Why I didn’t use deep learning for my image recognition problem
This is a how-things-went-wrong story of using deep learning to match images of climbing holds in gyms. But there’s a happy ending: a more traditional computer vision solution outperforms neural nets.
I am data scientist coming from a statistical background, always on the lookout for problems to solve using data.
By day, I work at GoCardless, a recurring payments provider, where I create machine learning products to reduce payment failure rates and tackle fraud.
My second life is rock climbing and working on Climbicus - an app for logging climbing routes at an indoor gym, leveraging computer vision techniques.