This Project Aims to develop face analytics models (CNN based models). Current project has models to estimate following information.
- Ethinicity - Detects ethinicity from face with four classes as [ Asian, Indian, Black, White]
- Gender- Detects human gender [Man , Woman]
- Emotion - Detects seven different facial emotions such as ['angry', 'disgust', 'fear', 'happy', 'sad','surprise', 'neutral']
- Download the race detection weights from https://drive.google.com/open?id=1KI_ih8li_zmK8GyaEizKmJT2k3EeQ6TH and place in "trained_models" folder ("Fweights" and "Mweights")
- To test the output on new image siply run -
"python Api.py test_image_path PATH_TO_SAVE_OUTPUT"
Api saves the the image at PATH_TO_SAVE_OUTPUT with bounding box drawn around the detected face with detected Ethinicity, Gender and Emotion
Simply clone the Repo. Project is built with python 2.7 and NOT tested with python 3
Some Test case results are shown below
Trained models are available in folder "trained_models". All models are trained with carefully designed CNN models. Ethinicity/ race detection model uses the transfer learning approach
Contact developer for training scripts.
-- Version 1.0 (Beta)
- Rupesh Wadibhasme
- Amit Nandi (co-author)