MyungJoo Ham [Wed, 14 Aug 2019 04:39:13 +0000 (13:39 +0900)]
Release of 0.3.0
Please do SR with this.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Wed, 4 Sep 2019 07:21:28 +0000 (16:21 +0900)]
[Test] Prevent unit-test errors in slow systems
In slow systems such as QEMU-ARM, a few test cases fail often.
- Add dynamic timeout for pipeline start
- Add mutex for sink callbacks
Changes in v2:
- Loop of wait_for_start runs if it's READY || PAUSED
- Wait 50us right after start() before state-checking.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jaeyun [Tue, 3 Sep 2019 07:08:12 +0000 (16:08 +0900)]
[Android] update script to remove warning
Clean up build script and remove warning about duplicates.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Parichay Kapoor [Fri, 30 Aug 2019 15:04:24 +0000 (00:04 +0900)]
[filter_single] Common functions between tensor_filter and tensor_filter_single
Simplified common functions between tensor_filter and tensor_filter_single
Added more implementations in tensor_filter_single
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Thu, 29 Aug 2019 10:51:50 +0000 (19:51 +0900)]
[filter_single] Added tensor_filter_single prototype
Added prototype for tensor_filter_single
tensor_filter_single is going to form the backend for singleshot api
to minimize the latency. This is made to be independent of GStreamer,
supporting all the existing tensor_filter plugins
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jaeyun [Tue, 3 Sep 2019 12:47:33 +0000 (21:47 +0900)]
[Filter/Transform] code clean
1. remove unnecessary code to get output size in tensor-filter.
2. set null to find operator in tensor-transform.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Tue, 3 Sep 2019 10:19:02 +0000 (19:19 +0900)]
[Api/Single] invalid condition to check model
1. validate model path before checking the file extension.
2. set pre-defined so extension for custom-filter.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
MyungJoo Ham [Tue, 3 Sep 2019 05:37:16 +0000 (14:37 +0900)]
[TF-Lite] Add terminating null
The string list requires a terminating null.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Dongju Chae [Mon, 2 Sep 2019 06:48:33 +0000 (15:48 +0900)]
[TESTHUB] Append 'gtest_output' for reporting unittest results to TESTHUB.
This commit appends 'gtest_output' into .spec file to generate .xml gtest output files.
These xml files are reported to TESTHUB (for daily x86_64 build).
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
Sangjung Woo [Mon, 2 Sep 2019 06:21:42 +0000 (15:21 +0900)]
[Filter] Fix svace issue
If the parameter hw of constructor is NNAPI_UNKNOWN, the member variable
accel is not initialized. This patch fixes that bug.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Jaeyun [Fri, 30 Aug 2019 07:09:06 +0000 (16:09 +0900)]
[Doc] fix invalid extension
fix invalid file extension in transform header
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Thu, 29 Aug 2019 06:56:29 +0000 (15:56 +0900)]
[Build] update build script
Fix duplicated and unnecessary code with the updation of nnstreamer features.
1. remove src-iio in android-jni (currently excluded to register the element, so unnecessary.)
2. remove option enable-nnfw in .spec (default false)
3. duplicated feature definition for tizen
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Fri, 2 Aug 2019 10:57:25 +0000 (19:57 +0900)]
[Api/Single] set timeout
Add new function to set the timeout to wait for the output.
The function try_pull is supported from gst 1.10, check gst version and return error code if is not supported.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Fri, 2 Aug 2019 07:24:44 +0000 (16:24 +0900)]
[Api/Tizen] function to get data size
Current function returns total byte size of info handle.
Change this to get the byte size with data index.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Sangjung Woo [Thu, 29 Aug 2019 11:22:27 +0000 (20:22 +0900)]
[Tizen/API] Fix the wrong boilerplate text
This patch fixes the wrong boilerplate text. (Apache -> LGPL-2.1)
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Jaeyun [Thu, 29 Aug 2019 07:07:58 +0000 (16:07 +0900)]
[Filter/Decoder] sub-plugins extension
Check sub-plugins file extension after supporting mac.
Wook defined the file extension for build target. With this, set sub-plugins name in filter/decoder.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
jijoong.moon [Wed, 14 Aug 2019 06:24:55 +0000 (15:24 +0900)]
[FILTER] add nnapi property to handle
Add nnapi property in the tensor_filter element.
