Converting model from onnx to TensorRT fails

When trying to convert a model to TensorRT I get the following error:

[8] Assertion failed: cond.is_weights() && cond.weights().count() == 1 && "If condition must be a initializer!" 

I have download efficientnetD2 from Tensorflow Model Zoo and converted it to onnx-format. When I try to convert the model to TensorRT I get the error mentioned above. I used the --opset 11 in order to make it into a onnx-model.

Edit: I have now tried on many different models from the Tensorflow Model Zoo. All of them giving the same error.
Are there anyone that has been able to convert any of the models on that list into TensorRT format?

Are there any solutions to this?

Hi,

We want to check this further.
Could you share the generated ONNX model with us?

Thanks.

model.zip (22.3 MB)
Sure, model can be found attached to this post.
I can upload more converted models if desirable.
Thanks for your support!

Hi again, any updates on this issue?

Hi,

Thanks for sharing the model.

We are checking this internally.
Will get back to you later.

Thanks.

Hi,

We are preparing an Efficientnet example from the TensorFlow Model Zoo.
Please wait for our announcement for the new JetPack and sample release (in TensorRT OSS).

Thanks.

Sweet!
Any idea on the target date for the next release (approx)?

Please check Tensorflow Object Detection - Really slow loading time for the details.
The example is released and the JetPack software will be available soon.

1 Like