How to run yolov8s 320x320 trained model in hef

General
1

I have converted yolov8s model with 320x320 imagesize and converted it onnx and then to HEF. When i used the default settings which is 640x640 i was able to see detection and now when i use the 320x320 im not able to see any detection when i pass a video file or through usb cam.

Im using the detection_pipeline from git

Is there any changes i need to make ?

my resources/gender.json file

{
“iou_threshold”: 0.6,
“detection_threshold”: 0.4,
“score_threshold”: 0.4,
“output_activation”: “none”,
“labels”: [
“unlabelled”,
“male”,
“female”
]
}

1 Like
2

Hi @sudhir,

Welcome to the Hailo Community!

For your use case, I recommend using detection.py instead of detection_pipeline.py. While detection_pipeline.py handles the basic pipeline and functionality setup, detection.py is our main application that allows you to run custom HEF files.

Note that in our upcoming update, we’ll be separating these into distinct infrastructure and examples repositories. For now, please proceed with detection.py for the best experience.

Let us know if you need any further assistance!

Best Regards,
Omria

1 Like