release-gate-records / run-b-production-gate /evaluation_metrics.json
jzloko's picture
Update run-b-production-gate/evaluation_metrics.json
7afc35b verified
{
"evaluator": {
"tool": "ultralytics/yolov5 val.py",
"yolov5_version_string": "YOLOv5 v7.0-460-g3fb11111",
"yolov5_git_commit": "3fb11111c6a8088fbc91430a1f99d207c16f0620",
"python": "3.13.5",
"torch": "2.10.0+cpu",
"device": "cpu"
},
"dataset_scan": {
"images": 7361,
"instances": 7459,
"backgrounds": 4704,
"corrupt": 0,
"notes": "YOLOv5 restored multiple corrupt JPEGs during scan."
},
"metrics": {
"precision": 0.776,
"recall": 0.68,
"map50": 0.747,
"map50_95": 0.365
},
"runtime": {
"preprocess_ms_per_image": 0.3,
"inference_ms_per_image": 35.0,
"nms_ms_per_image": 2.2,
"batch_size": 32,
"imgsz": 640
},
"artifacts": {
"yolov5_predictions_json": "receipt_bundle/raw/yolov5s_predictions.json",
"yolov5_runs_dir_source": "C:/loko_release_gate/yolov5/runs/val/exp4"
},
"notes": "Run B is a low-light operational surface derived from ExDark metadata.csv (People -> COCO class 0 person). COCO pycocotools evaluation is not applicable to this custom ground-truth set."
}