YOLO11-EMME v2.0: A 10,000-image Eye-Tracking Dataset for Archaeology, Ophthalmology and HCI

Data-only rebuild from three YOLO-format ZIPs (YOLO11-IRIS EMME-v2). We provide consolidated CSVs rebuilt directly from the dataset labels, synchronized with the manuscript and the GitHub repository. We rebuilt data-only tables from two YOLO-format archives (train=8,760; val=2,497; test=770; Total=12,027 images; 75,904 objects). Per-class metrics are intentionally NA, and figures are omitted to avoid confusion; they will be regenerated once validation outputs are available. Files (this version):- results/results_summary.csv- results/_ALL/dataset_summary.csv- results/_ALL/class_distribution.csv- results/_ALL/bbox_stats.csv- results/_ALL/integrity_checks.csv What’s in the tables:- results_summary.csv — overall metrics only (fixed in this release): P = 0.890, R = 0.830, F1 = 0.859, AP@0.5 = 0.995. Per-class = NA (no prediction artifacts are used in this release).- dataset_summary.csv — per-split counts of images, label files, and objects.- class_distribution.csv — object counts by class (iris, pupil) and split.- bbox_stats.csv — bounding-box geometry (normalized area mean/median, aspect ratio mean, small/medium/large ratios; thresholds: S < 0.02, M 0.02–0.10, L > 0.10).- integrity_checks.csv — label QC counters (missing_label_files, empty_label_files, malformed_rows, out_of_bounds_boxes). Release policy (to avoid confusion):- No model predictions, weights, or figures are included in this version. - PR curve and confusion matrices will be added in a later version once validation outputs (e.g., runs/detect/val*/metrics.csv) or paired predictions with ground truth are available. Reproducibility:- Baseline settings: seed=42, imgsz=640.- Source: three YOLO-format ZIPs with train/val/test splits under images/ and labels/.- The same tables are available in the linked GitHub repository under results/ (with per-ZIP tables and an _ALL/ aggregate folder). License:- CC BY 4.0 (same as the project’s public release). Please cite the dataset and the Zenodo DOI.