Gather is an alternative approach used by model optimizer
to implement --reverse_input_channels option.
Currently, if it's in use, convolutions in GPU plugin
may use other formats that leads to choosing less performant kernels.
* Add check for python version to networkx dependency
* Update condition, task CVS-72806
* Update check to not break old Python versions support
* Update POT NetworkX dependency
* Allign requirements list format with setuptools documentation
I'll merge this. We need to talk with someone who know openvino build system better than we to "use cmake function ov_ncc_style_check to perform such check automatically". As far as I can see, currently it used only in one place, so it is not common approach for openvino components
* Implement detection_output shape infer
* revise and update the code flow
* update based on review.
* Update based on review
* Implement the shuffle_channels Op shape inference.
* Fix CI coding style issue.
* Implement the select OP shape inference.
* Update based on the review comments
* Update based on the review comments.
* Add pragma once for the shape inference head.
* Add new shape_infer test file for detection_output OP.
* Ensure the header would only be included once.
* Add shuffle_channels OP shape infer test.
* Add shape_infer() invocations into shape_inference() API
shape_inference() API support Select, ShuffleChannels, DetectionOutput OPs
Fix extra pragma, unnecessary friend function declaration.
* Update based on the review comments.
* Move the shape infer API helpers into new folder.
* Applied review comments.
* Applied 2nd review comments
* Applied review comments
* Fix coding style.
* Update
* Applied review comments.
* Fix comipling issue of unused variable.
* Fix the CI issue.
* Update the coding style
* Move test cases into new folder
* Applied review comments.
* cli_parser.py fix to accept scalar value for freezing
* update cli help
* fixed unit-tests, clarified help for specifying data type
* typos correction
* Support for common rt_info attribute in MO IR Reader
* Add missed change
* Moved back wrong change
* Change attr name
* Add support for rt_info for out ports
* Add emitting for rt_info
* Fix restoration error
* Add support for rt_info for input ports
* Add more comments
* Set correct layout attr to restored graph
* Test MO and IR Reader on attacking inputs
Signed-off-by: Roman Kazantsev <roman.kazantsev@intel.com>
* Add test to check IR Reader against untrusted well-formed IR
Signed-off-by: Roman Kazantsev <roman.kazantsev@intel.com>
* Refactor IR Reader tests with corrupted IR
Signed-off-by: Roman Kazantsev <roman.kazantsev@intel.com>
* Test for regular expression denial of service
Signed-off-by: Roman Kazantsev <roman.kazantsev@intel.com>
* Remove undesired word like bomb
Signed-off-by: Roman Kazantsev <roman.kazantsev@intel.com>
* Move tests to new location
Signed-off-by: Roman Kazantsev <roman.kazantsev@intel.com>
* Use correct import
Signed-off-by: Roman Kazantsev <roman.kazantsev@intel.com>
* Revert blank line
Signed-off-by: Roman Kazantsev <roman.kazantsev@intel.com>
* Create mvp
* Implement new API & Refactoring
* Fix -oname for models whose name of output layer contains a port number
* Fix step numbers
* Create utils.py
* Remove shebang from utils.py
* Fix `-iname` option
* [IE Python Speech Sample] Migrate to OV 2.0 API
* improvements
* flake notes
* improved code style like as C++
* linters changes
* changed data.py
* sync output with C++ sample
Co-authored-by: Maxim Gordeev <maxim.gordeev@intel.com>
