FaceSleuth: Learning-Driven Single-Orientation Attention Verifies Vertical Dominance in Micro-Expression Recognition.
Abstract
Micro-expression recognition (MER) has achieved impressive accuracy in controlled laboratory settings. However, its real-world applicability faces a significant generalization cliff, severely hindering practical deployment due to poor performance on unseen data and susceptibility to domain shifts. Existing attention mechanisms often overfit to datasetspecific appearance cues or rely on fixed spatial priors, making them fragile in diverse environments. We posit that robust MER requires focusing on quasi-invariant motion orientations inherent to micro-expressions, rather than superficial pixel-level features. To this end, we introduce FaceSleuthR, a framework centered on our novel Single-Orientation Attention (SOA) module. SOA is a lightweight, differentiable operator that enables the network to learn layer-specific optimal orientations, effectively guiding attention towards these robust motion cues. Through extensive experiments, we demonstrate that SOA consistently discovers a universal near-vertical motion prior across diverse datasets. More critically, FaceSleuth-R showcases superior generalization in rigorous Leave-One-Dataset-Out (LODO) protocols, significantly outperforming baselines and state-of-the-art methods * Equal contribution. (a) when confronted with domain shifts. Furthermore, our approach establishes state-of-the-art results across several benchmarks. This work highlights adaptive orientation-aware attention as a key paradigm for developing truly generalized and high-performing MER systems. Index Terms- Micro-Expression Recognition, OrientationAware Attention, Cross-Dataset Generalization 1. INTRODUCTION Micro-expressions (MEs) are involuntary, millisecond-long facial muscle activations that reveal concealed emotions [1]. Their diagnostic value has found applications in psychotherapy [2], criminal investigations, and national security [3]. However, MER has reached a critical juncture. While current models demonstrate high accuracy in intra-dataset evaluations, they suffer a dramatic performance decline when tested across datasets. This "generalization gulf" highlights a fundamental flaw: existing methods often overfit to dataset-specific featur