ThuraTunScibotics/Detection-of-Face-Mask

Face mask detection of two classes with_mask and without_mask using Transfer Learning with MobileNetV2.

Detection-of-Face-Mask

Background - One of the tasks for Computer Vision internship at The Sparks Foundation.

Face Mask Detection project by classification of with_mask & without_mask classes is using TensorFlow, Keras and transfer learning with MobileNetV2 DNN architecture having weights of pre-trained on the imagenet. Training for the model and the dataset of this project is done on free GPU of GoogleColab notebook.

Algorithm

Phase:1 - Train Face Mask Detector

1. Load Face Mask Dataset
2. Train Face Mask Classifier with Keras/TensorFlow and using Transfer Learning with MobileNetV2 as baseModel
3. Serialize trained face mask classification model to disk

Phase:2 - Apply Face Mask Detector

1. Load face mask classification model from disk
2. Detect faces in image/videostream
3. Extract each face ROI
4. Apply face mask classifier to each face ROI to determine "mask" or "no mask"
5. Save/Show the result

Requirements

Check the package manager, conda which will be required to install required libraries & packages under specific virtual environment. Install anaconda on your machine, and run the following cell on terminal/command prompt after installed.

conda create -n FaceMaskDetector jupyter tensorflow keras python opencv imutils scipy numpy pandas matplotlib

Model Architecture

MobileNetV2 BaseModel -> Average Pooling -> Flatten -> Dense -> Dropout -> Dense

• Transfer Learning with MobileNetV2 with weights of pretrained on imagenet
• Average Pooling with pool_size=(7, 7)
• flatten into vector
• Dense layer with units=128 and activation='relu'
• Dropout 50%/0.5 of neurons
• Final Dense layer with units=2 and activation='softmax'

Compile and train the model

• learning_rate with 1e-4 & using adam optimizer & loss with binary_crossentropy & metric with ["accuracy"]
• Train with DataAugmentation for better accuracy & batch_size with 32 & epochs with 20

Classification Report & Train Accuracy & Loss Evaluation

Demo

References

https://towardsdatascience.com/my-quarantine-project-a-real-time-face-mask-detector-using-tensorflow-3e7c61a42c40

https://pyimagesearch.com/2020/05/04/covid-19-face-mask-detector-with-opencv-keras-tensorflow-and-deep-learning/