KeyError: "The name 'input_image:0' refers to a Tensor which does not exist. The operation, 'input_image', does not exist in the graph."? about mrcnn_serving_ready

[AI-ML-Enthusiast]

During creating the freeze_graph, just print model.inputs and model.outputs. In my case, the name was
input_image_7. Thanks

[bendangnuksung]

The tensor input_image should be in the Graph, are you loading the right model? and have you tried loading the same config class which you used for training?

[i-chaochen]

Hi, thanks for sharing this serving_ready @bendangnuksung

I can create and load the frozen graph now.

After I loaded the graph, I can get_tensor_by_name('input_image:0'), but what about the outputs?

I can see in your main.py, the output contains output_detection and output_mask, does output_detection contain bounding_box{rois: [N, (y1, x1, y2, x2)]}, class_ids and scores?

[AI-ML-Enthusiast]

@bendangnuksung @i-chaochen
problem is solved. The tensor name was not same as .pb file. thanks

[i-chaochen]

problem is solved. The tensor name was not same as .pb file. thanks

so where do you know the tensor name from the frozen graph?

[Asher94]

How did you get the output for the model.inputs? I tried printing them but nothing appeared.