GRACE pairs motion-weighted video tokens with AI-refined emotion text tokens using optimal transport, reporting new UAR and WAR records on DFEW, FERV39k, and MAFW.
The extended cohn-kanade dataset (ck+): A complete dataset for action unit and emotion-specified expression,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CV 1years
2025 1verdicts
CONDITIONAL 1roles
background 1polarities
background 1representative citing papers
citing papers explorer
-
From Coarse to Nuanced: Cross-Modal Alignment of Fine-Grained Linguistic Cues and Visual Salient Regions for Dynamic Emotion Recognition
GRACE pairs motion-weighted video tokens with AI-refined emotion text tokens using optimal transport, reporting new UAR and WAR records on DFEW, FERV39k, and MAFW.