Updated on 2022-08-19 GMT+08:00

Model Input Directory Specifications

The input directory of a model (oriented to Ascend chips) developed locally or developed in ModelArts and converted on Huawei HiLens must meet certain specifications. Huawei HiLens has the following requirements on the model input directory:

  • Input directory specifications for Caffe models:
    |
    |---xxxx.caffemodel         Model parameter file, mandatory. Only one model parameter file can exist in the input directory. 
    |---xxxx.prototxt           Model network file, mandatory. Only one model network file can exist in the input directory.
    |---insert_op_conf.cfg      Insertion operator configuration file, optional. Only one insertion operator configuration file can exist in the input path.
    |---plugin                  Custom operator directory, optional The input directory can contain only one plugin folder. Only custom operators developed based on Tensor Engine (TE) are supported.
  • Input directory specifications for TensorFlow models (in frozen_graph or saved_model format):
    frozen_graph format
    |
    |---xxxx.pb                 Model network file, mandatory. Only one model network file can exist in the input path. The model must be in frozen_graph or saved_model format.
    |---insert_op_conf.cfg      Insertion operator configuration file, optional. Only one insertion operator configuration file can exist in the input path.
    |---plugin                  Custom operator directory, optional The input directory can contain only one plugin folder. Only custom operators developed based on Tensor Engine (TE) are supported.

    saved_model format

    |
    |---saved_model.pb          Model network file, mandatory. Only one model network file can exist in the input path. The model must be in frozen_graph or saved_model format.
    |---variables               Fixed subdirectory name, including the model weight deviation, mandatory
        |---variables.index     Mandatory
        |---variables.data-00000-of-00001 Mandatory
    |---insert_op_conf.cfg      Insertion operator configuration file, optional. Only one insertion operator configuration file can exist in the input path.
    |---plugin                  Custom operator directory, optional The input directory can contain only one plugin folder. Only custom operators developed based on Tensor Engine (TE) are supported.