Gichan Jang [Fri, 12 Mar 2021 01:13:12 +0000 (10:13 +0900)]
[TEST] Add flexbuf test for converter and decoder
Add flexbuf test cases for converter and decoder
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Gichan Jang [Thu, 11 Mar 2021 08:45:48 +0000 (17:45 +0900)]
[Converter] Add flex buffer converter
Add tensor converter subplugin of flexbuffer.
Convert flexbuffer to tensor(s)
Example launch line:
gst-launch-1.0 videotestsrc ! video/x-raw,format=RGB,width=320,height=240 ! tensor_converter ! tensor_decoder mode=flexbuf ! other/flexbuf ! tensor_converter ! tensor_decoder mode=direct_video ! videoconvert ! videoscale ! video/x-raw,width=320,height=240 ! xvimagesink
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Gichan Jang [Thu, 11 Mar 2021 08:00:46 +0000 (17:00 +0900)]
[Decoder] Add flex buffer decoder
Add tensor decoder subplugin offlexbuffer.
Decode tensor(s) to flexbuffer.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Sangjung Woo [Wed, 24 Mar 2021 05:16:20 +0000 (14:16 +0900)]
[Spec] Add configuration for VD build
* Disable NNFW, ArmCL, Viviante, pytorch, caffe2 & tensorflow.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Jaeyun [Thu, 11 Mar 2021 10:12:23 +0000 (19:12 +0900)]
[Transform] option for DC removal
Add new option for DC removal (average value) in stand mode.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Thu, 11 Mar 2021 08:02:03 +0000 (17:02 +0900)]
[Common] tensor data struct
Now tensor-transform and tensor-if use common data struct to handle data.
Add new struct and common functions to handle tensor type and value.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Yongjoo Ahn [Thu, 11 Mar 2021 05:28:14 +0000 (14:28 +0900)]
[API] Remove API directory
- Finish migration of API related code to ML API repo (github.com/nnstreamer/api)
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Yongjoo Ahn [Thu, 11 Mar 2021 05:25:03 +0000 (14:25 +0900)]
[Doc] Fix Android build guide after ML API repo
- Fix Android library build guide for ML API with nnstreamer
- REF: https://github.com/nnstreamer/api/java
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Yongjoo Ahn [Mon, 8 Mar 2021 09:40:56 +0000 (18:40 +0900)]
[api/Android] Remove api related declarations in nnstreamer.mk
- Remove `NNSTREAMER_CAPI_*` from jni/nnstreamer.mk
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Xavier Roumegue [Thu, 18 Mar 2021 18:04:48 +0000 (18:04 +0000)]
[Filter/Python3] Include <stdexcept> for std::runtime_error support
This fixes the following build issue:
error: 'runtime_error' is not a member of 'std'
| 192 | throw std::runtime_error (dlerror ());
https://en.cppreference.com/w/cpp/error/runtime_error
Signed-off-by: Xavier Roumegue <xavier.roumegue@nxp.com>
MyungJoo Ham [Sun, 28 Feb 2021 13:01:16 +0000 (22:01 +0900)]
[Filter/Lua] Design of LUA script rules for nnstreamer.
Provided comments on how to write nnstreamer filters with LUA.
Related with #3095
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Sangjung Woo [Wed, 17 Mar 2021 10:05:38 +0000 (19:05 +0900)]
[Spec] Fix the buildbreak issue for VD
* Disable OpenVino
* Omit the test case execution & testcase summary
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Jihoon Lee [Wed, 17 Mar 2021 23:28:34 +0000 (08:28 +0900)]
[Fix] Add missing file section in rpm build
As nnstreamer mata package was not containing `files` section, it led to
the recent SR fail, this patch resolves the issue.
Also add %config directive to the configuration file
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: Jihoon Lee <jhoon.it.lee@samsung.com>
Gichan Jang [Tue, 9 Mar 2021 00:38:23 +0000 (09:38 +0900)]
[Doc] Change hotdoc generation script
Change hotdoc generation script.
