当我们拿到别人提供的pb模型时,我们很可能不清楚inputs所需的参数,可以使用以下方法来查看模型信息:
找到你的tensorflow安装位置,依次定位到以下目录:
tensorflow_core/python/tools
执行如下命令:
python saved_model_cli.py show --dir 模型路径 --all
之后可以得到如下信息:
signature_def['serving_default']: The given SavedModel SignatureDef contains the following input(s): inputs['image_shape'] tensor_info: dtype: DT_FLOAT shape: (2) name: Placeholder_366:0 inputs['input_image'] tensor_info: dtype: DT_FLOAT shape: (-1, -1, -1, 3) name: input_1:0 The given SavedModel SignatureDef contains the following output(s): outputs['concat_11/concat:0'] tensor_info: dtype: DT_FLOAT shape: (-1, 4) name: concat_11/concat:0 outputs['concat_12/concat:0'] tensor_info: dtype: DT_FLOAT shape: (-1) name: concat_12/concat:0 outputs['concat_13/concat:0'] tensor_info: dtype: DT_INT32 shape: (-1) name: concat_13/concat:0 Method name is: tensorflow/serving/predict
转自:https://blog.csdn.net/github_34790294/article/details/103725850