YOLOvME: The Mysterious Case of Dog Poop

Next time you let your dog poop in someone’s yard, you may want to think twice. Ben Farahmand realized he had this exact problem when his daughter started alerting him of the influx of dog poop that had started appearing in his yard. Unsure of who was behind this, Ben started brainstorming how he could catch the culprit red-handed.

We needed to know how the story ended, so we sat down with Ben and asked him a few questions about his experience with YOLOv5 and vision AI.

What Was Your Process Of Creating Your Object Detection Tool Like?

Identifying the Challenge

“I saw a problem. I was like, okay, my daughter plays in the yard, and she notices the dog poop in the yard. I wanted to be able to talk to the person that was letting the dog poop and not picking it up.“, says Ben.Ben needed to find a better solution than staring out his window 24/7.

Choosing The Right ML Model

So, after a quick Google search, Ben discovered YOLOv5 and got to work training his model.“I came across a couple of different available models. YOLOv5 seemed to have good documentation, so I just started with that one.”Ben then turned to Twitter and documented the information from his stakeout, keeping us on the edge of our seats.

Training the ML Model

In the beginning, Ben encountered a couple of obstacles. Ben’s model initially had trouble detecting small dogs, who were still considered suspects in the case. Sometimes it would classify small dogs as birds, and other times it would miss them completely. In a specific instance, a big, fluffy white dog was classified as a sheep.To remedy this problem, Ben swapped out the small neural net he had been using for a larger one – finding success here. Small dogs were no longer identified as birds, and the former sheep was finally correctly identified as a dog.

Did You Ever Catch the Culprit?

“I did and what ended up happening is that often the owners would walk by my yard. Sometimes the father would walk by, other times the daughter. When I looked at my data, I actually found that the father would always pick up the dog poop, but the daughter would not. “It seems that things have improved in Ben’s neighborhood. “We wrote a letter and spoke to them and so far there hasn't been any more dog poop. But, it now feels a little creepy because sometimes I go outside and I recognize these dogs walking past. I have to wonder if they know that I watch them.“

How Did You Get Started With Ultralytics and YOLOv5?

“Well, by accident, because I knew about OpenCV, I've kind of played around with it back in graduate school. But, my graduate program was more about design for social impact, without any kind of engineering focus. I just kind of tinkered with it.Recently, I knew it was possible to do what I wanted to, I just wasn't 100% sure if I could do it. I also didn't realize how easy it was. Kudos to you guys for having excellent documentation. Whoever wrote your documentation needs to get a raise.”

What Other Challenges Would You Like to Solve With YOLOv5?

“At the moment, the first thing that comes to mind is that we have a stop sign in our neighborhood that people run through all the time. Maybe someone else has already created an application like this, but I would like to find out who continues to run the stop sign. However, I can’t speak to how legal it would be for me to set up something like this.“

‍

‍Ben Farahmand is a product manager for a small education company called Tuva Labs, which catalyzes data literacy education and empowers teachers to bring real-world learning into math and science instruction. Tuva Labs supplies datasets, lessons, and graphing software allowing for the incorporation of any kind of statistics lesson into their curriculums. Ben also founded FAZA, a nail-biting cooperative board game with a pulp sci-fi theme retelling the classic extraterrestrial invasion story.

‍

‍Check out Ben’s Twitter thread, documenting the dog poop narrative.‍

‍Find Ben’s GitHub repo here.