MiRA is a parameter-free frame-marginal attention redistribution technique for ViT video models that improves sensitivity to localized facial cues on FER benchmarks.
arXiv preprint arXiv:2205.04749 (2022) 18 Yoon et al
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Reweighting Framewise Attention in Video Transformers for Facial Expression Understanding
MiRA is a parameter-free frame-marginal attention redistribution technique for ViT video models that improves sensitivity to localized facial cues on FER benchmarks.