Missing support of MatMul (for ONNX import) and Unsqueeze (Core + backend CPU + ONNX import)

changed the description

Hello Mary,

It looks like your ONNX contains an operator which is not yet supported by the platform.

Aidge import by default operator as a Generic Operator. A generic operator does not have an implementation and thus cannot run a forward pass.

We cannot help you with the current logs you sent us. Can you please send us the log of the cell where you import the ONNX model ? This will help us to see which operator you are missing.

Cheers, Cyril

added IssueMissing Feature label

Hello Cyril,

Thank you for your response.

The code snippet used for the conversion is as follows:

import tensorflow as tf
from keras.models import load_model

model1 = load_model('CD_model.h5')
tf.saved_model.save(model1, "CD_model")

This script is executed in a terminal using the following command:

python -m tf2onnx.convert --saved-model CD_model --output CD_model.onnx

Unfortunately, due to policy security reasons, we cannot provide specific details about the architecture of our model. However, it predominantly consists of the following operations:

Unsqueeze|GenericOperator
Transpose|GenericOperator
perm
Conv
Squeeze
Add
Relu
Reshape|GenericOperator
MaxPool
MatMul|GenericOperator

Cheers, Mary

Transpose and Reshape should be supported with the most recent version of aidge_core / aidge_cpu /aidge_onnx.

For MatMul and Unsqueeze, further developments are required as they are not supported yet.

Cheers, Cyril

Thank you for your response!

Regarding MatMul and Unsqueeze, please keep me posted on any updates or progress in supporting these operations.

Cheers, Mary

&2 you can find with this EPIC the latest updates for the operator support.

I will add this issue to the epic !

I will just add that we welcome the contributions from outside so if you have time to contribute to the framework by adding the support of MatMul (for ONNX import) and Unsqueeze (Core + backend CPU + ONNX import). I will gladly help you by pointing you to the right ressources and answering your questions !

Cheers, Cyril

mentioned in epic &2

changed title from Aidge::GenericOperator_Op::outputDimsForwarded() const: Assertion `false && "GenericOperator cannot forward dims"' failed to Missing support of MatMul (for ONNX import) and Unsqueeze (Core + backend CPU + ONNX import)

added Help Wanted label

MatMul supported since

!76 (merged) in aidge_core
aidge_backend_cpu!34 (merged) in aidge_backend_cpu
aidge_onnx!29 (merged) in aidge_onnx for the import

changed milestone to %aidge_core - v0.2.0

changed health status to on track

added to epic &2

added LanguageC++ PriorityMedium StatusWIP labels

removed StatusWIP label

added Bugfound 🐛 label

added #96 (closed) as child task

added #97 (closed) as child task

Hi @malo12ol98 , I'm going to start working on the last missing operator (squeeze, unsqueeze #96 (closed)) If the network is publicly available and in order to test the operator I would like to try on the network you are trying to load.

If possible could you send me a link to download it in the issue #96 (closed) ? Thanks in advance and have a nice day!

mentioned in merge request !102 (closed)

mentioned in merge request aidge_backend_cpu!49 (closed)

mentioned in merge request aidge_backend_cpu!51 (closed)

mentioned in merge request aidge_onnx!34 (closed)

added StatusWIP label

Hi @gregkub, Thank you for your help. Unfortunately I cannot send you the actual network for confidential reason. I could send you a changed version like I did in the comments above.

Thanks in advance and have a nice day!