cppe5-ppe-detection-fasterrcnn-mobilenetv3

CPPE-5 PPE Detection — Faster R-CNN MobileNetV3

This repository contains a TorchVision Faster R-CNN model fine-tuned on rishitdagli/cppe-5.

Architecture

• Faster R-CNN
• MobileNetV3-Large 320-FPN backbone
• 5 foreground classes plus background

Classes

• 1: Coverall
• 2: Face_Shield
• 3: Gloves
• 4: Goggles
• 5: Mask

Bounding boxes

Predictions use absolute pixel coordinates in [x_min, y_min, x_max, y_max] format.

Evaluation

{
  "map": 0.2159441113471985,
  "map_50": 0.4564204812049866,
  "map_75": 0.16165156662464142,
  "map_small": 0.04356435686349869,
  "map_medium": 0.10920310020446777,
  "map_large": 0.2874097526073456,
  "mar_1": 0.20925933122634888,
  "mar_10": 0.3011750280857086,
  "mar_100": 0.3011750280857086
}

The official CPPE-5 test split contains only 29 images, so interpret per-class and overall metrics with appropriate caution.

This checkpoint was produced in notebook run mode quick. The source dataset card currently reports its license as unknown; users should review the dataset terms before redistribution or commercial use.