This change will prevent broken link in github.io
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Jaeyun [Fri, 12 Mar 2021 10:18:12 +0000 (19:18 +0900)]
[Transform] handle empty option string case
If option string is empty, tensor-transform cannot parse the mode and will crash application.
Add error case for empty option string.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Xavier Roumegue [Fri, 12 Mar 2021 16:47:59 +0000 (16:47 +0000)]
[Decoder/Pose] Add support for position offset output tensor
The decodeur can change its decoding scheme thanks to option4.
The possible mode are:
- 'heatmap-only' (default):
* The confidence score is within [0:1] range.
* The keypoint position can be directly used.
- 'heatmap-offset
* This decoding scheme is applicable for posenet:
https://www.tensorflow.org/lite/examples/pose_estimation/overview
* Sigmoid has to be applied to get condidence score
* Additional offset tensor has to be considered to map a keypoint
in the NN input resolution range prior to downscale to heatmap
grid resolution size.
Example for posenet_resnet50_16_513 NN (Same Input/Output resolution) :
... ! tensor_decoder mode=pose_estimation \
option1=513:513 option2=513:513 \
option3=key_point_labels.txt option4=heatmap-offset ! ...
Signed-off-by: Xavier Roumegue <xavier.roumegue@nxp.com>
Xavier Roumegue [Thu, 11 Mar 2021 20:05:01 +0000 (20:05 +0000)]
[Decoder/Pose] Get key body connections from optional file
The body connections are added to each keypoint line in the
optional key body listing file.
A space separated number list is expected to describe other keypoints
connected to the current's one.
For example, default pose decoder configuration could be described
as the following:
top 1
neck 0 2 5 8 11
r_shoulder 1 3
r_elbow 2 4
r_wrist 3 1
l_shoulder 1 6
l_elbow 5 7
l_wrist 6
r_hip 1 9
r_knee 8 10
r_ankle 9
l_hip 1 12
l_knee 11 13
l_ankle 12
Signed-off-by: Xavier Roumegue <xavier.roumegue@nxp.com>
Jihoon Lee [Mon, 15 Mar 2021 11:11:20 +0000 (20:11 +0900)]
[Package] Seperate nnstreamer into core and conf
This patch separates nnstreamer into nnstreamer-core and nnstreamer-configuration.
From this patch, nnstreamer becomes a meta package that requires
`nnstreamer-core` and `nnstreamer-configuration`.
**Evaluation Method**
gbs and pdebuild
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: Jihoon Lee <jhoon.it.lee@samsung.com>
Cc: Myungjoo Ham <myungjoo.ham@samsung.com>
Jihoon Lee [Thu, 11 Mar 2021 09:44:24 +0000 (18:44 +0900)]
[Doc] Add documentation of using test-devel
Add documentation of how to utilize test-devel package in the
perspective of subplugin developer
**Self evaluation:**
1. Build test: [ ]Passed [ ]Failed [X]Skipped
2. Run test: [ ]Passed [ ]Failed [X]Skipped
Signed-off-by: Jihoon Lee <jhoon.it.lee@samsung.com>
Jihoon Lee [Thu, 11 Mar 2021 09:15:49 +0000 (18:15 +0900)]
[Packaging] Add test-devel to rpm/dpkg
Add test-devel to rpm/dpkg
This patch...
1. offers `nnstreamer-template.ini.in` from test-devel package
1. offers `subplugin_unittest_template.cc.in` from test-devel package
1. for rpm test-devel, it renames /etc/nnstreamer.ini until test-devel
is unintalled
1. add pkg-config variable to locate test related templates
Resolves #3124
Tested using gbs and pdebuild
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: Jihoon Lee <jhoon.it.lee@samsung.com>
Gichan Jang [Fri, 12 Mar 2021 04:47:03 +0000 (13:47 +0900)]
[TEST] Test protobuf and flatbuf support tensors
Test whether flatbuffer and protobuf support other/tensors.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Jaeyun [Wed, 10 Mar 2021 10:10:59 +0000 (19:10 +0900)]
[Transform] tensor type in stand mode
Currently only float32 can use stand mode in tensor-transform.
