There is not a lot of good "sunglasses" detection model, so I created one with the least amount of effort possible using yolov5
Use an existing yolov3 model to detect eyeglasses
Use normal glass data to create yolo labels with normal glass
Use sunglasses data to create yolo labels with sunglasses
Use those 2 labels to train a new yolo model to detect normal glass and sunglasses
This model detect any kind of eyeglasses
https://github.com/Leotiv-Vibs/sunglasses_detection
Use preprocess_meglass_data.py to get normal glasses data
https://github.com/cleardusk/MeGlass
https://github.com/shreyas0906/Selfies-with-sunglasses
Run glass_detection.py to create normal glasses labels and sunglasses labels dataset
Train new dataset with eyeglass_data.yaml file using yolov5
Using only 3 epochs already give acceptable result
python train.py --img 120 --batch 16 --epochs 3 --data eyeglasses_data.yaml --weights yolov5s.pt
Check /results for more
