platform/upstream/dldt.git
4 years agoFixed README (#2948)
Ilya Churaev [Tue, 3 Nov 2020 07:56:28 +0000 (10:56 +0300)]
Fixed README (#2948)

4 years agoFix cases when identities inserted identites are separated by nonFunctional layers...
Kamil Magierski [Tue, 3 Nov 2020 07:49:58 +0000 (08:49 +0100)]
Fix cases when identities inserted identites are separated by nonFunctional layers (#2771)

4 years agoHandle multiple GNA devices and modes HW/SW (#2720)
Krzysztof Bruniecki [Tue, 3 Nov 2020 07:49:09 +0000 (08:49 +0100)]
Handle multiple GNA devices and modes HW/SW (#2720)

* Introduce Gna2DeviceGetCount

* Check for the number of GNA devices

* Add Gna2DeviceGetCount() mock

4 years ago[GNA] Added export input/output names (#2601)
Andrey Dmitriev [Tue, 3 Nov 2020 07:45:08 +0000 (10:45 +0300)]
[GNA] Added export input/output names (#2601)

* [GNA] Added export input/output names

* fix comment

* Change version

4 years agoRemove obsoleted v0::Reshape operator (#2878)
Mateusz Tabaka [Tue, 3 Nov 2020 04:01:34 +0000 (05:01 +0100)]
Remove obsoleted v0::Reshape operator (#2878)

* Remove obsoleted v0::Reshape operator

* Fix handling negative dims on 32 bit platforms

Change-Id: Ia18e20df9a79fd0b7b38c434bee7140599833952

* apply code format

Change-Id: I4ddfb9e7033cbc97e167922f6ee5873d5487e551

4 years agoAdded check in hello sample (#2896)
Anton Romanov [Tue, 3 Nov 2020 03:34:06 +0000 (19:34 -0800)]
Added check in hello sample (#2896)

* Added check in hello sample

* move check before inference

4 years agoCompilation with MSVC for ARM (#2928)
Ilya Lavrenov [Mon, 2 Nov 2020 21:55:49 +0000 (00:55 +0300)]
Compilation with MSVC for ARM (#2928)

* Enable compilation with MSVC for ARM / ARM64

* Condition fix

4 years agoAdded transformation config to support automl efficientdet models (#2894)
Evgeny Lazarev [Mon, 2 Nov 2020 16:21:05 +0000 (19:21 +0300)]
Added transformation config to support automl efficientdet models (#2894)

* Added transformation config to support automl efficientdet-4 model

* Added configuration file to convert Automl EfficientDet model

* Updated unit test for Pack

* Added instruction on how to convert EfficientDet Tensorflow model

* Updated documentation on how to convert EfficientDet model

* Updated a documentation with instruction on how to convert Automl EfficientDet.

4 years ago[CPU] Added eltwise Round-5 (#2347)
Alexandra Sidorova [Mon, 2 Nov 2020 15:44:50 +0000 (18:44 +0300)]
[CPU] Added eltwise Round-5 (#2347)

4 years agoFixed FAST build mode (#2926)
Ilya Lavrenov [Mon, 2 Nov 2020 14:58:52 +0000 (17:58 +0300)]
Fixed FAST build mode (#2926)

4 years ago[IE TESTS] Split Layer Test refactoring (#2915)
Aleksandr Korolev [Mon, 2 Nov 2020 13:29:59 +0000 (16:29 +0300)]
[IE TESTS] Split Layer Test refactoring (#2915)

Co-authored-by: kora6 <kora6@github.com>
4 years agoRemove remaining Myriad-2 references (#2853)
Andrey Kamaev [Mon, 2 Nov 2020 12:10:50 +0000 (15:10 +0300)]
Remove remaining Myriad-2 references (#2853)

* Fix Myriad firmware copying command for multi-target cmake generators

* Remove remaining Myriad-2 references

* Remove remaining Myriad-2 compute stick from documentation

4 years agoChange pip install documentation (#2884)
Anton Romanov [Mon, 2 Nov 2020 11:09:54 +0000 (03:09 -0800)]
Change pip install documentation (#2884)

4 years agoVisitor API deserialization - create models for ops (#2889)
Bartek Szmelczynski [Mon, 2 Nov 2020 10:48:53 +0000 (11:48 +0100)]
Visitor API deserialization - create models for ops (#2889)

* add tests for 11 operators, which compare V7 and V10 version of the model optimizer

* add test for space_to_depth model

4 years ago[IE CLDNN] Fixed unitialized parameters in resample primitive which led to kernels...
Vladimir Paramuzov [Mon, 2 Nov 2020 10:31:07 +0000 (13:31 +0300)]
[IE CLDNN] Fixed unitialized parameters in resample primitive which led to kernels recompilation (#2923)

4 years agoUnmark passed ONNX test
Alexander Peskov [Thu, 29 Oct 2020 15:14:39 +0000 (18:14 +0300)]
Unmark passed ONNX test

