MPEblink2.0

Real-time eyeblink detection in unconstrained scenarios can widely serve for applications including fatigue detection, face anti-spoofing, emotion analysis, etc. Existing research efforts mainly focus on single-person cases in trimmed video. However, detecting eyeblinks in multi-person scenarios within untrimmed videos is also crucial for real-world applications but has not been well addressed yet, with key challenges including consistent multi-instance awareness and precise instance-level eyeblink detection in long videos. To fill this gap, we construct a large-scale dataset termed MPEblink that involves 891 untrimmed videos with 12,456 eyeblink events is proposed under multi-person conditions. Details and other contributions on methodology and experiments can be found in our research paper.