Reproduce “ARGen: Affect-Reinforced Generative Augmentation towards Vision-based Dynamic Emotion Perception” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Huanzhen Wang, Ziheng Zhou, Jiaqi Song, Li He, Yunshi Lan, Yan Wang, Wenqiang Zhang

ARGen is a framework that improves vision-based dynamic facial emotion perception by generating high-quality, affectively consistent video augmentations for scarce emotion categories in real-world datasets. It achieves substantial gains in Unweighted Average Recall for underrepresented emotions on DFEW and FERV39k datasets and demonstrates superior generative quality with a Fréchet Video Distance of 56.75 compared to baselines.

Key insights

  • Injecting explicit affective semantic priors (via ASI) into the generative process is crucial for producing naturalistic and emotionally consistent dynamic expression augmentations, especially for scarce categories.
  • An adaptive reinforcement learning strategy for diffusion models (via ARD) can dynamically optimize generation parameters, leading to artifact-free videos that preserve facial integrity and improve efficiency.
  • Targeted augmentation for underrepresented classes, rather than uniform enhancement, effectively mitigates data scarcity and long-tail distribution issues in DFER, improving model robustness without requiring external datasets.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.