Researchers have developed EquiME, a novel synthetic dataset for micro-expression recognition, addressing limitations in existing datasets such as scale and demographic coverage. This dataset comprises 75,000 videos generated using an AU-guided image-to-video pipeline, incorporating five target emotions and automatically inferred metadata. Evaluations indicate that models trained on EquiME achieve competitive performance on established datasets like SAMM and CASME II, demonstrating its potential as a valuable resource for micro-expression research. AI
IMPACT Enables more robust micro-expression recognition models by providing a larger, more diverse synthetic dataset.
RANK_REASON The item describes a new dataset and methodology for micro-expression recognition, presented in an arXiv paper. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