Signed-off-by: Alexander Peskov <alexander.peskov@intel.com>
4 years ago[TEST] One more trivival loop test
Alexander Peskov [Wed, 28 Oct 2020 22:34:18 +0000 (01:34 +0300)]
[TEST] One more trivival loop test

Signed-off-by: Alexander Peskov <alexander.peskov@intel.com>
4 years ago[CPU] Fix case with unused port in Loop body
Alexander Peskov [Wed, 28 Oct 2020 21:15:53 +0000 (00:15 +0300)]
[CPU] Fix case with unused port in Loop body

Signed-off-by: Alexander Peskov <alexander.peskov@intel.com>
4 years ago[TEST] One more trivial test case on Loop
Alexander Peskov [Tue, 27 Oct 2020 22:42:52 +0000 (01:42 +0300)]
[TEST] One more trivial test case on Loop

Also fixed compilation with gcc4.8

Signed-off-by: Alexander Peskov <alexander.peskov@intel.com>
4 years ago[CPU] Improve replication of passthrough Loop body
Alexander Peskov [Tue, 27 Oct 2020 22:36:33 +0000 (01:36 +0300)]
[CPU] Improve replication of passthrough Loop body

There was some problem with replicatioon of simple loop body where
input was used as output as is.

Also was voided usage of special prefixes like "in_" for Loop body inputs.

Signed-off-by: Alexander Peskov <alexander.peskov@intel.com>
4 years agoMin. Additional assert in copyTIBoody
Alexander Peskov [Tue, 27 Oct 2020 22:34:16 +0000 (01:34 +0300)]
Min. Additional assert in copyTIBoody

Signed-off-by: Alexander Peskov <alexander.peskov@intel.com>
4 years ago[TEST] Several more Loop test with static shapes
Alexander Peskov [Wed, 21 Oct 2020 23:49:51 +0000 (02:49 +0300)]
[TEST] Several more Loop test with static shapes

Signed-off-by: Alexander Peskov <alexander.peskov@intel.com>
4 years ago[CPU] implement Loop specification of TI node
Alexander Peskov [Wed, 21 Oct 2020 23:48:48 +0000 (02:48 +0300)]
[CPU] implement Loop specification of TI node

Signed-off-by: Alexander Peskov <alexander.peskov@intel.com>
4 years ago[NG] Keep special purposes port info in Loop after conversion to legacy
Alexander Peskov [Wed, 21 Oct 2020 23:37:20 +0000 (02:37 +0300)]
[NG] Keep special purposes port info in Loop after conversion to legacy

Also:
Simplified logic of data object name restoring. Avoid duplicatin of input ports
in case of multiple consumers. Provided code has WA comment in corresponding
naming restore section. Also added WA section with restore U8 precision for ouputs.
Avoid to eliminate limitation of CNNNetwork converter.

Signed-off-by: Alexander Peskov <alexander.peskov@intel.com>
4 years ago[TEST] improve data utils
Alexander Peskov [Wed, 21 Oct 2020 23:30:42 +0000 (02:30 +0300)]
[TEST] improve data utils

Signed-off-by: Alexander Peskov <alexander.peskov@intel.com>
4 years agoAvoid '-' symbol for test name
Alexander Peskov [Tue, 20 Oct 2020 09:37:43 +0000 (12:37 +0300)]
Avoid '-' symbol for test name

Signed-off-by: Alexander Peskov <alexander.peskov@intel.com>
4 years agoLoop tests enabled back
Alexander Peskov [Mon, 19 Oct 2020 17:17:50 +0000 (20:17 +0300)]
Loop tests enabled back

Signed-off-by: Alexander Peskov <alexander.peskov@intel.com>
4 years ago[IE] Improve Network topological sort in case of disconnected graph
Alexander Peskov [Wed, 28 Oct 2020 22:37:20 +0000 (01:37 +0300)]
[IE] Improve Network topological sort in case of disconnected graph

Signed-off-by: Alexander Peskov <alexander.peskov@intel.com>
4 years agoImprove CopyTIBody util to cover disconnected graphs
Alexander Peskov [Mon, 19 Oct 2020 17:16:34 +0000 (20:16 +0300)]
Improve CopyTIBody util to cover disconnected graphs

Signed-off-by: Alexander Peskov <alexander.peskov@intel.com>
4 years agoRemove obsoleted v0::StopGradient op (#2935)
Mateusz Tabaka [Mon, 2 Nov 2020 09:22:01 +0000 (10:22 +0100)]
Remove obsoleted v0::StopGradient op (#2935)

4 years agorefactor one-hot nGraph backend test (#2578)
Tomasz Jankowski [Mon, 2 Nov 2020 09:12:39 +0000 (10:12 +0100)]
refactor one-hot nGraph backend test (#2578)

4 years agoAvoid verbose return-type warnings (#2420)
Tomasz Jankowski [Mon, 2 Nov 2020 09:11:35 +0000 (10:11 +0100)]
Avoid verbose return-type warnings (#2420)

