multiple input/output model, slice index error

[frk1993]

Hi,
I am trying to serve a complex model with multiple input and multiple output. To test the saved model (pb model), I try to load the model and run an example. But I am facing slice index error. Here is the information about my model:

And here my code to load and run the mode:

And here the slice index error:

I hope you can help me!

best Regards

[vbezgachev]

Hi @frk1993 ,
How do you create anchors, input_image and image_meta?
When do you get the error? By executing the session.run()?

[frk1993]

Hi,
thanks for you response. I am saving l the anchors, input_image and image_met as numpy arrays from the original script. Where they are use for the eras.predict method. Then I load this arrays when I want to test the saved_model.
I get this error when I am running session.run()

I could solve the problem by increasing the batch size to two. This means I am appending anchors, input_image and image meta.
Do you have an idea why this error is occurring when I just use batch size one?

[vbezgachev]

Hi @frk1993 ,

It sounds for me like with a batch of size 1, all arrays are missing one dimension. Could you print out a shape of, for instance, input_image? To my understanding, the array should have 4 dimensions.
It is difficult to say seeing just a few details. I suppose that a problem lays in the loading of arrays. Probably, you need to expand the arrays in case of batch size 1.

[frk1993]

Hi @Vetal1977 ,
Thanks for you quick response. I could solve the problem. The problem was the following: Before saving the protobuf model I am creating the model and here I had to define the variable images_per_gpu. And the default value for this variable was set to two. After setting this variable to one, the problem was solved.