Understanding State of the Art Facial Recognition
Saurabh Ghanekar (~saurabh29) |
Description:
Talk Summary :-
Recently, there is a boom in concept of face recognition system with the introduction of Face ID by Apple in their iPhone X mobile phones. This was also incorporated by OnePlus for their mobile phones too. The most notable use of this technology is at Baidu, an internet company, are using face recognition instead of ID cards to allow their employees to enter their offices. Another place where this technology is prominently seen is in auto photo and video tagging feature of Facebook.
In this talk we will build a Facial Recognition program using python library “face_recognition” and then we will dive deep in the behind the scenes action of this library and will try to build a One Shot Learning face recognition model using PyTorch. We will be implementing a Siamese neural network on AT&T Laboratories Cambridge dataset. We will also cover the basics of this neural network, triple loss function and and will discuss the reason for choosing this architecture. I will explain how the network models a relation between two images and relates them.
Outcome of this Talk :-
Attendees will be able to possess the power to implement state of the art Facial Recognition program in a few minutes. They will also get to know how facial recognition works when we have very small dataset. They will be able to make a state of the art One Shot Learning face recognition based on Siamese Network (the working force of face_recognition and implementation of Google’s FaceNet).
Agenda :-
- Introduction to Face Recognition [2 mins]
- Introduction of python library “face_recognition” [1 min]
- Building a face recognition program using “face_recognition” library (possible live demo of the output) [6 min]
- How “face_recognition” encodes faces [2 min]
- Introduction of Triplet Loss and Siamese Network and reason to choose one shot learning (which is used to encode faces) [5 min]
- Implementation of Siamese Network using PyTorch on AT&T Laboratories Cambridge dataset and its results [10 min]
- Q&A Session [3 min]
Prerequisites:
Basic Knowledge of Machine Learning and Neural Networks
Love for Python
Content URLs:
Will be updated soon!
Speaker Info:
Saurabh Ghanekar
Hi, a Computer Science sophomore whose research interests lie in Machine Learning and Artificial Intelligence , occasionally working on Virtual and Augmented Reality projects. I’m part of a QS award winning student-led multidisciplinary lab called Next Tech Lab where we research in Artificial Intelligence, Mixed Reality, Internet of Things, and Blockchain. I am also co-organiser of PyData Amaravati. I also regularly participate and give talks in paper-reading groups and meetups like PyData.
Speaker Links:
E-mail me at : ghanekarsaurabh8@gmail.com