Custom yolov8n model evaluation

General Discussion
1

Hi everyone,

I’m working with a custom YOLOv8n model trained on my own dataset, and I’d like to evaluate its accuracy and performance both before and after optimization without the hailo hardware.

I’m not sure whether I should use the DFC pipeline (SDK Native / SDK FP Optimized / SDK Quantized) for this evaluation, or instead use the Model Zoo tools, specifically hailomz eval with my exported .har file and a custom calibration TFRecord or image directory.

Could someone clarify the recommended workflow for this use case?

  • Should evaluation before and after optimization be done entirely through the DFC pipeline, or can hailomz eval handle both stages?

  • Are there any specific arguments I should include when using hailomz eval with a custom HAR and calibration dataset?

Thanks in advance for your help!

2

Welcome to the Hailo Community!

I recommend doing that. Then you can use your newly acquired skills for any model not just the ones in the Model Zoo.