- https://www.kaggle.com/andrewmvd/face-mask-detection
- https://www.kaggle.com/wobotintelligence/face-mask-detection-dataset
- with_mask, without_mask, mask_weared_incorrect are the 3 class labels.
- SSD Single Shot MultiBox Detector - https://arxiv.org/abs/1512.02325
- Faste Rcnn - https://arxiv.org/abs/1506.01497
- Pytorch
- The Main motivation in developing Facemask Detection is to deploy model in mobile Device and Heroku therefor i need to reduce the detection latency low as possible while keeping decent MAP. Therefor i have given more emphasis on SSD Model with MobileNetV3 large as backbone.
- I trained pytorch pretraind ssd_mobilenet_v3_large object detection model with heavy augmentation, given 0.38 MAP. The Model was trained for above 100 epochs.
- The pretrained model separately trained with Mosaic Augmentation but i have trained for 40 EPOCHS and got 0.39MAP. Model performance can be further improved by for more epochs.
- Model is deployed to Heroku using Flask.
- Using REST API i have developed an android application , where user can capture picture from camera and can detect mask.
- Another android application is developed where The Pytorch scripted model is deployed on device.
- Check out the URLS below for related repos.
- Android Application using Heroku Rest API - https://github.com/green93/FaceMaskDetectionHerokuApiMVP
- Android Application using pytorch scripted model - https://github.com/green93/FaceMaskDetectionAndroidPytorch
- Heroku API Repo - https://github.com/green93/facemask-detection-heroku
- Rest API - https://facemask-detection-api.herokuapp.com/predict