Optimized YOLOv5 model trained on the PWMFD medical masks dataset using transfer learning from COCO with frozen backbone, data augmentations such as mosaic, and an input image size of 320 x 320.

Architecture: here

AP:

• Evaluation from pycocotools: 67%
• Evaluation from yolov5 val.py script: 71%

fps:

• Nvidia Geforce GTX960, 4 GB: 69 fps