Use nnapi=(true|false)(:(cpu|gpu|npu)) option for tensor_filter
element. For example,
gst-launch-1.0 --gst-plugin-path=${PATH_TO_PLUGIN} filesrc
location=${PATH_TO_IMAGE} ! pngdec ! videoscale ! imagefreeze !
videoconvert ! video/x-raw,format=RGB,framerate=0/1 ! tensor_converter
! tensor_filter framework=tensorflow-lite model=${PATH_TO_MODEL} nnapi=true:gpu ! filesink location=tensorfilter.out.log
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
Parichay Kapoor [Wed, 28 Aug 2019 08:02:22 +0000 (17:02 +0900)]
[filter plugins] Remove dependency on gst headers
Remove dependency of tensor_filter plugins and tensor_subplugin on gst headers
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Wook Song [Wed, 28 Aug 2019 04:20:13 +0000 (13:20 +0900)]
[Tests/Common] Use macro as file extension of shared object
This patch replaces '.so' with a macro that is defined as file
externsion of shared object according to the platform.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Tue, 27 Aug 2019 03:52:16 +0000 (12:52 +0900)]
[macOS/Conf] Support .dylib as a file extension of sub-plugins
This patch adds .dylib as a file extension of sub-plugins to support.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Tue, 27 Aug 2019 02:18:10 +0000 (11:18 +0900)]
[macOS/Tests] Let unit test cases for tensor_sink work on macOS
This patch updates unit test cases for tensor_sink to make them work on
macOS.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Mon, 26 Aug 2019 07:27:38 +0000 (16:27 +0900)]
[macOS/Meson/IIO] Exclude tensor source plugin for Linux IIO
Linux IIO is not compatible with macOS so that this patch excludes
tensor source plugin for Linux IIO from build tree.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Mon, 26 Aug 2019 02:29:28 +0000 (11:29 +0900)]
[Meson] Support macOS
In order to support macOS, this patch updates the meson build script in
the source root directory to pass macro definition, '__MACOS__', to the
compiler.
Signed-off-by: Wook Song <wook16.song@samsung.com>
jijoong.moon [Mon, 12 Aug 2019 23:45:52 +0000 (08:45 +0900)]
[nnfw] Adding nnfw support with nnapi
Support nnfw. If enable-nnfw is on, it will use nnfw.
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
Jaeyun [Thu, 22 Aug 2019 12:43:21 +0000 (21:43 +0900)]
[Api/Android] add option for tf-lite version
update script to set tensorflow-lite version (1.9 and 1.13 available)
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Thu, 22 Aug 2019 08:20:02 +0000 (17:20 +0900)]
[Api/Android] add target abi
Update build script to set target abi.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Thu, 22 Aug 2019 03:47:34 +0000 (12:47 +0900)]
[Api] code clean
1. add struct for state change callback
2. check element pointer before free the resources
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Thu, 22 Aug 2019 03:28:09 +0000 (12:28 +0900)]
[Api/Pipeline] remove unnecessary function
Remove unnecessaty code to set buffer-free function.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Wed, 31 Jul 2019 10:19:36 +0000 (19:19 +0900)]
[Android/Api] add script to build api library
Add script to build android-api library.
If runs this script, ext-files are downloaded and run gradle build.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Thu, 8 Aug 2019 11:57:45 +0000 (20:57 +0900)]
[Api/Pipeline] element restriction
Handles element restriction (whitelist for API release)
If needs restriction in pipeline elements, simply set the list in meson option.
e.g. meson -Denable-capi=true -Denable-element-restriction=true -Drestricted-elements='videotestsrc videoconvert' build
this allows videotestsrc and videoconvert with nnstreamer elements.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Wed, 7 Aug 2019 09:41:42 +0000 (18:41 +0900)]
[Api/Pipeline] set state paused while constructing the pipeline
when called construct(), set the pipeline state paused.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Fri, 2 Aug 2019 04:40:32 +0000 (13:40 +0900)]
[Api/Tizen] base code to check plugin name
prepare code to validate the plugins in pipeline.
TODO: update code to initialize white-list for tizen
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
MyungJoo Ham [Wed, 14 Aug 2019 14:46:23 +0000 (23:46 +0900)]
[Test/bmp2png] Use g_strdup to avoid gcc9 strict checks
strncpy is required not to use strlen for the length.
Use g_strdup and let glib handle it.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Sangjung Woo [Tue, 6 Aug 2019 04:12:49 +0000 (13:12 +0900)]
[Spec] Fix the bug of post uninstall script.
Post uninstall script with `-p` option invokes a single program directly
instead of spawning shell script. Because of this reason, below error
occurs when uninstalling package. This patch fixes that bug.