4 years agonGraph backend tests refactor: gather, numeric (#2367)
Tomasz Jankowski [Mon, 2 Nov 2020 09:10:25 +0000 (10:10 +0100)]
nGraph backend tests refactor: gather, numeric (#2367)

4 years agoFixed compilation errors / warnings with clang (#2929)
Ilya Lavrenov [Mon, 2 Nov 2020 08:56:38 +0000 (11:56 +0300)]
Fixed compilation errors / warnings with clang (#2929)

4 years agoRemove obsoleted v0::Or operator (#2857)
Mateusz Tabaka [Mon, 2 Nov 2020 03:49:03 +0000 (04:49 +0100)]
Remove obsoleted v0::Or operator (#2857)

* Remove obsoleted v0::Or operator

* apply style format

4 years agoRemove deprecated v0::Round op (#2905)
Mateusz Tabaka [Mon, 2 Nov 2020 03:47:35 +0000 (04:47 +0100)]
Remove deprecated v0::Round op (#2905)

4 years agoRemove deprecated v0::Reverse op (#2899)
Mateusz Tabaka [Mon, 2 Nov 2020 03:46:30 +0000 (04:46 +0100)]
Remove deprecated v0::Reverse op (#2899)

* Remove deprecated v0::Reverse op

* fix unit tests

* add include for std::inserter

4 years agoTemporarily skip ExecGraphTests.CheckExecGraphInfoSerialization on GPU (#2921)
Andrey Babushkin [Sun, 1 Nov 2020 09:55:50 +0000 (12:55 +0300)]
Temporarily skip ExecGraphTests.CheckExecGraphInfoSerialization on GPU (#2921)

* Skip ExecGraphTests.CheckExecGraphInfoSerialization on GPU

* [execution_graph_tests] Add test skipping macro

* Add missing import

4 years agoAdding sinks to ngraph::Function (#2741)
Svetlana Dolinina [Sat, 31 Oct 2020 16:41:05 +0000 (19:41 +0300)]
Adding sinks to ngraph::Function (#2741)

* initial draft of adding sinks to ngraph::Function

* style fixes

* code style fixes

* code style fixes

* code style fix

* review fix+build fix

* code style fix

* fix build

* API changed according to latest discussion

* review fixes

* review fixes + tests

* added 1 more ctor

* style fixes

* used new api in ir parser

* fixed build

* review fixes

* remove validate_nodes_and_infer_types from remove_sink/remove_result

* removed validate_.. after discussion

* style fix

4 years ago[IE][VPU]: GatherND layer & tests (#2710)
Maxim Kurin [Fri, 30 Oct 2020 23:02:23 +0000 (02:02 +0300)]
[IE][VPU]: GatherND layer & tests (#2710)

* GatherND layer & test
* Update vpu firmware 1452

4 years ago[LPT] nGraph nodes naming fix (#2822)
Edward Shogulin [Fri, 30 Oct 2020 20:23:35 +0000 (23:23 +0300)]
[LPT] nGraph nodes naming fix (#2822)

* [LPT] functional tests: FakeQuantize with dynamic intervals

* [LPT] decomposeFakeQuantize: removed debug info

* [LPT] Add NetworkHelper::mark_as_dequantization_op function

[ngraph] Fix compare runtime info function

[LPT] Fix test cases with no DEQUANTIZATION runtime attribute

[LPT] Change include path for dequantization op

* [LPT] Remove Subtract functional test, enable and rename legacy tests

Co-authored-by: Vladislav Golubev <vladislav.golubev@intel.com>
Co-authored-by: Aleksandr Pertovsky <aleksandr.pertovsky@intel.com>
4 years agoAdded support for Visual Studio VERSIONINFO (#2900)
Ilya Lavrenov [Fri, 30 Oct 2020 18:38:34 +0000 (21:38 +0300)]
Added support for Visual Studio VERSIONINFO (#2900)

* Generate vs_version file

* Fixed comments

* Fixed comments

4 years agoFixed cmake stage with BUILD_SHARED_LIBS=ON (#2914)
Ilya Lavrenov [Fri, 30 Oct 2020 18:37:51 +0000 (21:37 +0300)]
Fixed cmake stage with BUILD_SHARED_LIBS=ON (#2914)

4 years agoFixed compilation with ENABLE_V7_SERIALIZE=ON (#2898)
Ilya Lavrenov [Fri, 30 Oct 2020 18:19:12 +0000 (21:19 +0300)]
Fixed compilation with ENABLE_V7_SERIALIZE=ON (#2898)

* Fixed compilation with ENABLE_V7_SERIALIZE=ON

* Removed legacy code

4 years agoadded OpenVINO Model Server to docs (#2541)
Trawinski, Dariusz [Fri, 30 Oct 2020 16:57:01 +0000 (17:57 +0100)]
added OpenVINO Model Server to docs (#2541)

