TensorRT 6.0.1 + TensorFlow 1.14 - No conversion function registered for layer: FusedBatchNormV3 yet

AI & Data Science Deep Learning (Training & Inference) TensorRT
1

Hi,

I had my models working fine with TensorFlow 1.13.0 + TensorRT 5.1.

Now I updated to TensorRT 6 and in the release notes it says it was tested with TensorFlow 1.14.0, so I updated also Tensorflow to 1.14.0.

When I try my deployment procedure now, the UFF converter complains:

Converting to UFF graph
Warning: No conversion function registered for layer: FusedBatchNormV3 yet.

  • Why is this the case? The release notes say it was successfully tested with TF 1.14, which doesn’t seem to be the case.

  • Is there a website with a list of the supported layers of the UFF converter? I can only find that list for TensorRT, but not for UFF converter.

Thanks!

2

Hi,
FusedBatchNormV3 operation is currently not supported by UFF parser.

You can try to convert your model to ONNX instead of UFF using tf2onnx:
https://github.com/onnx/tensorflow-onnx
tf2onnx supports converting this op to BatchNormalization op in ONNX:
https://github.com/onnx/tensorflow-onnx/blob/master/tf2onnx/onnx_opset/nn.py#L470

And BatchNormalization op is supported by the TensorRT ONNX parser:
https://github.com/onnx/onnx-tensorrt/blob/master/operators.md

Thanks

3

That’s great, thanks a lot for the detailed answer!

4

Hi

Did you solved the problem? The onnx supported this layer.

https://github.com/onnx/tensorflow-onnx/blob/master/tf2onnx/onnx_opset/nn.py#L501

I don’t know, when we work with Tensorflow models, Why we use UFF/ONNX converter and then parse them to TensorRT, we can use directly TF-TRT API, right? If so, what’s advantage of this UFF/ONNX converter method?

5

Its because TF-TRT still being a TF graph- model with some optimizations however a model converted from UFF/ONNX is completely optimized for TRT or any other deployment in DeepStream for example.