AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Most Important Person-Guided Dual-Branch Cross-Patch Attention for Group Affect Recognition

Hongxia Xie, Ming-Xian Lee, Tzu-Jui Chen, Hung-Jen Chen, Hou-I Liu, Hong-Han Shuai, Wen-Huang Cheng

ICCV 2023 · Label-efficient and robust learning › Recognition and long-tail · approach: Uncertainty-aware facial expression recognition · 18 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use only the Most Important Person face instead of all faces (ablation)CrossViT with global + faces · GroupEmoWaccuracy (%): 85.01 -> 88.86Improved
add MIP input to global input (dual-path) (ablation)CrossViT · GAF 3.0accuracy (%): 75.56 -> 77.29Improved
use MIP only without global image (ablation)DCAT · GroupEmoWaccuracy (%): 89.55 -> 80.08Hurt
replace CrossViT class-token cross-attention with Cross-Patch Attention (ablation)DCAT-ViT · GroupEmoWaccuracy (%): 87.86 -> 89.55Improved
add token ranking module to dual-path + CPA (ablation)DCAT-ViT · GAF 3.0accuracy (%): 76.21 -> 77.29Improved
add CPA module on top of dual-path input (ablation)DCAT-CSwin-L · GroupEmoWaccuracy (%): 90.12 -> 90.47Improved
add dual-pathway input (MIP) to CrossViT (ablation)CrossViT · GroupEmoWaccuracy (%): 88.48 -> 88.86Improved
drop a certain percentage of tokens in MIP branch (ablation)DCAT · dropping ratio studyaccuracyImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy