CMNet: Contrastive Magnification Network for Micro-Expression Recognition

Micro-Expression Recognition (MER) is challenging because the Micro-Expressions' (ME) motion too weak to distinguish. This hurdle can be tackled by enhancing intensity for a more accurate acquisition of movements. However, existing magnification strategies tend use features facial images that include not only clues as features, leading representation deficient credibility. In addition, variation over time, which crucial encoding movements, also neglected. To this end, we provide reliable scheme extract while considering their on time scale. First, devise an Intensity Distillation (ID) loss acquire contrasting difference between frames, given in same video lies intensity. Then, are calibrated follow trend original video. Specifically, due lack truth annotation video, build tendency setting each vacancy uncertain value, guides extracted converge towards rather some fixed values. A Wilcoxon rank sum test (Wrst) method enforced implement calibration. Experimental results three public ME databases i.e. CASME II, SAMM, and SMIC-HS validate superiority against state-of-the-art methods.