/sbin/ldconfig: relative path `1' used to build cache
warning: %postun(nnstreamer-0.2.1-6.4.i686) scriptlet failed, exit
status 1
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
MyungJoo Ham [Wed, 14 Aug 2019 11:35:59 +0000 (20:35 +0900)]
[gcc9-fix] fix strncpy '__builtin_strncpy' issue
This fixes:
FAILED: tests/59830eb@@bmp2png@exe/bmp2png.c.o
cc -Itests/59830eb@@bmp2png@exe -Itests -I../tests -I/usr/include/libpng16 -fdiagnostics-color=always -pipe -D_FILE_OFFSET_BITS=64 -Werror -std=gnu89 '-DVERSION="0.2.1"' -Wredundant-decls -Wwrite-strings -Wformat -Wformat-nonliteral -Wformat-security -Winit-self -Waddress -Wno-multichar -Wvla -Wpointer-arith -Wmissing-declarations -Wmissing-prototypes -Wnested-externs -Waggregate-return -Wold-style-definition -Wdeclaration-after-statement '-DNNSTREAMER_CONF_FILE="/etc/nnstreamer.ini"' -DHAVE_ORC=1 -DENABLE_TENSORFLOW_LITE=1 -DENABLE_PYTORCH=1 -DENABLE_CAFFE2=1 -g -O2 -fdebug-prefix-map=/<<PKGBUILDDIR>>=. -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2 -MD -MQ 'tests/59830eb@@bmp2png@exe/bmp2png.c.o' -MF 'tests/59830eb@@bmp2png@exe/bmp2png.c.o.d' -o 'tests/59830eb@@bmp2png@exe/bmp2png.c.o' -c ../tests/bmp2png.c
In file included from /usr/include/string.h:494,
from ../tests/bmp2png.c:36:
In function ‘strncpy’,
inlined from ‘main’ at ../tests/bmp2png.c:291:3:
/usr/include/aarch64-linux-gnu/bits/string_fortified.h:106:10: error: ‘__builtin_strncpy’ specified bound depends on the length of the source argument [-Werror=stringop-overflow=]
106 | return __builtin___strncpy_chk (__dest, __src, __len, __bos (__dest));
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
../tests/bmp2png.c: In function ‘main’:
../tests/bmp2png.c:291:34: note: length computed here
291 | strncpy (pngfilename, argv[1], strlen (argv[1]) + 1);
| ^~~~~~~~~~~~~~~~
cc1: all warnings being treated as errors
This is Case 3 of #1598
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Wed, 14 Aug 2019 12:03:42 +0000 (21:03 +0900)]
[Debian/Unittest] Disable capi-unittest in arm/launchpad.
It takes too much time in the launchpad arm emulator.
Disable it to pass unittest-capi.
This resolves another case of #1598
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Wed, 14 Aug 2019 00:45:13 +0000 (09:45 +0900)]
[Debian] Remove tensorflow-dev build-dep for arm64
We are removing tensorflow-arm64 from our PPA.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Tue, 13 Aug 2019 10:08:23 +0000 (19:08 +0900)]
[Debian] Use Tensorflow only if it's amd64
If it's ARM, don't use tensorflow.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jaeyun [Mon, 5 Aug 2019 05:28:49 +0000 (14:28 +0900)]
[Android/Api] method to get data size
add method to calculate byte size of tensor data.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Fri, 2 Aug 2019 10:46:54 +0000 (19:46 +0900)]
[Api/Single] invalid output if error occurs
1. allocate output buffer after invoke is done.
2. allow one buffer in appsink and drop old buffer.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Tue, 6 Aug 2019 02:35:19 +0000 (11:35 +0900)]
[Test] mark negative testcase
1. Add postfix in tc name for the negative testcases.
2. Update comment for future work.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Mon, 12 Aug 2019 07:06:37 +0000 (16:06 +0900)]
[Build] set python encode for meson
set python-encode utf-8 for meson build.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Fri, 9 Aug 2019 07:19:13 +0000 (16:19 +0900)]
[Filter/Py] remove global val to register subplugin
SAM result, remove global val about filter-framework.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Wook Song [Fri, 9 Aug 2019 04:43:35 +0000 (13:43 +0900)]
[Debian/armhf] Add build dependency on protobuf
Caffe2-related features, which are enabled by default, require
libprotobuf-dev at build time. This build dependency is omitted only for
armhf architecture among those we support. This patch resolves this
issue.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Wed, 7 Aug 2019 05:32:55 +0000 (14:32 +0900)]
[Meson] Apply work-around for gtest-related failure on meson 0.50.1
In the case that gtest has been installed via apt (i.e., apt install
libgtest-dev), meson provides the -I options indicating the directory
that does not exist. Therefore, it is failed to build when
'-Wmissing-include-dirs' is set. This patch works around this issue.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Jaeyun [Thu, 8 Aug 2019 05:58:25 +0000 (14:58 +0900)]
[Meson] option to disable orc
Add meson-option to disable orc library (default true).
Set -Denable-orc=false if you need to block the orc library.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Dongju Chae [Thu, 8 Aug 2019 08:06:23 +0000 (17:06 +0900)]
[PythonFilter] Fix svace errors in python filter
This commit fixes svace errors.
