Showing results for 
Show  only  | Search instead for 
Did you mean: 
Registered: ‎05-31-2019

In CHaiDNN, XportDNN calibration_directory

I wanna implement yolov2 with CHaiDNN.


When I wanna do quantization for the weights and parameters with xilinx mode,

$ python XportDNN.pyc --quant_type "Xilinx" \
--deploy_model ./models/bvlc_googlenet_without_lrn/bvlc_googlenet_without_lrn_deploy.prototxt \
--weights ./models/bvlc_googlenet_without_lrn/bvlc_googlenet_without_lrn.caffemodel \
--quantized_deploy_model ./models/bvlc_googlenet_without_lrn/RistrettoDemo/bvlc_googlenet_without_lrn_quantized_deploy.prototxt \
--calibration_directory ./data/ilsvrc12/ILSVRC2012_img_val --calibration_size 32 \
--bitwidths 6,6,6 --dims 3,224,224 --transpose 2,0,1 \
--channel_swap 2,1,0 --raw_scale 255.0 \
--mean_value 104,117,123 --input_scale 1.0

i dont know should be in the --calibration_directory. Imagenet dataset is different from the dataset which can be used for yolov2.

In theory , there should be  images and labels. Yolov2 is for detection network and you have to draw the boxes of the objects. In darknet(the yolo original framework), i use VOCdekit to do the training and validation. But, for caffemodel, i dont know the image format for calibration.

Thanks for anyone replys.

0 Kudos
0 Replies