What should the inputs and outputs of the SavedModel SignatureDef look like?

[cpoptic]

It is unclear precisely what the SavedModel inputs and outputs should look like for the Mask RCNN model, once frozen and exported in a Servable format.

What should the inputs and outputs of the SavedModel SignatureDef look like?

I'm seeing these 3 input tensors and these two output tensors. Is this correct?

The given SavedModel SignatureDef contains the following input(s):
inputs['input_anchors'] tensor_info:
dtype: DT_FLOAT
shape: (-1, -1, 4)
name: input_anchors:0
inputs['input_image'] tensor_info:
dtype: DT_FLOAT
shape: (-1, -1, -1, 3)
name: input_image:0
inputs['input_image_meta'] tensor_info:
dtype: DT_FLOAT
shape: (-1, 33)
name: input_image_meta:0
The given SavedModel SignatureDef contains the following output(s):
outputs['mrcnn_detection/Reshape_1'] tensor_info:
dtype: DT_INT64
shape: (-1)
name: mrcnn_detection/Reshape_1:0
outputs['mrcnn_mask/Reshape_1'] tensor_info:
dtype: DT_FLOAT
shape: (-1, 100, 28, 28, 21)
name: mrcnn_mask/Reshape_1:0
Method name is: tensorflow/serving/predict

[chrigui94]

Hi @cpoptic ,
did you understand it? , I have the same problem I am implementing the mrcnn frozen model in c# application, but it requires these 3 inputs but I have only the input_image tensor.. how can I get the other 2 variables?

[bendangnuksung]

Please check inferencing/saved_model_inference.py on how the inferencing is done on saved_model.