YOLOv6

The latest in the class of single shot object detection models

You Only Look Once (YOLO) was first introduced in 2015 with the idea of doing inference to perform object detection on a picture/video with a single neural network. YOLO picked up steam with YOLOv3 released in 2018. YOLOv5 introduced in 2020 was based of PyTorch, which made it extremely easy to consume and play with. We have previously researched into many aspects of the YOLO family of models, such as this article which evaluates which model could be best suited to your use case.

YOLO v6

The latest model in this family, YOLOv6 was released by Meituan Research 2 weeks back. Every generation of YOLO models aims to improve the speed of inference, while maintaining the same accuracy. YOLOv6 is no different and claims a best case inference speed of 1242 frames a second. However, these metrics aren’t sufficient by themselves to determine if YOLOv6 is the best model to use for a specific use case, as a lot of performance attributes depend on the training dataset, hardware available, and acc…