Driver Drowsiness Detection Dataset

Detecting driver drowsiness from in-cabin cameras requires extreme precision in eye state classification—something that's nearly impossible to label accurately with real-world data. This synthetic dataset provides pixel-perfect eye keypoint annotations generated by computer simulation, guaranteeing ground-truth accuracy that manual labeling cannot achieve. The dataset covers a wide demographic range of synthetic drivers with varied skin tones, facial features, and accessories (glasses, sunglasses, hats). Lighting conditions include daytime, nighttime, tunnel transitions, and oncoming headlight glare. Eye states are annotated across a spectrum from fully open to fully closed, with intermediate drowsy states. The open-source sample includes 168 images; the full package contains 10,000 images with 20 validation videos showing real-time drowsiness detection performance.