Using your own data to create a robust computer vision model

Following on from my previous post here, I wanted to see how feasible it would be to reliably detect and segment a Futoshiki puzzle grid from an image without using a clunky capture grid. It works surprisingly well even when trained on a tiny dataset!

Image for post
Image for post
An example model prediction (image by author)

So what is Semantic Segmentation?

Semantic Segmentation is a step up in complexity versus the more common computer vision tasks such as classification and object detection. The goal is to produce a pixel-level prediction for one or more classes. This prediction is referred to as an image ‘mask’. …

Image for post
Image for post

Futoshiki is a type of Japanese puzzle in the same vein as Sudoku — in so far as the aim is to completely fill a grid of boxes with numbers. The grid in this case is much smaller, typically 5 x 5 rather than 9 x 9, and includes additional inequality constraints between neighbouring numbers. A couple of years back my ideal Saturday morning routine would invariably include lazily solving one from the back of the Guardian. Simpler times.

Around this time I had also gotten into the world of data science, starting my first professional role. I knew solving…

Sam Watts

Data Scientist

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store