* added OpenVINO Model Server

* updated documentation to include valid links

* minor fixes

* Fixed links and style

* Update README.md

fixed links to model_server

* more corrections

* dropped reference in ie_docs and minor fixes

* Update README.md

Fixed links to Inference Engine pages

Co-authored-by: Alina Alborova <alina.alborova@intel.com>
Co-authored-by: Andrey Zaytsev <andrey.zaytsev@intel.com>
4 years ago[IE CLDNN] Added int8 output suppirt into bfyx_to_fsv16 fp kernel (#2906)
Vladimir Paramuzov [Fri, 30 Oct 2020 12:19:02 +0000 (15:19 +0300)]
[IE CLDNN] Added int8 output suppirt into bfyx_to_fsv16 fp kernel (#2906)

4 years ago[IE CLDNN] Fix for some unwanted behaviour in layout optimizer (#2669)
Mikołaj Życzyński [Fri, 30 Oct 2020 12:18:14 +0000 (13:18 +0100)]
[IE CLDNN] Fix for some unwanted behaviour in layout optimizer (#2669)

4 years ago[MO] Add CMake install for Model Optimizer (28291) (#2831)
Sergey Lyubimtsev [Fri, 30 Oct 2020 11:37:06 +0000 (14:37 +0300)]
[MO] Add CMake install for Model Optimizer (28291) (#2831)

* [MO] Add CMake install for Model Optimizer

* [MO] Update test for version.py

* [MO] Add CMake install for Model Optimizer

* [MO] Update test for version.py

* [MO] Add CMake install for Model Optimizer

* [MO] Update test for version.py

* [MO] fix file permissions for install location

4 years ago[IE][VPU][Myriad]: NMS layer optimization; support big buffers (#2839) (#2839)
Andrey Sokolov [Fri, 30 Oct 2020 10:30:21 +0000 (13:30 +0300)]
[IE][VPU][Myriad]: NMS layer optimization; support big buffers (#2839) (#2839)

* Myriad: new firmware version (1445)

4 years ago[IE][VPU][GT]: Fixes Eltwise+ReLU merge for dynamic networks (#2777)
Gladilov, Gleb [Fri, 30 Oct 2020 08:42:09 +0000 (11:42 +0300)]
[IE][VPU][GT]: Fixes Eltwise+ReLU merge for dynamic networks (#2777)

Eltwise + ReLU merge is expected to be performed unconditionally
in all cases and since it does not require strides to be defined
could be performed before adjustDataLayout pass.

Unfortunately, there are cases with unexpected degradation after
such a change is introduced. In specific case it seems to be
caused by degradation in HW operation (convolution). It was not
investigated completely and reason is still unknown (convolution
itself remains unchanged in network, but for some reason works
slower).

It has been decided to introduce change only in case of dynamic
models to have performance benefit for some cases and avoid
degradations in others.

Moving mergeEltwiseAndReLU pass before adjustDataLayout for
dynamic cases allows to get additional performance gain due to
lack of extra copy stages introduced in adjustDataLayout.

Signed-off-by: Gladilov, Gleb <gleb.gladilov@intel.com>
4 years agoImprove ngraph python api in debug mode (#2836)
Rafal Blaczkowski [Fri, 30 Oct 2020 08:17:58 +0000 (09:17 +0100)]
Improve ngraph python api in debug mode (#2836)

* Update MacOS machine

* Update setup.py

* Add debug statement

* Improvements

* Update ngraph/python/setup.py

Co-authored-by: Michał Karzyński <4430709+postrational@users.noreply.github.com>
* Rollback the last change

Co-authored-by: Michał Karzyński <4430709+postrational@users.noreply.github.com>
4 years agoRemoved legacy travis script (#2888)
Ilya Churaev [Fri, 30 Oct 2020 07:18:09 +0000 (10:18 +0300)]
Removed legacy travis script (#2888)

* Removed legacy travis script

* Removed legacy variables

4 years ago[IE CLDNN] Fixed compilation error for tutorial samples (#430)
kobeyu [Fri, 30 Oct 2020 07:07:17 +0000 (15:07 +0800)]
[IE CLDNN] Fixed compilation error for tutorial samples (#430)

Co-authored-by: Kobe Yu <kobe.yu@kneron.us>
Co-authored-by: Alexander Zhogov <alexander.zhogov@intel.com>
4 years ago[IE][VPU][GT]: Introduce Split by dynamic dimension check (#2802)
Gladilov, Gleb [Fri, 30 Oct 2020 06:12:11 +0000 (09:12 +0300)]
[IE][VPU][GT]: Introduce Split by dynamic dimension check (#2802)

* [IE][VPU][GT]: Introduce Split by dynamic dimension check

At the moment, myriad plugin does not support split operation
by dynamic axis. To be sure there is no issue with optimized-out
split operation which should have been replaced with copy
stage - assertion before DTS transformation is introduced.

Check should be performed before loop with DTS transformations
because it requires dynamic context (dynamic dimension should be
visible as dynamic), otherwise dynamic dimension would be
replaced with upper-bound estimation and check will always pass.

