A Manager’s Guide to Starting a Computer Vision Program

Abstract: With every passing day, deep learning and computer vision are taking the world by storm. But what does this mean for your typical data science team? How do you know if you should be using computer vision in your work, and how do you get started?

This talk will provide a guide for data science leaders and managers who are thinking about dipping their toes into computer vision. It will help you think through a set of sequential questions that you’re likely to encounter along the way:

Should you use a fully out-of-the-box solution like AWS Rekognition or Google Cloud Vision?

Should you build your own custom models without writing code by making use of a service like Google AutoML Vision or Clarifai? When should you decide to code your own models instead?

If you code your own models, what infrastructure should you use? Do you need to invest in your own on-premise GPU’s or are cloud GPU’s sufficient? Or maybe you should you use a hosted service for training and deployment?

Keras, PyTorch,Tensorflow (Eager), MXNet, Gluon--it seems like a new framework emerges every month. What factors should you consider when you’re deciding on a language and framework?

Where do you find talent? Should you hire it or grow it?


This talk will draw heavily on my experiences at ShopRunner, an e-commerce network offering its members free two-day shipping, free returns, and seamless checkout at hundreds of retailers, where we’ve been building a computer vision program to turn tens of millions of product images into compelling consumer features across our website, mobile app, and browser extension.

Bio: Michelangelo D’Agostino is the senior director of data science at ShopRunner, leading a team that leverages e-commerce data from a network of over 140 retailers to build personalization products for ShopRunner's 6 million members. Prior to ShopRunner, he led the data science R&D team at Civis Analytics, a Chicago-based data science consulting and software startup that helps companies and nonprofits better utilize their data. Michelangelo was a senior analyst in digital analytics with the 2012 Obama re-election campaign, where he helped to optimize the campaign’s email fundraising juggernaut and analyzed social media data. He's also been a mentor with the Data Science for Social Good Fellowship. Michelangelo holds a PhD in particle astrophysics from UC Berkeley and got his start in analytics sifting through neutrino data from the IceCube experiment. He’s also written about science and technology for the Economist.

Open Data Science Conference