I’ve implemented a YOLOv7 model for object detection within data center racks, integrating it into AR applications for both Android and iOS platforms. Initially, the model performed satisfactorily with up to 150 classes. However, upon expanding the training to include 500 classes, I observed a significant decline in performance, resulting in sluggish application behavior.
Additionally, I encountered an issue during the initial execution of the application, where it froze for several seconds, particularly when updating the model or launching it for the first time. This freeze is exclusive to these instances and resolves itself once the application resumes normal operation.
Furthermore, it’s worth noting that we intend to further increase the number of classes to over 3000 in the future.
I’m seeking advice on optimizing the YOLOv7 model’s performance, especially with an increasing number of classes, as well as addressing the freezing issue during the application’s initial execution on both Android and iOS platforms.
Any insights or recommendations would be immensely helpful. Thank you for your assistance!