Have you ever heard of the site What-Dog.net? It’s a website created by Microsoft that uses object recognition to attempt to determine a dog’s breed from a picture. I wrote a review for What-Dog.net way back in 2016. Since it’s been so long, I thought it might be fun to take another look at the site’s dog breed recognition capabilities. Has it improved any in the past 6 years? This post will be an update of my 2016 review!
The first thing I noticed is that the look, and display URL, of the website has changed. What-Dog.net still directs to the dog breed recognition site, so that’s good. It’s hard for me to remember exactly what the site looked like 6 years ago, outside of the screenshots I have saved, but from what I can tell it looks a bit more sophisticated now. Change is good, right!?
I only have my own two dogs to test the software on, so take that as you will. A super small sample size is never a great thing. Oh well. First up we have my papillon, Kitsune. He is indeed a purebred papillon, but he doesn’t always look like one. He’s oversized and I often cut his hair, which isn’t typical for the breed. Back in 2016, I found that the breed recognition software recognized Kitsune as a papillon 75% of the time, seemingly depending on what picture I submitted. Here’s my old screen shot from when I was testing out pictures of Kit…
Testing the software with pictures of Kitsune now, it does seem to be more accurate! I didn’t try a ton of different pictures, but out of all the ones I tried, there was only one where it listed two breeds instead of one – King Charles spaniel, and papillon. I even dug up that old picture of Kit in a bow tie that I tested back in 2016. Back then, using that picture, the software thought Kitsune was a Tibetan Spaniel. With that same photo now, it guessed papillon!
Fen was always going to be the real test, though. In 2016, the Microsoft software didn’t guess his breed correctly a single time. My theory was that maybe it didn’t have Alaskan Klee Kai in its breed database. In 2016, depending on what photo I used, it seemed to think Fenrir was either a chihuahua or a Siberian husky.
Will it do better in 2022?…..
Not really. In all the newer photos I tried of Fen it mostly guessed him to be a Siberian Husky or a Canadian Eskimo dog. One time it guessed him to be a malamute. Based mostly off of Fen’s colors, it does make sense to guess those breeds. But Fen is only 15 lbs, a fact I guess the software can’t really determine from photos.
The Microsoft dog breed recognition software does seem to be a bit better at determining when there is no dog in a photo. Back in 2016 it seemed to recognize one of my 2lb house rabbits, Barnaby (RIP), as a chow chow!
This time I tried photos of cats, to see what dog breeds they most closely resembled, but the software just displayed a message saying it couldn’t find any results. I’m not sure whether that was because it couldn’t detect a dog in the photo, or it was just a coincidence. Because strangely, there were a couple of photos of Fen that I received the same error message with.
Overall I’d say that the Microsoft dog breed recognition software does seem to have improved somewhat. Still, I feel like programs that predict your pet’s breed from a photo are more of a novelty than something you should 100% rely on. I wouldn’t base my dog’s breed solely on the results from this site, although I think the results can be interesting. It’s also a fun website to play around with, if you have a bit of free time.
Have you ever used the What-Dog website or a similar photo app? How accurate was it for your pet?