A cleaned binary classification dataset for driver drowsiness detection, last updated on Hugging Face in November 2025. It contains image crops of faces with 'drowsy' and 'not_drowsy' labels, created by unifying labels from four source datasets and removing eye-only crops using MediaPipe face-mesh. The dataset is structured into train, validation, and test splits.
Use Cases
- Train a binary classifier to detect drowsy drivers based on facial image crops.
- Benchmark drowsiness detection models using the provided train, valid, and test splits.
- Fine-tune a face mesh model for fatigue recognition based on the cleaned and unified labels.
Strengths
- Labels are unified from four source datasets, suggesting consistency.
- Dataset is cleaned, with eye-only crops removed using MediaPipe face-mesh.
- Includes predefined train, validation, and test splits for model development.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Data may reflect source bias inherent to the four original datasets.
Provenance
- Source
- Hugging Face user 'n7i5x9'.
- Collection Method
- Labels unified from four source datasets; eye-only crops removed with MediaPipe face-mesh.
- Time Range
- null
- Freshness
- Last updated 2025-11-09 12:33:41; freshness should be verified.
- Geography
- null