Model compiler error



  • I'm trying to compile my own model using the command this command:

    python3 __main__.py --pb_path pb_files/tiny-yolo-voc-1c.pb --tensor_output_name output --dataset_input_name input:0 --dataset_pic_path dataset/yolo_240_320/dog.bmp --image_w 320 --image_h 240 --eight_bit_mode True
    

    I'm getting this error:

    Traceback (most recent call last):
      File "__main__.py", line 145, in <module>
        main()
      File "__main__.py", line 118, in main
        k210_layers = model_loader_module.load_model(dataset_val, rfb, args)
      File "/home/kendryte/Kendryte_K210/kendryte-model-compiler/model_loader/pb/__init__.py", line 65, in load_model
        converter.convert()
      File "/home/kendryte/Kendryte_K210/kendryte-model-compiler/model_loader/pb/tensor_head_to_tensor_list.py", line 233, in convert
        while self.convert_step():
      File "/home/kendryte/Kendryte_K210/kendryte-model-compiler/model_loader/pb/tensor_head_to_tensor_list.py", line 228, in convert_step
        raise ValueError('no converter for', self.output_tensor.op.type, 'name:', self.output_tensor.op.name)
    ValueError: ('no converter for', 'Identity', 'name:', 'output')
    

    Any ideas how to fix this?



  • @sunnycase but.....

    1. Is there any difference between the files converted by nncase and ModelCompiler?
    2. Don't you need to use Model Complier anymore?


  • @Set

    1. You don't need to install tensorflow, but you need toco which is included in tensorflow to convert pb to tflite.
    2. Yes, you can use toco to convert pb to tflite. (toco is the official tensorflow lite converter)
    3. There are some prebuilt binaries at Release.


  • @latyas Excuse me:

    1. Do I have to install tensorflow through the source code?
    2. Convert the model file to tflite?
    3. Need the source code of nncase be compiled? Or just "install tinyos-tools"?

  • Staff



  • I have the same problem.