Is it a unicycle?

I posted this to the Unicycle Chat group on Facebook and wanted to post here in case it would interest anyone that doesn’t follow that group.

I wasn’t very surprised to find that Google’s Inception model was trained to recognize unicycles. I was curious to see when it was right and wrong so put together a little app to show Instragram posts tagged with “#unicycle” and the prediction from InceptionV3. I started gathering data a few days ago.

It doesn’t seem to recognize drawings very well although it did well recognizing a screenshot of Daphne from Scooby Doo riding a unicycle (on the 2017-12-13 page).

http://iiau.rtbg.io/

That’s really neat, great work! Actually, flipping through the pages and just seeing the pictures was fun by itself. I kept doing that after I’d used up my interest in the machine learning part. :slight_smile:

But I was impressed to see that it did very well most of the time, including one photo of a juggler where you can’t actually see any part of the unicycle. I also learned that Unipiper refrigerator magnets are a thing.

Thanks! I find it useful to find unicycle photos and people to follow on Instagram. Processing a relatively small set of photos each day takes 3-4 hours running time on Cloud Machine Learning Engine which seems a bit high to me so I want to see how long it takes just running on my laptop.

I took a brief look (fascinating!) and noticed the common denominator in the pictures from yesterday (12/15) seems to be a hashtag mentioning unicycle. Like a picture of a Nest thermostat. Obviously nothing to do with unicycling but the hashtag is there for whatever reason…

Well, it’s only looking at photos tagged with #unicycle for simplicity. If I was going to improve it I’d add #unicycles, #unicyclelife, #monocycle, and the Cyrillic transliteration. However, Instagram recently added a feature which lets you follow tags which is probably more useful for my purposes. It won’t let you filter out photos of self-balancing “unicycles” and thermostats like my page could but that’s a feature they might add down the line.

Long live the unipiper!