Signed-off-by: Gladilov, Gleb <gleb.gladilov@intel.com>
* [IE][nGraph]: Fixes normalize_axis symbol exporting

Signed-off-by: Gladilov, Gleb <gleb.gladilov@intel.com>
4 years agoRemove operators with existing functional tests from layer creator (#2772)
Bartosz Lesniewski [Fri, 30 Oct 2020 04:42:27 +0000 (05:42 +0100)]
Remove operators with existing functional tests from layer creator (#2772)

* Remove 49 operators from layer creator

* Remove debug code

* Add specificCreator for deconvolution

* add specificCreator for logical not

* Fix concat creator behavior for negative axis

* add creator for TopK, overrides for visit_attribute methods

* remove layerCreator templates for removed ops

* Disable exception check for Deconvolution and TopK

* add specificCreator for stridedSlice

* add specificCreator for DetecionOutput

* resolve conflict with batchNormInference changes

* fix detection output visit_attributes for some of the bools

* Remove Ceiling from LayerCreator

* Change detectionOutput param validation to expect bool instead of uint

* detectionOutput specificCreator to set bool params as int strings

* Add vector of integer value accessor handling to xml deserializer.

* Fix indentation

4 years agoRemove obsoleted v0::ReplaceSlice (#2887)
Mateusz Tabaka [Fri, 30 Oct 2020 04:37:15 +0000 (05:37 +0100)]
Remove obsoleted v0::ReplaceSlice (#2887)

4 years agoRemove obsoleted v0::Not operator (#2846)
Mateusz Tabaka [Fri, 30 Oct 2020 04:33:26 +0000 (05:33 +0100)]
Remove obsoleted v0::Not operator (#2846)

* Remove obsoleted v0::Not operator

* restore tests from ngraph/test/backend

4 years agoSerialization of experimental and custom ops (#2862)
Jozef Daniecki [Fri, 30 Oct 2020 04:31:27 +0000 (05:31 +0100)]
Serialization of experimental and custom ops (#2862)

* Add IEGeneric node type handling.

* Remove dependency on plugin_api library.

IEGeneric type name is passed via Visiotr API as new syntetic atribute.

* Add custom operations support

* Fix charachter literals comparison.

* Pass custom opsets to CNNNetwork:serialize().

IE extensions are stored in ngraph CNNNetwork and later used to pass
custom opsets to serialization transformation.

* Refactor custom ops tests to use template_extension library.

* Add comment on __generic_ie_type__ purpose.

4 years agoFixed convertFunctionToCNNNetwork to support non unique names (#2864)
Gleb Kazantaev [Fri, 30 Oct 2020 04:24:37 +0000 (07:24 +0300)]
Fixed convertFunctionToCNNNetwork to support non unique names (#2864)

* Unique names normalization during nGraph to CNNNetwork conversion

* Added tests

* Code refactoring

4 years agoUpdate MemCheckTests OMZ part: get_testdata.py and configs (#2722)
Vitaliy Urusovskij [Thu, 29 Oct 2020 21:37:02 +0000 (00:37 +0300)]
Update MemCheckTests OMZ part: get_testdata.py and configs (#2722)

4 years ago[CPU] Enable GatherND tests in ONNX. (#2904)
Nikolay Shchegolev [Thu, 29 Oct 2020 19:53:26 +0000 (22:53 +0300)]
[CPU] Enable GatherND tests in ONNX. (#2904)

4 years ago[IE CLDNN] Added 6d crop support (#2841)
Vladimir Paramuzov [Thu, 29 Oct 2020 18:41:05 +0000 (21:41 +0300)]
[IE CLDNN] Added 6d crop support (#2841)

4 years ago[Python API] Remove IENetLayer class (#2755)
Anastasia Kuporosova [Thu, 29 Oct 2020 18:34:55 +0000 (21:34 +0300)]
[Python API] Remove IENetLayer class (#2755)

* [Python API] Remove IENetLayer class

* fix tests

* fix tests

* remove importing ngraph from several tests

4 years ago[CPU] GatherND implementation. (#2757)
Nikolay Shchegolev [Thu, 29 Oct 2020 16:28:31 +0000 (19:28 +0300)]
[CPU] GatherND implementation. (#2757)

4 years ago[IE CLDNN] Disabled fusing into eltwise node for fsv32 layout (#2844)
Vladimir Paramuzov [Thu, 29 Oct 2020 13:01:53 +0000 (16:01 +0300)]
[IE CLDNN] Disabled fusing into eltwise node for fsv32 layout (#2844)

4 years agoFix visual appearance (#2148)
Dominik Wernberger [Thu, 29 Oct 2020 12:27:33 +0000 (13:27 +0100)]
Fix visual appearance (#2148)

4 years agoImprove support ONNX Resize-10 created by PyTorch (#1350)
Maxim Vafin [Thu, 29 Oct 2020 12:26:23 +0000 (15:26 +0300)]
Improve support ONNX Resize-10 created by PyTorch (#1350)

4 years ago[IE][VPU][TESTS] Fix vpu split with unusable outputs & test (#2718)
Aleksandr Korolev [Thu, 29 Oct 2020 12:12:10 +0000 (15:12 +0300)]
[IE][VPU][TESTS] Fix vpu split with unusable outputs & test (#2718)

