My Intro to CV

I was lucky to have been awarded a grant to attend AWS re:Invent 2019. If you’re unaware, re:Invent is a behemoth of a conference in Las Vegas where AWS shows off all its new features, has keynote speakers, and offers lectures on how to use their products. It’s such a big event that AWS takes over all the casino conference areas and even nightclubs!

DeepLens

The first AWS product to catch my eye at the conference was the AWS DeepLens. It’s a great little device to have for a beginner CV project. It has an Intel processor, 8GBs of RAM and it has an Ubuntu OS.

During the lecture, some AWS employees walked us through a jupyter notebook in AWS SageMaker on how they gathered data, labels, and then trained a model to classify if a polar bear is in the frame or not.

After the lecture they were giving out coupons for the device so I picked one up once I got back home. This was my first official CV project that wasn’t just an MNIST classification project.

My DeepLens Project

My project was basically a CV replacement to badging in and out of an office. Instead of a scanner, the idea was to use the DeepLens to detect badges as people walk in a door.

Looking back, I definitely would have done everything differently. First, I would just train the model locally, or in an EC2 instance instead of SageMaker. Second, I wouldn’t have used the backbone I did. I found some tutorial online that used MXNet not knowing any better. It was really convenient to store the model file in s3 and just point to it from the DeepLens product site on AWS.