edgetpu: edgetpu-deeplab compile error

After edgetpu-deeplab model trained on my customer dataset which has only two class, I convert the frozen model to tflite format. When I compile the frozen tflite model using edgetpu_compiler, it runs forever and return nothing on the terminal except 'Edge TPU Compiler version 15.0.340273435’.

convert command:

tflite_convert \
--graph_def_file=./frozen_inference_graph_quant.pb \
--output_file=./edgetpu_deeplab_quant.tflite \
--output_format=TFLITE \
--input_shape=1,361,,361 \
--inference_type=QUANTIZED_UINT8 \
--inference_input_type=QUANTIZED_UINT8 \
--std_dev_values=128 \
--mean_values=128 \
--change_concat_input_ranges=true \
--input_arrays="MobilenetEdgeTPU/MobilenetEdgeTPU/input" \
--output_arrays="ArgMax"

I see that coral.ai provides a semantic segmentation model, deeplab_mnv2, can you tell me how to get a edgetpu-deeplab model running on edge tpu?