adding more images for entry for calibration

Hello everyone,

I am converting a YOLOv8s ONNX model to HEF format using the Hailo Model Zoo.
I am using the following command:

hailomz compile yolov8s \
  --ckpt=cybest.onnx \
  --hw-arch hailo10h \
  --calib-path=train/images \
  --classes=2 \
  --performance

The folder train/images contains many images, but during compilation the tool seems to use only 64 images for calibration.

Could someone please explain:

  • Is the calibration image count limited to 64 by default?

  • Is there any way to increase the number of calibration images used during compilation?

  • If not, what is the recommended best practice for selecting calibration images?

Thank you in advance for your support.

Hi @user555 ,

Please let me know if that was helpful: https://hailo.ai/developer-zone/documentation/dataflow-compiler-v5-2-0/?sp_referrer=sdk%2Fmodel_optimization.html#calibration
specifically calibset_size.

Thanks,

1 Like

I have a small confusion regarding calibration. Currently, my training image folder contains 800 images. Could someone please guide me on how to include additional images for calibration? An example command would be greatly appreciated.

thanks for the support finally its work

1 Like