Inspiration
We were first inspired to do a profile application using deep learning where any face detected on a camera could be then extracted and by which their age and gender would be extracted and displayed which is really helpful in dating apps or in a crime detection to profile the images
What it does
An application which recognizes gender using the extracted face which is obtained by cascading neural networks and the output of the same is fed into our custom neural network model to get the prediction
How I built it
We use a neural network to extract both the faces and to get a prediction. For extracting the faces we use cascading CNN which we referred through a research paper which led us to a github link https://github.com/ipazc/mtcnn and also we import a pretrained model VGG16 with trained weights called VGGFaces since these weights were trained on a millions of faces hence we get state of the art results
Challenges I ran into
We tried to implement our own custom trained network for predicting the gender however since our model weights were trained on scratch and with just training on 10000 images would really make the weights accurate to be able to predict any given general image, We also tried to predict the age which gave us poor results maybe due to large number of labels and the images were close to identical between a single age group , for example images in age 15 and 16 were very much identical to each other, hence the model became too complicated to implement
Accomplishments that I'm proud of
We made a working real time model which predicts the gender of an image with 95-96% accuracy
What I learned
i learnt a vast deal about learning convolution networks and how flask also works in setting up web pages
What's next for Gender Recognition through Faces using Deep learning
The above model could be improved further by the recognizing the age of a person through deep learning model and can also be used in dating and profiling applications
Log in or sign up for Devpost to join the conversation.