This repo supports face mask detection utilizing YoloV5. Ideally this repo could be deprecated and the face mask detection model could be integrated to be supported by the vsdkx-model-yolo-torch
repo.
'conf_thresh': 0.5, # Float class confidence threshold
'iou_thresh': 0.4 # Float Intersection of Union threshold
The model is currently configured to support a custom trained facemask detection on YOLOv5 as TFlite. The model configuration can be found in config/models.yaml
.
model_path: 'wight.tflite'
input_shape: [320, 320]
classes_len: 2
anchors: [ [ 5.98438, 7.60156, 12.57031, 15.35938, 21.73438, 27.95312 ],
[ 32.90625, 41.90625, 47.37500, 61.37500, 70.06250, 90.50000 ],
[ 94.75000, 123.12500, 125.25000, 161.12500, 192.00000, 198.25000 ] ]
anchor_len: 3
max_width_height: 4096
mask_on: 1
mask_off: 0
mask_on_threshold: 0.60