Real-time Object Recognition

A shootout with a high-performance machine vision camera and YOLOv4

Hello again! We've been quiet on the blogging front lately since things have been quite busy. But in the background, we've continued to create tons of great demos for interested customers and investors. Today, we wanted to share with you one of our favorites.

One of the biggest challenges with our previous demos was that they were "in the wild." They showed great results, but we had a hard time answering questions like "what if we had less light?" or "what if the objects were moving faster?"

So, our engineering team spent some time assembling a controllable scene with a pottery wheel and dimmable lights. With this setup, we were able to change speed and motion as needed to demonstrate performance. On top of that, we implemented an off-the-shelf object recognition algorithm (YOLOv4) to rule out subjective human judgments (such as "that one looks a bit sharper, but that one is less noisy... I'm not sure which one I prefer").

Check out our results below!

If you’d like to know more, please reach out to us here or inquire directly about an evaluation kit!