Micro-expression (ME) is a rapid, subtle and spontaneous motion of the human face that usually lasts between 1/25 and 1/5 seconds. Unlike macro-expression, which could be misleading on human emotion recognition, micro-expression is mostly expressed unconsciously where genuine emotion can be revealed. As indicators of emotional states, interpreting people's micro-expression correctlycan boost many practical applications around our daily life.
In the academic community, there have been several competitions held by international experts trying to solve the problem of micro-expression recognition, such as OMG2018, MEGC2018, MEGC2019, etc., but it is far from being solved due to insufficient annotation of the datasets. More samples are needed for robust automatic micro-expression recognition research and related applications. Also, the class imbalanced in the current database is a tough challenge for the recognition task. A dataset with a large size of even samples among classes is essential for a highlevel approach.
Considering the difficulties of fine-labeled ME dataset, we propose a novel synthetic ME dataset based on current manual annotated ME datasets, i.e., CASME II and SAMM, which contains over 10,000 sequences with balanced classes. We sincerely welcome the latest efforts and research advances from the scientific community to join this challenge in tackling the micro-expression recognition challenge on our published dataset!