Signed-off-by: Dongju Chae <dongju.chae@samsung.com>
MyungJoo Ham [Thu, 8 Aug 2019 02:04:52 +0000 (11:04 +0900)]
[Readme] Update release status
We now fully ready for Android & Yocto.
Update the release status
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jaeyun [Wed, 7 Aug 2019 05:32:55 +0000 (14:32 +0900)]
[Meson] remove missed option
remove missed option about tensorflow mem-optimize in meson.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Sangjung Woo [Wed, 7 Aug 2019 05:38:29 +0000 (14:38 +0900)]
[C-API] Add sample code for ml_pipeline_switch_get_pad_list()
This patch newly adds the sample code for ml_pipeline_switch_get_pad_list()
function since it looks hard to use. Moremover, it updates the remarks
section for finalizing its allocated memory space.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Sangjung Woo [Wed, 7 Aug 2019 04:40:00 +0000 (13:40 +0900)]
[C-API] Describe pre & post state of the pipeline
This patch newly add the pre & post state of the pipeline in C-API.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Sangjung Woo [Wed, 7 Aug 2019 02:10:51 +0000 (11:10 +0900)]
[C-API] Describe the pipeline state as image
This patch describes the the pipeline state as image instead of ASCII
text art.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Sangjung Woo [Tue, 6 Aug 2019 11:28:51 +0000 (20:28 +0900)]
[C-API] apply the review guide of C-API
* Fix the typo (targetting -> targeting)
* Use detailed comments (pipe is NULL? -> Not negotiated?)
* Use detailed URL (github.com/nnsuite/nnstreamer -> https://github.com/nnsuite/nnstreamer)
* Add '*' into omitted comment line
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Sangjung Woo [Tue, 6 Aug 2019 11:06:26 +0000 (20:06 +0900)]
[C-API] Add PERMISSION_DENIED state for privilege issue
This patch newly adds the ML_ERROR_PERMISSION_DENIED error code for
privilege issue and its checking logic in ml_get_feature_enabled()
function.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Wook Song [Tue, 6 Aug 2019 14:40:38 +0000 (23:40 +0900)]
[Tests] Apply bash script formatter to all runTest.sh files
In order to improve readability, this patch applies a bash shell script
formatter named 'shfmt' [1] to all runTest.sh files in the ./tests
directory. Note that 4 spaces are used for indentation instead of using
'tab'.
[1] https://github.com/mvdan/sh
Signed-off-by: Wook Song <wook16.song@samsung.com>
Jaeyun [Tue, 30 Jul 2019 04:02:22 +0000 (13:02 +0900)]
[Test] stream test for custom-filter without model
Add simple stream testcase for custom-filter without model file.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Tue, 30 Jul 2019 03:56:57 +0000 (12:56 +0900)]
[Common] function declaration
fix link error about plugin api in cpp file.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Tue, 6 Aug 2019 07:53:18 +0000 (16:53 +0900)]
[Api] common headers in nnstreamer
change nnstreamer headers include to prepare gradle build.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Hyoung Joo Ahn [Wed, 7 Aug 2019 02:07:32 +0000 (11:07 +0900)]
[Filter/Tensorflow] remove the unused option for memory optimization
after changing tensorflow to use `c-api`, the `mem_optmz` is not used anymore
Signed-off-by: Hyoung Joo Ahn <hello.ahn@samsung.com>
Hyoung Joo Ahn [Wed, 7 Aug 2019 02:17:18 +0000 (11:17 +0900)]
[Typo] fix typo at README.md
fix a typo of README file: technolog -> technology
Signed-off-by: Hyoung Joo Ahn <hello.ahn@samsung.com>
Jaeyun [Mon, 5 Aug 2019 04:00:34 +0000 (13:00 +0900)]
[Build] fix build warning
Add function declaration and fix enum conversion.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Fri, 2 Aug 2019 10:19:23 +0000 (19:19 +0900)]
[Api] update api description
Fix typo and update details about the APIs.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Fri, 2 Aug 2019 07:16:59 +0000 (16:16 +0900)]
[Api] check null param
In the functions for tensor data, add code to check null param.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
MyungJoo Ham [Fri, 2 Aug 2019 06:39:54 +0000 (15:39 +0900)]
[Tizen/C-API] Increate single-invoke timeout
Timeout of 1 sec is too short even for unit-test cases
in QEMU-ARM. Increase it to 3 sec.
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
MyungJoo Ham [Fri, 2 Aug 2019 04:49:43 +0000 (13:49 +0900)]
[UnitTest/TizenCAPI] Support in-emulator unit test
In QEMU, things are not that fast. Slow it down!
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Jaeyun [Thu, 1 Aug 2019 12:10:47 +0000 (21:10 +0900)]
[Tizen/Feature] invalid condition to check feature status
1. Fix invalid condition when mutex-lock of element handle locked.
2. