Quant_tool input wrong

VIM3
#1

Hey, I am running the demo: How To Use Tengine SDK
(How To Use Tengine SDK | Khadas Documentation)

However, for model quantization, there is a bug in setting

" -x 128,128,128 -y 128,128,128".

It should be as followed :

“-x 0,0,0 -y 255,255,255”

./quant_tool -f darknet -m ~/models/yolov3.weights -p ~/models/yolov3.cfg -o ~/models/yolov3.tmfile -a MINMAX -i /home/wudi/code/coco128/images/train2017/ -x 0,0,0 -y 255,255,255 -z 416,416,3 -c INTERNAL -t UINT8 -n 100

#2

@frank Please check.

#3

One more thing, when I converted other models, e.g., yolov4-tiny, there is a segmentation fault…
Is there a way to check other models

#4

@stevenwudi Both methods can be used.

Just need to do the corresponding processing in the post-processing process.

" -x 128,128,128 -y 128,128,128" or “-x 255,255,255 -y 0,0,0”.