A project that develops a Mask Object detection deep neural network model.
For object detection model development, a Custom pre-trained Faster-RCNN architecture was adapted and fine-tuned to a custom face-mask dataset. Refer to the notebooks within this repository for all code and process(es) applied for this.
Some basic examples of applying the trained model to new inputs are shown below:
As shown, the model can adapt to a range of scenes, both simple and complex. It also works surprisingly well on complex scenes with high numbers of people, which make the final trained model useful for real applications.
Predictions on new sets of images stored in a directory can be made using the inference dashboard, developed in Dash. In real-time the model will make predictions for the selected image using the chosen confidence level as a threshold.
Some examples from the dashboard application are shown below for illustration.
