TDA4VM: Edge AI Cloud - Not able to compile customized model

Kruti Yadav

Part Number: TDA4VM

Hello,

I'm trying to run customized model Efficientnet b0. I changed some of the parameters as follows -

weights=None, include_top=False, input_shape=(960,540,3), activation='relu' at the time of training & generated new model.

While compiling the model in Edge AI Cloud, I'm getting 1 error. Here I've attached its screenshot.

Can anyone suggest how to resolve it?

over 4 years ago

0 Anand Pathak over 4 years ago

TI__Genius 9065 points

Hi,

Can you please share the model you are trying to run so I can try to reproduce the issue you are observing?

Regards,

Anand

0 Kruti Yadav over 4 years ago in reply to Anand Pathak

Prodigy 20 points

Please find attached url of the model I used.

EfficientNet-B0

Thanks

0 Anand Pathak over 4 years ago in reply to Kruti Yadav

TI__Genius 9065 points

Hi Kruti,

I am still investigating this issue and need some more time. Will keep you updated.

Regards,

Anand

0 Anand Pathak over 4 years ago in reply to Anand Pathak

TI__Genius 9065 points

Hi Kruti,

This issue is coming because there are dynamic sized tensors in your network. I see Shape->StridedSlice->Pack layers at several places in the network. Are they intentional or they just got generated as part of model export?

Regards,

Anand

0 Anand Pathak over 4 years ago in reply to Anand Pathak

TI__Genius 9065 points

Hi Kruti,

You can refer to the following link to remove the dynamic tensors in your network and then give a try again, the error should be resolved:

https://github.com/tensorflow/tensorflow/issues/43882#issuecomment-731636562

Hope this helps. I am marking as resolved, please feel free to reopen the thread in case there are any further questions.

Regards,

Anand

Processors

Processors forum

TDA4VM: Edge AI Cloud - Not able to compile customized model