Month: October 2019 (page 4 of 4)

Recently Favorited Tweets

When you ask the Snook for a picnic, he doesn’t disappoint. ❤️


via Instagram

Current mood.


via Instagram

Recently Favorited Tweets

Recently Favorited Tweets

Recently Favorited Tweets

Recently Favorited Tweets

Recently Favorited Tweets

Knitted Disruption

Hey knitters! I’m working on a knitting + machine learning project and I need a collection of images of stockinette, garter, seed, and moss stitches. Images like these:

If you’re willing to spend a few minutes helping, I’d be so grateful! Just email your images to knittingml@krishoward.org. You can attach multiple images at once if you like.

They don’t have to be swatches; they can be closeups from finished articles. They don’t have to be perfectly straight or blocked or anything like that either. I’m looking for a wide variety, to be honest! Stripes and multiple colours are great! Even fairisle. I want to teach the model to disregard colour, so having photos with it is very helpful. Just no lace or cables or crochet… (yet).

More details

A few years back I gave a talk at several tech conferences about the overlap between knitting patterns and programming languages. As part of that, I talked a bit about KnitML (an old proposed standard for writing patterns in ways computers can understand) and how it could be used with special software to simulate knitted fabric. A few people asked me afterwards if it could go the other way – from a photo of knitting, can you reverse engineer the pattern?

It got me thinking. I know that I can “read” knitting. I do it all the time, and I’m sure other knitters do too. So if I can do it, why not a computer? I’m also fortunate in that I work with some very smart Machine Learning experts and have access to run experiments in the Cloud cheaply. So I decided to give it a try.

I’m starting with image classification. Think facial recognition for knitting swatches. Can I train a model to recognise the difference between stockinette, garter, moss, and seed stitch? The first step is gathering as much training data as I can, hence my request to you all!

So the more the better. Different wools, different colours, different lighting, different angles. It’s all super useful!

And I will definitely share the results back afterwards and thank all contributors. 🙂