-modal attention for audio-visual zero-shot learning

20 Jul 2022  ·  Otniel-Bogdan Mercea, Thomas Hummel, A. Sophia Koepke, Zeynep Akata ·

Audio-visual generalised zero-shot learning for video classification requires understanding the relations between the audio and visual information in order to be able to recognise samples from novel, previously unseen classes at test time. The natural semantic and temporal alignment between audio and visual data in video data can be exploited to learn powerful representations that generalise to unseen classes at test time. We propose a multi-modal and Temporal Cross-attention Framework (\modelName) for audio-visual generalised zero-shot learning. Its inputs are temporally aligned audio and visual features that are obtained from pre-trained networks. Encouraging the framework to focus on cross-modal correspondence across time instead of self-attention within the modalities boosts the performance significantly. We show that our proposed framework that ingests temporal features yields state-of-the-art performance on the \ucf, \vgg, and \activity benchmarks for (generalised) zero-shot learning. Code for reproducing all results is available at \url{https://github.com/ExplainableML/TCAF-GZSL}.

PDF Abstract

Datasets


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
GZSL Video Classification ActivityNet-GZSL (cls) TCaF HM 12.20 # 3
ZSL 7.96 # 3
GZSL Video Classification ActivityNet-GZSL(main) TCaF HM 10.71 # 4
ZSL 7.91 # 4
GZSL Video Classification UCF-GZSL (cls) TCaF HM 50.78 # 2
ZSL 44.64 # 3
GZSL Video Classification UCF-GZSL(main) TCaF HM 31.72 # 2
ZSL 24.81 # 2
GZSL Video Classification VGGSound-GZSL (cls) TCaF HM 8.77 # 2
ZSL 7.41 # 2
GZSL Video Classification VGGSound-GZSL(main) TCaF HM 7.33 # 3
ZSL 6.06 # 3

Methods