A multimodal fusion pipeline with pretrained encoders, LLM-enriched text, attention fusion, label refinement, and ensembling reaches 87.49% weighted F-score on MER2025-SEMI.
Title resolution pending
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
unclear 1representative citing papers
citing papers explorer
-
ECMF: Enhanced Cross-Modal Fusion for Multimodal Emotion Recognition in MER-SEMI Challenge
A multimodal fusion pipeline with pretrained encoders, LLM-enriched text, attention fusion, label refinement, and ensembling reaches 87.49% weighted F-score on MER2025-SEMI.