Deyu Huang

When I run [tf2onnx ](https://github.com/onnx/tensorflow-onnx)which is using `tensorflow_hub`, there are some `DeprecationWarning` encoutered. ``` lib/python3.9/site-packages/tensorflow_hub/__init__.py:74: DeprecationWarning: distutils Version classes are deprecated. Use packaging.version instead. if (distutils.version.LooseVersion(tf.__version__) < /home/deyu/anaconda3/envs/py39/lib/python3.9/site-packages/tensorflow_hub/__init__.py:75: DeprecationWarning: distutils...

fixes https://github.com/onnx/tensorflow-onnx/issues/2002 Signed-off-by: Deyu Huang

Support new onnx signal operators

2

There are several signal operators added in [ONNX 1.12](https://github.com/onnx/onnx/releases/tag/v1.12.0). Tracking issue for it. - DFT ONNX: https://github.com/onnx/onnx/blob/main/docs/Operators.md#DFT TensorFlow: https://www.tensorflow.org/api_docs/python/tf/signal/fft https://www.tensorflow.org/api_docs/python/tf/signal/ifft - STFT ONNX: https://github.com/onnx/onnx/blob/main/docs/Operators.md#STFT TensorFlow: https://www.tensorflow.org/api_docs/python/tf/signal/stft - HannWindow ONNX: https://github.com/onnx/onnx/blob/main/docs/Operators.md#HannWindow...

fixes https://github.com/onnx/tensorflow-onnx/issues/1952 Signed-off-by: Deyu Huang

Support for tensorflowjs 3.17 -- maxNumElements attr in TensorListReserve dynamic list shoule be ignored

3

5 tests are failed with latest [tfjs vesion 3.17](https://github.com/tensorflow/tfjs/releases/tag/tfjs-v3.17.0), which have skiped so far. Need to support it in the future. https://github.com/onnx/tensorflow-onnx/blob/772dbe605e94880d4cbf8e9e1ed9ce9517b89c64/tests/test_cond.py#L121 https://github.com/onnx/tensorflow-onnx/blob/772dbe605e94880d4cbf8e9e1ed9ce9517b89c64/tests/test_loops.py#L69 https://github.com/onnx/tensorflow-onnx/blob/772dbe605e94880d4cbf8e9e1ed9ce9517b89c64/tests/test_loops.py#L162 https://github.com/onnx/tensorflow-onnx/blob/772dbe605e94880d4cbf8e9e1ed9ce9517b89c64/tests/test_loops.py#L186 https://github.com/onnx/tensorflow-onnx/blob/772dbe605e94880d4cbf8e9e1ed9ce9517b89c64/tests/test_loops.py#L224

Add shape inference as an arg for tools/onnx-optimize.py

1

In https://github.com/onnx/tensorflow-onnx/blob/1915fb52270856debc74ba5d28e332e46ccc1e1a/tools/onnx-optimize.py#L59 the `optimizer.optimize_graph()` method would eliminate infered shape. This PR adds the shape inference after model optimized. Signed-off-by: Deyu Huang

Support RaggedGather operator when rank > 1

1

**Describe the bug** Now, tf2onnx only supports **RaggedGather** op rank == 1. But there are some NLP models have rank > 1 scenarios. **System information** - OS Platform and Distribution:...

ssd-mobilenet 300x300 onnx model converted from tensorflow can be optimized.

1

[`ssd-mobilenet 300x300`](https://zenodo.org/record/4735652/files/ssd_mobilenet_v1_coco_2018_01_28.onnx) from https://github.com/mlcommons/inference/tree/r1.1/vision/classification_and_detection. The `ZerosLike inserts Casts before Mul` issue can be optimized refer: https://github.com/onnx/tensorflow-onnx/issues/1997 which can improve the inference performance. Optimized model `opset 11` and `opset 15`. [ssd_mobilenet_v1_coco_2018_01_28_opset11.zip](https://github.com/mlcommons/inference/files/9149255/ssd_mobilenet_v1_coco_2018_01_28_opset11.zip)...