NVIDIA-AI-IOT / cuDLA-samples

YOLOv5 on Orin DLA
Other
167 stars 17 forks source link

For qdq translator, may exit by node scales do't match #9

Closed Bobo-y closed 8 months ago

Bobo-y commented 8 months ago

image

As above, when use qdq translator, does this affect model precision if i force delete this "assert".

liuanqi-libra7 commented 8 months ago

Delete this assert won't affect the accuracy. This check is used when the calibration method is using max, but you can also avoid this issue by setting '--calibration_type=hist'.

Bobo-y commented 8 months ago

Got it, thanks.