Add option to set output type and calc stand mode.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Juyeong Lee [Fri, 12 Mar 2021 15:11:41 +0000 (00:11 +0900)]
[Test] Fix typo
Fix typo made in #3130
Signed-off-by: Juyeong Lee <2jy22@naver.com>
Jaeyun [Fri, 12 Mar 2021 07:05:43 +0000 (16:05 +0900)]
[Decoder] file content size
When loading data from file, use param to get the contents size.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Xavier Roumegue [Wed, 10 Mar 2021 12:56:56 +0000 (12:56 +0000)]
[tests/decoder_pose] Remove dummy option3 setting from command line
Option3 is now used to specify the labels file.
Signed-off-by: Xavier Roumegue <xavier.roumegue@nxp.com>
Xavier Roumegue [Tue, 9 Mar 2021 12:28:56 +0000 (12:28 +0000)]
[Decoder/Pose] Add support for optional pose label file
This adds an optional option (3), specifying the pose labels file path.
A label per line is expected, line number specifying the pose id.
If this option is not selected, legacy label mapping is used as fallback.
Signed-off-by: Xavier Roumegue <xavier.roumegue@nxp.com>
Xavier Roumegue [Wed, 10 Mar 2021 16:41:00 +0000 (16:41 +0000)]
[Decoder/Pose] Run gst-indent
Signed-off-by: Xavier Roumegue <xavier.roumegue@nxp.com>
Xavier Roumegue [Tue, 9 Mar 2021 19:46:13 +0000 (20:46 +0100)]
[Decoder/Utils] Remove last empty label while enumerating labels from file
While parsing the labels file, a label per line is expected.
g_strsplit(file_contents, '\n', -1) is used to parse the file and
returns the label as a string vector.
The last '\n' in the file introduces an additional empty element, as the
delimiter involves 2 elements.
Hence, the patch is removing the last trailing '\n' to avoid this
additional dummy element in the labels list.
Signed-off-by: Xavier Roumegue <xavier.roumegue@nxp.com>
Xavier Roumegue [Wed, 10 Mar 2021 16:37:24 +0000 (16:37 +0000)]
[Decoder/Utils] Run gst-indent on tensordecutil.c
Signed-off-by: Xavier Roumegue <xavier.roumegue@nxp.com>
Juyeong Lee [Thu, 11 Mar 2021 20:53:06 +0000 (05:53 +0900)]
Remove underscore in test names
Resolves #2541
Update test suit names and test names not to contain underscore.
This follows gtest recommendations.
https://github.com/google/googletest/blob/master/docs/faq.md#why-should-test-suite-names-and-test-names-not-contain-underscore
Signed-off-by: Juyeong Lee <2jy22@naver.com>
Dongju Chae [Fri, 12 Mar 2021 01:38:55 +0000 (10:38 +0900)]
[Coverity] Fix coverity issues (CID 314892, 314893)
This patch fixes coverity issues (CID 314892, 314893)
- UNINTENDED_INTEGER_DIVISION
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Jaeyun [Mon, 8 Mar 2021 06:11:14 +0000 (15:11 +0900)]
[Filter/TF-lite] tensor name in meta
Set tensor name in tensors info struct.
TODO: we should update testcases in api repo later.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
hyeonseok lee [Wed, 3 Mar 2021 09:52:17 +0000 (18:52 +0900)]
Fix md file link
1. Fix all the md files in
$NNST_ROOT/*.md
$NNST_ROOT/Documentation/*.md
$NNST_ROOT/tools/*md
which has a link to other md files
Signed-off-by: hyeonseok lee <hs89.lee@samsung.com>
MyungJoo Ham [Wed, 3 Mar 2021 10:44:32 +0000 (19:44 +0900)]
[Edge/MQTT] Manifest for MQTT implementation
This describes more specific for #3104.
The assignees of #3104 should refer to this document for implementation.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Dongju Chae [Thu, 4 Mar 2021 06:17:15 +0000 (15:17 +0900)]
[Python] Change the logic to find the python library name
This patch changes the logic to find the python library name.
The library name depends on which options were used to build python.
For example, if python 3.7 was compiled with '--with-pymalloc',
the library name is 'libpython3.7m.so'
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Jaeyun [Wed, 3 Mar 2021 11:30:13 +0000 (20:30 +0900)]
[If] fix typo
Fix typo, property name of tensor-if element.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Dongju Chae [Thu, 4 Mar 2021 01:51:27 +0000 (10:51 +0900)]
Revert "[Debian] Temporally disable caffe2/tensorflow"
This reverts commit
dd053adff98fadd339eb9a5f50078bdcae73d7f7.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Jaeyun [Wed, 3 Mar 2021 11:46:37 +0000 (20:46 +0900)]
[Build] fix install path
fix build error, change unittest install path.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Wook Song [Wed, 24 Feb 2021 09:57:41 +0000 (18:57 +0900)]
[Meson] Revise build dependency checking mechanism for MvNCSDK2
The existing build dependency checking mechanism for MvNCSDK2 is
unnecessarily complex. This patch revises it.
See also: https://review.tizen.org/gerrit/#/c/platform/adaptation/npu/intel-libmvnc/+/254171/
Reported-by: Jaeyun Jung <jy1210.jung@samsung.com>
Signed-off-by: Wook Song <wook16.song@samsung.com>
Jaeyun [Fri, 19 Feb 2021 07:53:08 +0000 (16:53 +0900)]
[Test] unittest install dir
change unittest install dir (libdir to bindir) and update related build script.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Juyeong Lee [Thu, 25 Feb 2021 18:32:01 +0000 (03:32 +0900)]
[Docs] Fix broken links in contributing guide
This patch updates broken links in contributing guide.
Signed-off-by: Juyeong Lee <2jy22@naver.com>
MyungJoo Ham [Thu, 25 Feb 2021 00:51:47 +0000 (09:51 +0900)]
[Dist/RPM] Fix meson path options.
With the side-effects of meon join_paths,
absolute paths of bindir/libdir is working,
this is not recommended.
Fix the path inputs; use relative paths along with
prefix.
Fixes #2642
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Dongju Chae [Tue, 2 Mar 2021 06:23:21 +0000 (06:23 +0000)]
[Debian] Temporally disable caffe2/tensorflow
This patch temporally disables caffe2/tensorflow.
They are not compatible with protobuf 3.14.0, and need the corresponding patches. This will be reverted soon.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
MyungJoo Ham [Wed, 24 Feb 2021 10:13:40 +0000 (19:13 +0900)]
[Documentation] List up supported features for binary releases
For binary release list up supported features.
Fixes #1448
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Dongju Chae [Thu, 25 Feb 2021 01:53:47 +0000 (10:53 +0900)]
[Fix/Meson] Find the dependency using alternative name
This patch checks the dependency using alternative name even with 'enabled' option.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
MyungJoo Ham [Thu, 25 Feb 2021 04:08:09 +0000 (13:08 +0900)]
[Filter/Lua] Skeleton code for LUA filter subplugin
This is skeleton code for LUA filter subplugin, which
shows the idea of implementation and usage patterns.
This starts addressing #3095 (backlog of #3087)
Anyone may jump in and start implementing the contents.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Dongju Chae [Mon, 22 Feb 2021 03:35:07 +0000 (12:35 +0900)]
[meson] supports alternative target name to find pkg dependency
This patch supports alternative target name to find pkg dependency.
For example, in case of armnn, the dependency name should be
'Armnn' not 'armnn' to find its cmake config files.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Dongju Chae [Tue, 23 Feb 2021 01:50:57 +0000 (10:50 +0900)]
Revert "[gRPC/Test] Add gRPC unittests using asynchronous APIs"
This reverts commit
3cafaa57660a652bf45eaadfeb4ad44d75413a81.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Dongju Chae [Thu, 24 Dec 2020 06:23:47 +0000 (15:23 +0900)]
[gRPC/Test] Add gRPC unittests using asynchronous APIs
This patch adds gRPC unittests using asynchronous APIs.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Dongju Chae [Tue, 22 Dec 2020 09:45:13 +0000 (18:45 +0900)]
[gRPC/Async] Extend gRPC codes to support async/non-blocking APIs
This patch extends gRPC codes to support async/non-blocking APIs.
Let's discuss about how we can handle the tag of each RPC request
in the gstreamer pipeline.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Jaeyun [Thu, 18 Feb 2021 06:55:48 +0000 (15:55 +0900)]
[C-Api] remove script to build c-api
Remove build option for c-api and related testcases.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Hyoung Joo Ahn [Fri, 19 Feb 2021 08:25:44 +0000 (17:25 +0900)]
[LOG] unifying the unit of time checker
To make it simple, the value is changed to `ms`.
Signed-off-by: Hyoung Joo Ahn <hello.ahn@samsung.com>
Hyoung Joo Ahn [Fri, 19 Feb 2021 04:09:18 +0000 (13:09 +0900)]
[NNFW] add a timer and DBG flag
The `DBG` flag and log messages are added to check the time of `invoke` function
Signed-off-by: Hyoung Joo Ahn <hello.ahn@samsung.com>
MyungJoo Ham [Mon, 22 Feb 2021 10:32:04 +0000 (19:32 +0900)]
[Meson] Place main library dep at the first place of deps
1. Guarantee the main library's position of deps.
2. Print out tf/tf-lite version number during meson.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Mon, 22 Feb 2021 08:07:41 +0000 (17:07 +0900)]
[TF 2.4] TF C-API breaking changes at 2.4
Because of breaking changes of TF 2.4
https://github.com/tensorflow/tensorflow/blob/
e43be76009614be88454d2fdf2fe702acc5bab77/RELEASE.md#breaking-changes-1
, we need to use different C-API usages for TensorFlow.
The corresponding change is
> C-API functions TF_StringDecode, TF_StringEncode, and TF_StringEncodedSize are no longer relevant and have been removed; see core/platform/ctstring.h for string access/modification in C.
Fixing #3081
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jihoon Lee [Wed, 17 Feb 2021 08:59:57 +0000 (17:59 +0900)]
[Fix/test] fix unittest fail when no tflite
When testing auto option, *.tflite is used so nnstreamer tries to find
tflite in the framework. However when tflite is disabled, there is no
tflite support leading the test to segfault (tested in ubuntu 18.04)
This patch resolves the issue
v2. Fixed tizen_capi plugin. Added a convinient macro
`TEST_REQUIRE_TFLITE` to exploit `DISABLED_tcname` feature in gtest
v3. adding tizen_capi fix is reverted as tizen_capi is migrating
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: Jihoon Lee <jhoon.it.lee@samsung.com>
Jaeyun [Tue, 16 Feb 2021 06:51:40 +0000 (15:51 +0900)]
[Build] disable c-api
disable c-api build.
TODO: we should remove api directory and related testcases later.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Hyoung Joo Ahn [Wed, 17 Feb 2021 11:14:20 +0000 (20:14 +0900)]
[TFLite] fix a minor bug
during the DBG mode, the `g_critical` generate an error. Furthermore, when there are multiple models, the invoke time should be distinguished.
Signed-off-by: Hyoung Joo Ahn <hello.ahn@samsung.com>
Jaeyun [Wed, 17 Feb 2021 09:15:57 +0000 (18:15 +0900)]
[Build/Dev] internal header in cpp
add extern for cpp use in internal dev header.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Wed, 17 Feb 2021 04:55:36 +0000 (13:55 +0900)]
[Test] remove dependency to ml api
For tizen sensor element, remove dependency to ml api. Replace to gst functions.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Dongju Chae [Tue, 16 Feb 2021 08:34:16 +0000 (17:34 +0900)]
[gRPC] remove uncessary dependency declaration
This patch removes unnecessary dependency declaration.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Dongju Chae [Tue, 16 Feb 2021 07:50:30 +0000 (16:50 +0900)]
Revert "[grpc] Temporral disable grpc package"
This reverts commit
f563d989d6b26ce097574c21c6c1e6185b189614.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Gichan Jang [Mon, 15 Feb 2021 01:58:32 +0000 (10:58 +0900)]
[TEST/common] Fix memory leak
Fix memory leak of the tensor common unit test.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Gichan Jang [Mon, 15 Feb 2021 06:29:00 +0000 (15:29 +0900)]
[CAPI] Change comment to comply convention
Change comment for nntrainer enum to comply convention.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Gichan Jang [Wed, 10 Feb 2021 06:52:24 +0000 (15:52 +0900)]
[Bug] Fix memory leak
Fix memory leak detected by valgrind.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Gichan Jang [Wed, 10 Feb 2021 06:58:27 +0000 (15:58 +0900)]
[TEST/sink] Change free function
Some tests do not remove and free temporary files.
Change to remove and free temporary files from `_free_test_data`.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
MyungJoo Ham [Tue, 9 Feb 2021 15:35:14 +0000 (00:35 +0900)]
[OSX] Build error fix
OSX does not have sys/auxv.h
Add a macro assuming that SIMD is always available for
OSX-AARCH64 (e.g., M1) devices.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jaeyun [Thu, 4 Feb 2021 07:42:01 +0000 (16:42 +0900)]
[Build] remove unnecessary option
Remove unnecessary build option for sub-plugins, default val is auto.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Gichan Jang [Tue, 9 Feb 2021 10:58:21 +0000 (19:58 +0900)]
[TEST/single] Destory output data handle
After #3002, User should destory the output handle.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Jaeyun [Tue, 9 Feb 2021 08:07:35 +0000 (17:07 +0900)]
[C-Api/Tizen] restricted elements for internal function
Do not check restricted list when constructing a pipeline with internal API.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Gichan Jang [Tue, 9 Feb 2021 08:04:30 +0000 (17:04 +0900)]
[grpc] Temporral disable grpc package
The grpc version is upgraded, but the grpc package of the nnstreamer ppa is not upgraded properly, so it is temporarily disabled.
Discussed with dongju-chae
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Dongju Chae [Mon, 8 Feb 2021 05:55:24 +0000 (14:55 +0900)]
[gRPC/Test] Skip SSAT tests on qemu/aarch64
This patch makes gbs skip ssat tests on qemu/aarch64 env.
In gRPC 1.35.0, some behaviors are not supported on qemu/aarch64.
See the below error log
```
[ 119s] qemu: Unsupported syscall: 168
[ 119s] E0208 04:45:45.
854670498 21089 cpu_linux.cc:41] Error determining current CPU: Function not implemented
[ 119s]
[ 119s] Unsupported setsockopt level=1 optname=15
[ 119s] E0208 04:45:45.
874942631 21089 socket_utils_common_posix.cc:222] check for SO_REUSEPORT: {"created":"@
1612759545.
874068306","description":"Protocol not available","errno":92,"file":"/home/abuild/rpmbuild/BUILD/grpc-1.35.0/src/core/lib/iomgr/socket_utils_common_posix.cc","file_line":198,"os_error":"Protocol not available","syscall":"setsockopt(SO_REUSEPORT)"}
[ 119s] qemu: uncaught target signal 4 (Illegal instruction) - core dumped
```
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Jaeyun [Thu, 4 Feb 2021 07:42:01 +0000 (16:42 +0900)]
[Build] remove unnecessary option
Remove unnecessary build option for sub-plugins, default val is auto.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Jaeyun [Fri, 5 Feb 2021 08:35:46 +0000 (17:35 +0900)]
[CodeClean] fix indent
fix indent in c-api source.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Yongjoo Ahn [Mon, 18 Jan 2021 07:35:18 +0000 (16:35 +0900)]
[test/SSAT] Check flatbuf and protobuf decoder sub-plugins
- Check existence of flatbuf and protobuf sub-plugins for decoder SSAT test
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Jaeyun [Fri, 5 Feb 2021 09:07:35 +0000 (18:07 +0900)]
[Build] fix package name
fix package name required for devel packages.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Geunsik Lim [Thu, 4 Feb 2021 01:51:11 +0000 (10:51 +0900)]
TAOS-CI: Use .TAOS-CI/.gbs.conf by default
This commit is to use a (1) ./.TAOS-CI/.gbs.conf or (2) ./packaging/.gbs.conf file
by default to support customized GBS repositories for each GitHub repository
because the (3) ~/.gbs.conf file of user account (e.g., www-data) can be used among
lots of repositories.
**Changelog**
* Added ./.TAOS-CI/.gbs.conf
* Added ./packaging/.gbs.conf
* Replaced https with http to avoid ssl issue due to firewall
Signed-off-by: Geunsik Lim <geunsik.lim@samsung.com>
Gichan Jang [Wed, 3 Feb 2021 05:01:15 +0000 (14:01 +0900)]
[TEST] Code clean unittest plugins
- It causes confusion by performing free resource in the macros.
Changed to release the resource in same code block.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Gichan Jang [Tue, 2 Feb 2021 07:14:58 +0000 (16:14 +0900)]
[TEST] Add tensor filter fw autodetection unit test
Add tensor filter framework autodetection unit tests
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Gichan Jang [Tue, 2 Feb 2021 06:54:43 +0000 (15:54 +0900)]
[Filter] Framework auto detection without specifying the option
Modify the tensor filter to automatically detect the framework without specifying the framework option.
Default value of the framework is changed to "auto".
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Jaeyun [Fri, 29 Jan 2021 06:25:13 +0000 (15:25 +0900)]
[Test] separate nnfw testcase
separate testcases for nnfw plugin (functions for filter-plugin and ml-api)
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Gichan Jang [Wed, 27 Jan 2021 04:48:16 +0000 (13:48 +0900)]
[SSAT] Remove some log files generated by SSAT tests
- Remove some log files during SSAT test for humble yocto qemu environment
Signed-off-by: Yongjoo Ahn <yongjoo1.ahn@samsung.com>
Hyoung Joo Ahn [Thu, 4 Feb 2021 03:02:27 +0000 (12:02 +0900)]
[ML-API] add log messages
The additional log messages are required because returning the error code `ML_ERROR_STREAMS_PIPE` gives very little information for the abnormal situation.
Signed-off-by: Hyoung Joo Ahn <hello.ahn@samsung.com>
Gichan Jang [Wed, 27 Jan 2021 04:48:16 +0000 (13:48 +0900)]
[Doc] Move stable wiki doc to /Documentation
Copy stable wiki documents to /Documentation.
After this PR, copied wiki page will be deleted.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Gichan Jang [Mon, 1 Feb 2021 10:37:19 +0000 (19:37 +0900)]
[DOC] Add a description of the combination option
Add a description of the comparison between tee and input-combination.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Gichan Jang [Tue, 2 Feb 2021 07:47:00 +0000 (16:47 +0900)]
[Doc] Add nnstreamer tools to github.io
Add nnstreamer tools doc to github.io
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Gichan Jang [Tue, 2 Feb 2021 00:23:31 +0000 (09:23 +0900)]
[Doc] Add nnstreamer logo to github.io
Add NNStreamer logo to github.io.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Jaeyun [Fri, 29 Jan 2021 07:57:38 +0000 (16:57 +0900)]
[Build] internal header for APIs
Add internal header for sub-plugins and native APIs.
After separating API repo, native API build will include this header.
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
Gichan Jang [Mon, 1 Feb 2021 08:40:05 +0000 (17:40 +0900)]
[TEST] Change string property test
Current setting and getting string property test try to select tensor(s) that does not exist.
Changed to set and get framework of the tensor filter.
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Xavier Roumegue [Tue, 26 Jan 2021 17:15:07 +0000 (18:15 +0100)]
[Decoder/BoundingBox] Add option for detection threshold on tf-ssd reported scores
This limits boxes drawing to objects detected with reasonable confidence,
specifying the threshold in percentage thanks to the option-3 as below:
option-3=bbox idx:classes idx:scores idx:num detection idx,threshold
So this adds ",threshold" to the current option-3 configuration scheme.
The threshold has to be within [0 100] range, otherwise skipped.
Signed-off-by: Xavier Roumegue <xavier.roumegue@nxp.com>
Xavier Roumegue [Thu, 28 Jan 2021 14:12:54 +0000 (15:12 +0100)]
[Decoder/BoundingBox] Fix off-by-one error on valid TF-SSD labels handling
The class label indexes starts at 0 while parsing the labels file,
and hence 0 should be considered as a valid label in the general case.
Signed-off-by: Xavier Roumegue <xavier.roumegue@nxp.com>
Xavier Roumegue [Thu, 28 Jan 2021 14:17:40 +0000 (15:17 +0100)]
[Decoder/BoundingBox] Clamp bounding boxes locations
Some neural networks could return negative values, which can lead to
segfault errors while drawing the boxes.
For example, the model below outputs some locations boxes with values
out of [0.0 1.0] in some circumstances.
http://download.tensorflow.org/models/object_detection/ssd_mobilenet_v2_quantized_300x300_coco_2019_01_03.tar.gz
Signed-off-by: Xavier Roumegue <xavier.roumegue@nxp.com>
Xavier Roumegue [Fri, 22 Jan 2021 19:46:02 +0000 (19:46 +0000)]
[Decoder/BoundingBox] Add option to specify TF-SSD output tensors mapping
Recent tensor flow object detections neural network models use output
tensor mapping defined by google:
https://www.tensorflow.org/lite/models/object_detection/overview#output_signature
which is different than the one currently used on bounding boxes tensor
decoder element.
This patch adds the possibility to specify any output tensors mapping
through the option-3 parameter for tf-ssd mode using the following
convention:
locations index:classes index: scores index: number of detections index.
For example, option3=0:1:2:3 would map the output tensors as defined by
previous link, .i.e. :
- Bounding boxes locations tensor on index 0
- Classes tensor on index 1
- Scores tensor on index 2
- Number of detection tensor on index 3
option3 for tf-ssd mode is optional, and if not specified, the default
tensors mapping preserve the backward compatibility (.i.e. 3:1:2:0)
Signed-off-by: Xavier Roumegue <xavier.roumegue@nxp.com>
Xavier Roumegue [Thu, 28 Jan 2021 16:26:15 +0000 (17:26 +0100)]
[Decoder/BoundingBox] Apply gst-indent
Signed-off-by: Xavier Roumegue <xavier.roumegue@nxp.com>
Gichan Jang [Tue, 26 Jan 2021 02:15:07 +0000 (11:15 +0900)]
[C-API] Fix C-API header comments
Clarify the description of nnstreamer C-API comments
Signed-off-by: Gichan Jang <gichan2.jang@samsung.com>
Jihoon Lee [Wed, 20 Jan 2021 07:47:05 +0000 (16:47 +0900)]
[pkg-config] Update require to capi-nnstreamer
As users of capi-nnstreamer.pc need `capi-ml-common.pc`, it should be
added to the require although currently, capi-ml-common.pc has the
exactly same include path so nothing really changes.
This is more of a future proof that one day capi-ml-common.pc might
point to a different include header.
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: Jihoon Lee <jhoon.it.lee@samsung.com>
Jaeyun [Fri, 29 Jan 2021 09:43:12 +0000 (18:43 +0900)]
[Build] package for nnstreamer API
add package for nnstreamer API (separate from nnstreamer pkg)
Signed-off-by: Jaeyun <jy1210.jung@samsung.com>
MyungJoo Ham [Fri, 11 Dec 2020 06:36:14 +0000 (15:36 +0900)]
[Tools/Parser] Rule Implementation Done
Redefine "GstBin" and "GstElement" for nns-parser.
As we do not create actual pipelines from strings,
we only need symbolic data.
All the parsing rule implementations are fixed for nns-parser.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Mon, 30 Nov 2020 07:04:04 +0000 (16:04 +0900)]
[Tools/Parser] Remove GSTPARSE functions
GSTPARSE has defined its own functions,
which are replaced with general functions
by this commit.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Tue, 1 Dec 2020 08:04:29 +0000 (17:04 +0900)]
[Tools/Parser] Do ref counting with pseudo elements
With linked list, psuedo elements are required to be ref-counted.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>