CLIP-AE uses CLIP features, audio, cross-attention fusion, and two self-supervised losses to improve unsupervised temporal action localization, with new top scores on THUMOS14 and ActivityNet v1.2.
Pre-trained audio, VLP, and CBP feature extractors first extractF audio,F CBP , andF V LP features
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
CLIP-AE: CLIP-assisted Cross-view Audio-Visual Enhancement for Unsupervised Temporal Action Localization
CLIP-AE uses CLIP features, audio, cross-attention fusion, and two self-supervised losses to improve unsupervised temporal action localization, with new top scores on THUMOS14 and ActivityNet v1.2.