DIT4FUN/kendryte-model-compiler

model conversion tool

Dependent

you need python3 tensorflow and pillow.

pip3 install tensorflow
pip3 install pillow

Convert

clone the model-compiler.
convert .pb file to k210

python3 model-compiler --pb_path <your pb file> --tensorboard-mode

check your inout tensor name and your output tensor name, check your input dataset image width and height.

python3 model-compiler --pb_path <pb file path> --tensor_output_name <output tensor name> \
  --tensor_input_name <output input name> --dataset_input_name <input dataset tensor name> \
  --tensor_input_min <min value in input tensor> --tensor_input_max <max value in input tensor> \
  --image_w <image width> --image_h <image height> --dataset_pic_path <example image path> \
  --output_path <output path for compile result>

Arguments

###-h show more help ###--eight_bit_mode using 8bit mode or 16bit mode

###--tesorboard-mode run tensorboard for current .pb file