Gender-Detection

Dataset

• Images and Meta Data : https://data.vision.ee.ethz.ch/cvl/rrothe/imdb-wiki/static/wiki.tar.gz

• Cropped Images : https://data.vision.ee.ethz.ch/cvl/rrothe/imdb-wiki/static/wiki_crop.tar

Data Preprocessing

• Few of the Images were corrupted and hence had to be removed from the dataset.
• Face Detection Algorithm (Cascade Classifier) is used to check if an image is valid or not.
• If a face is detected in the Image, Collect the required details (Full path and Gender) of the Image.
• Store the required data in a different .csv file.

Framework/Library used

• Keras with Theano backend.
• OpenCV2 for Image Processing Algorithms.
• Pandas and NumPy for .csv and mathematical computations.

Keras

• Is a Deep Learning Library which supports Theano and TensorFlow at backend.
• Keras is easier to install compared to other frameworks and had a better Documentation.
• https://keras.io/

Data Model

• VGG-16 data model is used for training and testing.
• Such a data model contains a 16 layer network.
• Weights of the VGG16 Model used from : https://drive.google.com/file/d/0Bz7KyqmuGsilT0J5dmRCM0ROVHc/view?usp=sharing

Results and Outcome

• A live video captures an image if a face detected in the frame.
• The cropped image (face) is sent to the pipeline where the a prediction is made with already saved model which is loaded globally.
• An additional layer (Sigmoid) is added at the flatten layer of the model.
• The Prediction lies between 0-1.
• Values near to 0 signifies Female. Values near 1 signifies Male.
• A Cross Validation Accuracy of 90.92% is achieved on the Dataset trained.

Team Members

• Mohit Reddy (mohitreddy1996)
• Ambareesh Prakash (AeroDeath)
• Vilas M Bhat (vilas897)
• Aneesh Aithal (aneesh297)
• Somnath Sarkar (somnathsarkar)