Implement symmetric padding mode

[TomWildenhain-Microsoft]

Signed-off-by: Tom Wildenhain [email protected]

[lamhoangtung]

Thanks for the work @TomWildenhain-Microsoft.

What tensorflow version are you using for this? I've been trying this commit for issue #1686 but having Segmentation fault: 11 on both tensorflow 1.15 and 2.6.0.

What I've tried:

pip3 install git+https://github.com/onnx/tensorflow-onnx@tom/symmetric_padding
python3 -m tf2onnx.convert \
    --graphdef weights/COCO-MaskRCNN-R50C41x.pb \
    --output weights/test.onnx \
    --inputs image:0 \
    --outputs output/boxes:0,output/scores:0,output/labels:0,output/masks:0 \
    --opset 15

Output with tensorflow 2.6:

/Users/linus/anaconda3/envs/tf2_2/lib/python3.7/runpy.py:125: RuntimeWarning: 'tf2onnx.convert' found in sys.modules after import of package 'tf2onnx', but prior to execution of 'tf2onnx.convert'; this may result in unpredictable behaviour
  warn(RuntimeWarning(msg))
WARNING:tensorflow:From /Users/linus/anaconda3/envs/tf2_2/lib/python3.7/site-packages/tf2onnx/tf_loader.py:305: convert_variables_to_constants (from tensorflow.python.framework.graph_util_impl) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.compat.v1.graph_util.convert_variables_to_constants`
2021-09-03 22:25:52,312 - WARNING - From /Users/linus/anaconda3/envs/tf2_2/lib/python3.7/site-packages/tf2onnx/tf_loader.py:305: convert_variables_to_constants (from tensorflow.python.framework.graph_util_impl) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.compat.v1.graph_util.convert_variables_to_constants`
WARNING:tensorflow:From /Users/linus/anaconda3/envs/tf2_2/lib/python3.7/site-packages/tensorflow/python/framework/convert_to_constants.py:927: extract_sub_graph (from tensorflow.python.framework.graph_util_impl) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.compat.v1.graph_util.extract_sub_graph`
2021-09-03 22:25:52,313 - WARNING - From /Users/linus/anaconda3/envs/tf2_2/lib/python3.7/site-packages/tensorflow/python/framework/convert_to_constants.py:927: extract_sub_graph (from tensorflow.python.framework.graph_util_impl) is deprecated and will be removed in a future version.
Instructions for updating:
Use `tf.compat.v1.graph_util.extract_sub_graph`
2021-09-03 22:25:57,769 - INFO - Using tensorflow=2.6.0, onnx=1.10.1, tf2onnx=1.10.0/46e239
2021-09-03 22:25:57,769 - INFO - Using opset <onnx, 15>
2021-09-03 22:26:11,073 - INFO - Computed 2 values for constant folding
2021-09-03 22:26:17,887 - INFO - folding node using tf type=Mul, name=maskrcnn/deconv/mul
2021-09-03 22:26:17,888 - INFO - folding node using tf type=Mul, name=maskrcnn/deconv/mul_1
./to_onnx.sh: line 6: 15937 Segmentation fault: 11  python3 -m tf2onnx.convert --graphdef weights/COCO-MaskRCNN-R50C41x.pb --output weights/test.onnx --inputs image:0 --outputs output/boxes:0,output/scores:0,output/labels:0,output/masks:0 --opset 15

With tensorflow 1.15:

/Users/linus/anaconda3/envs/tf2_2/lib/python3.7/runpy.py:125: RuntimeWarning: 'tf2onnx.convert' found in sys.modules after import of package 'tf2onnx', but prior to execution of 'tf2onnx.convert'; this may result in unpredictable behaviour
  warn(RuntimeWarning(msg))
WARNING:tensorflow:From /Users/linus/anaconda3/envs/tf2_2/lib/python3.7/site-packages/tf2onnx/verbose_logging.py:76: The name tf.logging.set_verbosity is deprecated. Please use tf.compat.v1.logging.set_verbosity instead.

2021-09-03 22:20:47,185 - WARNING - From /Users/linus/anaconda3/envs/tf2_2/lib/python3.7/site-packages/tf2onnx/verbose_logging.py:76: The name tf.logging.set_verbosity is deprecated. Please use tf.compat.v1.logging.set_verbosity instead.

2021-09-03 22:20:55,109 - INFO - Using tensorflow=1.15.0, onnx=1.10.1, tf2onnx=1.10.0/46e239
2021-09-03 22:20:55,110 - INFO - Using opset <onnx, 15>
2021-09-03 22:21:07,678 - INFO - Computed 2 values for constant folding
2021-09-03 22:21:15,321 - INFO - folding node using tf type=Mul, name=maskrcnn/deconv/mul
2021-09-03 22:21:15,321 - INFO - folding node using tf type=Mul, name=maskrcnn/deconv/mul_1
./to_onnx.sh: line 6: 15310 Segmentation fault: 11  python3 -m tf2onnx.convert --graphdef weights/COCO-MaskRCNN-R50C41x.pb --output weights/test.onnx --inputs image:0 --outputs output/boxes:0,output/scores:0,output/labels:0,output/masks:0 --opset 15

[TomWildenhain-Microsoft]

@lamhoangtung opset 15 isn't released yet. Try opset 14. I'm using tensorflow==2.5.0 and onnx==1.10.1 with --opset 14 and it converts successfully.

[lamhoangtung]

Thanks @TomWildenhain-Microsoft. It's working with --opset 14