Data Leakage and Evaluation Issues in Micro-Expression Analysis

Micro-expressions have drawn increasing interest lately due to various potential applications. The task is, however, difficult as it incorporates many challenges from the fields of computer vision, machine learning and emotional sciences. Due spontaneous subtle characteristics micro-expressions, available training testing data are limited, which make evaluation complex. We show that leakage fragmented protocols issues among micro-expression literature. find fixing leaks can drastically reduce model performance, in some cases even making models perform similarly a random classifier. To this end, we go through common pitfalls, propose new standardized protocol using facial action units with over 2000 samples, provide an open source library implements manner. Code is publicly https://github.com/tvaranka/meb .