Custom Object Detection ACAP using YOLOv5

Update: DetectX 5.x now uses YOLOv8 and cannot load YOLOv5 models. For new training and deployments, see Run Your YOLOv8 Models on Axis Cameras. This article and its video describe the legacy YOLOv5 workflow; the last YOLOv5 release, 4.1.1, is no longer maintained.

While Axis cameras offer robust built-in object detection analytics for common use cases, some scenarios require more specialized detection. This package allows you to leverage a trained YOLOv5 model on the camera itself, bypassing the need for server-based processing. If you have a labeled dataset, you can train a YOLOv5 model, export it and create an ACAP to run it in the camera.

Please watch the video to understand the process

Legacy DetectX YOLOv5 release

There you will find the legacy instructions for training the model and building the ACAP. The current DetectX repository contains the updated YOLOv8 workflow.