* Fix vpu split with unusable outputs & test

Co-authored-by: kora6 <kora6@github.com>
4 years ago[IE][VPU][DTS][Tests]: Remove WAs for networks with multiple outputs (#2828)
Andrew Bakalin [Thu, 29 Oct 2020 10:44:52 +0000 (13:44 +0300)]
[IE][VPU][DTS][Tests]: Remove WAs for networks with multiple outputs (#2828)

* Remove WAs for networks with multiple outputs

4 years agoRemoved forced CMAKE_VERBOSE_MAKEFILE (#2891)
Ilya Lavrenov [Thu, 29 Oct 2020 10:12:01 +0000 (13:12 +0300)]
Removed forced CMAKE_VERBOSE_MAKEFILE (#2891)

4 years agoFixed issue #41619 (#2827)
Ilya Lavrenov [Thu, 29 Oct 2020 09:50:46 +0000 (12:50 +0300)]
Fixed issue #41619 (#2827)

* Fixed issue #41619

* More descriptive error message

4 years agoSplit java CMakeLists.txt into several ones (#2734)
Ilya Lavrenov [Thu, 29 Oct 2020 09:48:03 +0000 (12:48 +0300)]
Split java CMakeLists.txt into several ones (#2734)

* Split java CMakeLists.txt into several ones

* Replaced openjdk to OpenJDK

* Build Java library before tests

4 years agoUpdated apiValidator rules (#2785)
Ilya Lavrenov [Thu, 29 Oct 2020 09:38:15 +0000 (12:38 +0300)]
Updated apiValidator rules (#2785)

* Updated apiValidator rules: added custom filter to emulate BinaryExclusions.xml

* temp

* Temp

* debug log

* Print a message about success

* Fixed file name

4 years agoFix for collision of #1932 and #2731 (#2890)
Konstantin Satunin [Thu, 29 Oct 2020 09:37:55 +0000 (12:37 +0300)]
Fix for collision of #1932 and #2731 (#2890)

4 years agoRevise ReLU (#2863)
Piotr Szmelczynski [Thu, 29 Oct 2020 08:37:52 +0000 (09:37 +0100)]
Revise ReLU (#2863)

* remove relu_backprop

* Update ReLU spec

* change inputs and outputs subsections of ReLU spec

* Update Mathematical Formulation subsection

* Update Category of ReLU in spec

* Update Short description of ReLU in spec

4 years ago[CPU] Add tests for SetBlob + I64 (#2402)
Maxim Andronov [Thu, 29 Oct 2020 08:34:29 +0000 (11:34 +0300)]
[CPU] Add tests for SetBlob + I64 (#2402)

4 years agoRemove obsoleted Min, Max operators (#2832)
Mateusz Tabaka [Thu, 29 Oct 2020 08:32:50 +0000 (09:32 +0100)]
Remove obsoleted Min, Max operators (#2832)

4 years agoRemoved InferenceEngine dependency on Legacy library (#1932)
Ilya Lavrenov [Thu, 29 Oct 2020 07:54:57 +0000 (10:54 +0300)]
Removed InferenceEngine dependency on Legacy library (#1932)

* Split libraries

* Small fixes

* Fixes after self-review

* Minimized legacy usage

* Fixed IE_CPU.onnx_prior_box

* Fixed linkage

* Temp WA for private plugins

* removed comments

* Reverted OneHot changes back

4 years agoONNX Loop (#2847)
Mateusz Bencer [Thu, 29 Oct 2020 06:51:51 +0000 (07:51 +0100)]
ONNX Loop (#2847)

* Loop op ngraph implementation, update IE IR Reader and ngraph to cnn converter

* refactoring SubGraphOp class

* type prop unit tests

* ngraph code style

* update comment

* single layer tests for Loop operation

* fix file name

* Add SpecialBodyPorts attribute in Loop op, update single layer tests

* first debug version

* more tests

* missing test file

* removed not needed shapes from test data

* move test data to new folder

* shape infer tests

* Added execution tests

* add several new tests cases, strict checks in Loop impl, temporary disable single layer tests

* ngraph codestyle, refactoring, clone_new_args test

* resolve review remarks

* fix build

* fix tests

* more execution tests

* add a new constructor of Loop op, resolve review remarks

* execution tests

* synchro with current version

* handle scalars and more tests

* scalar test enabled

* loop reference impl

* bug fixes in tests, onnx importer part and in the ref implementation of the Loop op

* applied remarks

* handle unsupported cases

* rewrite unit tests

* update INTERPRETER manifest

* is_termination_condition_always_true simplification

* [TEST] update python models tests

* review remarks

* added xfail to tiny_yolov3

* missing model test

* revert test data

* fixed numbers of failing tests

* fixed failed test description

* fix test message

* fix xfail test

* zoo models tests clean-up

* missing comma

Co-authored-by: Ivan Tikhonov <ivan.tikhonov@intel.com>
4 years ago[IE CLDNN] Fix accuracy bug in fsv16 imad conv + other minor fixes (#2876)
Jedrzej Hajduczenia [Thu, 29 Oct 2020 06:33:05 +0000 (07:33 +0100)]
[IE CLDNN] Fix accuracy bug in fsv16 imad conv + other minor fixes (#2876)

4 years ago[Transformations] Hswish fusion to be called before Hsigmoid fusion (#2845)
Gorokhov Dmitriy [Thu, 29 Oct 2020 06:07:51 +0000 (09:07 +0300)]
[Transformations]  Hswish fusion to be called before Hsigmoid fusion (#2845)

4 years ago[IE] Add RTTI macro to ReshapeFullyConnectedFusion ngrap pass (#2837)
Zoran Zomborat [Thu, 29 Oct 2020 04:33:55 +0000 (06:33 +0200)]
[IE] Add RTTI macro to ReshapeFullyConnectedFusion ngrap pass (#2837)

4 years agoAdded information about unsupported subgraphs for set affinity (#2872)
Ilya Churaev [Thu, 29 Oct 2020 03:22:25 +0000 (06:22 +0300)]
Added information about unsupported subgraphs for set affinity (#2872)

4 years agoAdd `validate_test_case` fixture with using of `jsonschema`. Specify all required...
Vitaliy Urusovskij [Wed, 28 Oct 2020 21:11:01 +0000 (00:11 +0300)]
Add `validate_test_case` fixture with using of `jsonschema`. Specify all required fields for test cases (#2821)

4 years agosetBatchSize: getting rid of ConstantFolding (#2842)
Evgenya Stepyreva [Wed, 28 Oct 2020 19:49:12 +0000 (22:49 +0300)]
setBatchSize: getting rid of ConstantFolding (#2842)

* setBatchSize: getting rid of setBatchSize

* Trigger CI

* Feedback adressed

* Trigger CI

* f -> specialized_function

4 years agosetupvars.bat: Renamed all Python version related variables. (#2854)
Artyom Anokhov [Wed, 28 Oct 2020 18:02:25 +0000 (21:02 +0300)]
setupvars.bat: Renamed all Python version related variables. (#2854)

4 years ago[ SSR ] Transpose->MatMul (#2525)
Evgenya Stepyreva [Wed, 28 Oct 2020 17:06:42 +0000 (20:06 +0300)]
[ SSR ] Transpose->MatMul (#2525)

* [ SSR ] Reshape->Softmax->Reshape

* Call DepthToSpaceFusion during SmartReshape

* rtti

* remove softmax wa

4 years agoExtended test infrastructure to call per plugin configure function before running...
Bartosz Sochacki [Wed, 28 Oct 2020 11:57:46 +0000 (12:57 +0100)]
Extended test infrastructure to call per plugin configure function before running a test. The function allows to pass plugin specific configuration (#2731)

4 years ago[IE][CMAKE] Fix build issues with custom configuration (#2851)
Vladislav Vinogradov [Wed, 28 Oct 2020 09:20:26 +0000 (12:20 +0300)]
[IE][CMAKE] Fix build issues with custom configuration (#2851)

* FASTER build enabled
* MYRIAD plugin disabled

4 years ago[nGraph] Add opset5 ops to Python nGraph (#2833)
Jan Iwaszkiewicz [Wed, 28 Oct 2020 09:18:14 +0000 (10:18 +0100)]
[nGraph] Add opset5 ops to Python nGraph (#2833)

4 years ago[IE TESTS][IE MKLDNN] Disable CtcLoss SLT due to sporadic failures (#2852)
Irina Efode [Wed, 28 Oct 2020 09:00:11 +0000 (12:00 +0300)]
[IE TESTS][IE MKLDNN] Disable CtcLoss SLT due to sporadic failures (#2852)

4 years ago[IE][VPU]: Improves myriad plugin API (#2816)
Gladilov, Gleb [Wed, 28 Oct 2020 08:14:14 +0000 (11:14 +0300)]
[IE][VPU]: Improves myriad plugin API (#2816)

LoadNetwork takes network argument by constant reference.
Myriad plugin implementation applies transformations to
given network in order to get compiled model.
Transformations take network argument by non-constant
reference, so at some point of time network copy must be
acquired. ICNNNetwork is neither copyable nor movable, so
the only way to get network copy is using special utility
returning std::shared_ptr.

Myriad plugin does not expose any ownership strategy,
so prefers to take network argument by simple reference.
Plugin also requires nGraph -> CNN conversion during
LoadNetwork implementation. Conversion utilities returns
std::shared_ptr, which makes plugin to use workaround for
lifetime of converted object (to have 2 "pointers" to
network: raw pointer to input network and smart pointer to
converted network). Such workarounds make code more
error-prone, because using wrong pointer to semantically
the same object may lead to unexpected results.

To overcome these issues API has been changed in a way to
make interfaces more clear (do not expose ownership strategy
or mutability) and get rid of unnecessary workarounds.

