It’s Monday morning and I’m channeling my inner Jeremy Howard (again). Over the last few months, I have seen plenty of facial recognition guides and some of them are really quite good.

The problem is, each student of deep learning or machine vision becomes interested through different venues. Maybe you’re a musician interested in sounds, or a statistics student interested in tabular data.

I know I personally became interested when started reading about TensorFlow and image classification — this idea that if I have a well curated dataset, I might be able to train an architecture to classify images with high accuracy.

Google Colab didn’t exist then and I recently ported something similar over to Colab and I’m in the process of a walkthrough.

My initial foray was done on my iMac using Anaconda, Spyder, no GPU usage, just letting my Core i5 trudge through thousands of images while I left to do other things (read: study for emergency medicine boards). Of course, before I could train, I had to read through two dozen articles of people using different styles to write code. What resulted was a chimeric hodge-podge of code that was difficult to understand when I looked at it weeks later.

Then Colab entered the scene and while I still use Jupyter and Spyder on my machine, I use Colab to speak to you and efficiently share my code. It’s not perfect but they’ve done a great job in making high end computing available to the masses and all you need is a Google account.

#computer-vision #haar-cascades #face-recognition #python #opencv