Using YOLOv3 for detection and custom 3 classes trained mobilenetv2 model from tensorflow for prediction.
OpenCV and streamlit was used to deploy using webcam.
The dataset includes images of team members, captured by webcam using data_creator.py.
Each image using YOLO to crop to take only the face area.
Dataset has been put in structural folders with each folder name as label (name of team member):
- long: 597 images
- minh: 669 images
- tung: 507 images
Link for the dataset: images_260921.zip
MobilenetV2 model was used as based model and customized classification dense layers to predict the team members.
The model was trained using Train_model.ipynb
The best model was exported to .h5 file for deployment: model_1.h5
Streamlit was used to deploy the file: face_recog_app.py
- Download YOLOv3 model and weight at: https://drive.google.com/drive/folders/1QO9ydq_cUHlfpKK78DSUymHii5aix2jf?usp=sharing
- Load YOLOv3 for face detection using weights and config files
- Load trained CNN model for face prediction
- Create bounding box around each face detected and show name with confidence level (in probability)
Contributions:
- Minh Tran: https://github.com/tranducminh1902
- Long Nguyen: https://github.com/longnguyentruong0607
- Vu Thanh Tung: https://github.com/tung151078
References: