Blink Detection Model Outputs from Sleep-EDF and GazeBase Experiments

Processed research outputs from a cross-sensor blink detection study using publicly available Sleep-EDF and GazeBase v2.0 datasets. The dataset includes trained CNN-LSTM and attention-based CNN-LSTM model weights, evaluation metrics, and experimental results. It was deposited by Koushik Chowdhury on Harvard Dataverse in June 2026.

Use Cases

• Benchmarking blink detection models based on CNN-LSTM and attention-based architectures.
• Analyzing cross-sensor transfer performance between electrooculography (EOG) and optical eye-tracking signals.
• Conducting few-shot adaptation experiments for model generalization.
• Visualizing model behavior using generated training curves, confusion matrices, and UMAP plots.
• Reproducing results from leave-one-subject-out validation and bootstrap confidence interval analyses.

Strengths

• Includes trained model weights for CNN-LSTM and attention-based CNN-LSTM architectures.
• Contains outputs for multiple experimental setups: cross-sensor transfer, few-shot adaptation, and leave-one-subject-out validation.
• Provides generated visualizations such as training curves, confusion matrices, and UMAP feature-space plots.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Derived from the publicly available Sleep-EDF and GazeBase v2.0 datasets.

Collection Method

Processed outputs from a machine learning study evaluating blink detection models.

Freshness

Last updated 2026-06-03 09:33:51; freshness should be verified.