Signed-off-by: Gladilov, Gleb <gleb.gladilov@intel.com>
4 years agoAdd dynamic shape checks for legacy transformations (#2783)
Gleb Kazantaev [Wed, 28 Oct 2020 07:36:16 +0000 (10:36 +0300)]
Add dynamic shape checks for legacy transformations (#2783)

* Added dynamic shape checks for ConvertInterpolate pass

* Added dynamic checks for ConvertLRNToLegacy pass

* Added dynamic checks for ConvertMatMul* pass

* Added dynamic checks for ConvertPadToLegacy pass

* Updated TileIE; added dynamic checks to ConvertTileToLegacy pass

* Added dynamic checks to FCBiasFusion pass

* Added dynamic checks to Reshape1DOps pass

* Added dynamic checks to ReshapeFCFusion pass

* Added dynamic checks to ReshapeFC pass

* Updaed Reshape1DConvolution pattern

4 years ago[IE CLDNN] Removed unused DO gpu impl (#2809)
Vladimir Paramuzov [Wed, 28 Oct 2020 06:24:22 +0000 (09:24 +0300)]
[IE CLDNN] Removed unused DO gpu impl (#2809)

4 years ago[CPU] Generic JIT Eltwise implementation (#1464)
Gorokhov Dmitriy [Wed, 28 Oct 2020 06:16:28 +0000 (09:16 +0300)]
[CPU] Generic JIT Eltwise implementation (#1464)

4 years agoRemove obsoleted v0::OneHot operator (#2855)
Mateusz Tabaka [Wed, 28 Oct 2020 04:16:31 +0000 (05:16 +0100)]
Remove obsoleted v0::OneHot operator (#2855)

4 years agoRemove obsoleted v0::Product op (#2860)
Mateusz Tabaka [Wed, 28 Oct 2020 04:12:52 +0000 (05:12 +0100)]
Remove obsoleted v0::Product op (#2860)

4 years agofix path to code snippets in Custom_ONNX_Ops.md (#2764)
Mateusz Tabaka [Wed, 28 Oct 2020 03:20:52 +0000 (04:20 +0100)]
fix path to code snippets in Custom_ONNX_Ops.md (#2764)

4 years ago[IE][VPU]: Implement HSwish layer with tests (#2775)
Maksim Doronin [Tue, 27 Oct 2020 21:38:10 +0000 (00:38 +0300)]
[IE][VPU]: Implement HSwish layer with tests (#2775)

* Implement HSwish layer with tests
* Disable HSwish decomposition by a predicate
* Update vpu firmware

4 years agoONNX Loop operation support (#2756)
Evgeny Lazarev [Tue, 27 Oct 2020 20:04:43 +0000 (23:04 +0300)]
ONNX Loop operation support (#2756)

* Generate TensorIterator without back edges from TensorFlow models

* Added a check in the MarkSubgraphsWithCorrectLayout to not fail when port is not connected

* Updated the 'protobuf2nx' to consume the graph protobuf message

* Cleanup TI from the IRv7 specific code

* Do not run some front transformations recursively

* Draft support for the ONNX Loop operation when 'cond' = True

* LoopToTI transformation changes

* Added draft of Loop operation and parser for ONNX Loop operation body

* Updated Loop body parser + added shape and type infer for the Loop operation

* Fixes for ONNX Loop operation parser

* Moved Loop parsing to Loop op extractor. Added generation of external edges for the Loop body ops

* Added support for ThresholdedRelu using decomposition

* Added support for Min ONNX operation

* Draft fixes for port_map generation for the Loop

* Rename transformation file and fix BOM

* Fixed shape inference for Loop scan outputs (axis is not None)

* Fixed shape inference for ONNX Loop operation

* Refactor checks in the TensorIteratorMerge transformation

* Code refactoring. Enabled commented transformations

* Documentation update for ONNX Loop, ThresholdedRelu and Min

* Fixed typo in the Loop front transformation where execution condition input is connected. Other refactorings

* Fixed in the Loop extractor

* Added printing 'internal_layer_id' attribute in the graph dumper

* Updated calculation of iterations number for the Loop

* Added missing code

* Fixed output port shapes and types generation for Loop operation

* Update function names and variable names in the Loop operation

* Fixed type inference for iteration count input

* Added removal of input/output ports of the Loop if they are not used

* Fixed renumbering Loop operations input/output ports to keep mandatory

* Fixed ThresholdedReluDecomposition transformation

* Updated MO IR Reader to know about Loop operation. But it is still not supported by the MO IR Reader

* Added unit test for Slice op shape infer (reverse the sequence of elements)

* Reverted changes in the ONNX loader function call to protobuf2nx

* Enable Reshape0DToSqueeze transformation recursively

* Refactored Loop operation support implementation

* Changed ThresholdedReluDecomposition to generate Const with shape [1] instead of scalar

* Code style and wording fixes

* Restored accidentally removed 'return' statement in the TI shape infer function

* Fixed comments

* Fixed comment

Co-authored-by: Evgeny Lazarev <elazarev.nnov@gmail.com>