Training different YOLOv5 models on 'Face Mask Detection' dataset for comparison. Paper
Pretrained weights for each model can be found in weights directory.
Detailed results for each model are located in results directory.
Jupyter notebook is written with Google Colab in mind, but can be used with Kaggle for bigger datasets (33h GPU usage time limit). Every Kaggle batch session will end in an irrelevant error caused by differences in working directory structures between Kaggle and Collab. Results still can be downloaded. Notebook utilizes ultralytics' YOLOv5 library.
Upload your dataset to Roboflow and export it in YOLOv5 PyTorch format with show download code enabled. Replace #API# and #PROJECTNAME# and you're ready to go.
Face Mask Detection dataset used for training is available here. Dataset consists of 1677 images.
