We, humans, are extremely good at noticing fundamental and sophisticated things such as beauty from people's face. Nevertheless, we could seamlessly read emotions, feelings and even personality in our faces however diverse we are. Its quite clear that most of our personality and current situations, even our current health status is just written somewhere between our forehead and chin and this is truly backed by research.

Smart computers, powered by Artificial intelligence have recently surpassed average human visual capabilities. This is due to the availability of big data and advances in computing resources. Machines are now able to figure out patterns that us mere mortal beings could never get a clue of. Computer vision to be specific has been used by various industries including the pharmaceutical industry to diagnose diseases. A lot has been achieved, albeit, none has been used that exploits the facial landmarks for medical information and disease diagnosis.

What it does

We decided to exploit this conventional application of Artificial intelligence in facial recognition and facial landmark detection to achieve a Face Sickness Detection system. What you need is a picture of a victim, it might be a prospective patient or just another healthy being captured from the perfect frontal face area. The image is then uploaded to the model for inferencing after which it predicts the candidate's health status with a significant order of precision

How we built it

Deep metric learning has been an approach taken time and again for measuring the distance between two things by applying some discriminative loss function to the deep learning model. Most precisely, most researchers have always preferred this approach and applied it t in challenging environments and applications such as face recognition. It has traditionally been used to embed facial images to some vector space such that faces of the same person appear to be closer, contrary, different people have a wide margin from each other. Consequently, a classifier is then used to determine and represent each embedding to its corresponding class label, which in this case -the face ID.

Using this sophisticated approach, we decided not to train a model from scratch since their exits pretrained opensource projects that utilize the same. Facebook Research released DeepFace in 2015, Google Research also released Facenet in the same year among others. However, these models are not designed with Face Sickness Detection in mind, they provide the embedding required, it's then upon our smart classifier after training to dissimilate the image candidate and demystify the candidate's health status. We use Pytorch(FaceBook library for Deep Learning) and Django (Python web framework ) to make this a reality

Challenges we ran into

Sincerely, we didn't run into lots of challenges since most of the algorithms we used were off the shelf with extensive documentation.

Accomplishments that we're proud of

We are the first group, that I know, to develop a face diagnostic tool for doctors and many others and with this project, we could improve human life and diagnose diseases at an early-stage so people could be watched keenly before diseases worsen and lives could be saved.

What we learned

Its always been a norm on the society that every emerging technology is meant to improve lives and possibly impact every industry. Did we have to earn an MBA in medicine to do this? No. We had to learn. Understand the underlying principles of human psychology and basic facial anatomy. We had to understand how the face responds to any ailment. Then we came in with what we do best, technology

What's next for Face Sickness Detection

We are aiming to improve our model to remarkable precision. Moreover, want to train a custom deep metrics learning algorithm with Face Sickness Detection in mind. The dataset needed will be heavily influenced by that. We are going to a level where we will expound on the problem and even get into the details of the infection if someone is considered sick. We need to study. We need to consult. Clearly, we need help

Share this project: