In this session we will provide an introduction to TensorFlow.js. We will then use a step by step walkthrough for building a fully functional real time face detection using nothing but a webcam, a browser, Tensorflow.js and face-api.js.
Data Con LA 2019 - Real Time Face Detection through a Webcam Using AI by Obaid Sarvana
1. ACCENTURE
APPLIED
INTELLIGENCE
Accenture’s unique approach to combiningAI with data,
analytics and automation under a bold strategic vision
to transform your business—not in silos, but across
every function and every process, at scale.
2. AGENDA
• Introductions
• Brief background on Machine Learning &Tensorflow
• Edge Learning
• Tensorflow & FaceAPI
• Code Demo
• Come connect with us
3. WHAT AREWETALKINGABOUT?
In this session we will provide an introduction toTensorFlow.js.We will then use
a step by step walkthrough for building a fully functional real time face detection
using nothing but a webcam, a browser,Tensorflow.js and face-api.js.
Want to learn more aboutTensorflow?Check out Google’sCrash course!
https://developers.google.com/machine-learning/crash-course/
4. REACH OUT TO CONNECT!
WHO ARE WE?
OBAID SARVANA
obaid.sarvana@accenture.com
JACOB REDDING
Jacob.redding@accenture.com
Data Architects, Software developers
Latest Projects
• J: Open Source strategies/Governance
• O: Software Innovation, Connected Health,
Innovation Strategy
5. HOW DIDWE GET HERE?
Jungle > Candy Store
AI Packaged goods
Learning on the edge!
9. MAKES IT EASY…EASIER….ISH
The packages make it “Easier”, but, of course, that is relative.
Let’s dig into the code and we can discuss the current state of software and where we are and where this is
going.
Machine Learning – not a Jungle, a candy shop
Feels dangerous, scary
Still have math PhDs and superexperts, but their findings and insights are now getting translated to more accessible tools
Actually – getting easier and easier to access
This shift is driven by 2 phenomena:
(funded) AI summer – Google, Facebook, etc. whose bottom lines are helped by AI, are funding investments into AI (pyTorch, Tensorflow, etc)
API revolution – developers, and technologists exposing an abstraction of their knowledge/tools to 1) protect their IP and 2) Allow for easy use/access their tool in a way that lets’ them hit the groud running
Leading to high quality AI tools, which can be accessed through APIs requiring minimal knowledge of AI Math etc
Democratization of AI
As these packages mature, they are also getting ported to other languages, and in the process becoming more and more lightweight.
Allowing for the reverse of the shift to cloud. We’ve heard about the rise of edge computing, but now we’re seeing the enablement of edge learning. Smart things that don’t need to rely on master control - browsers, phones, raspberry Pis that can learn and formulate their own theories of the world
What was the paradigm shift that led to the current explosion
Brief historical reference (pre-tensor learning and post-tensor)
2 phenoms – Tensorflow (Keras, etc.) & Edge Learning
Software packages do the heavy lifting so you can focus on the experimentation
Today – Edge learning (edge compute)
Code Demo