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.
Induced disgust, happiness and surprise: an addition to the mmi facial expression database,
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
-
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.