The keras model is created by training from scratch on around 2200 face images (~1100 for each class). Face region is cropped by applying face detection using cvlib on the images gathered from Kaggle. It acheived around 96% training accuracy and ~90% validation accuracy. (20% of the dataset is used for validation)
- numpy
- opencv-python
- tensorflow
- keras
- requests
- progressbar
- cvlib
We would first need to generate model by executing the train_model.ipynb file. Once the model is generated successfully,it is saved and then we can use detect_gender.ipynb to get realtime gender classification.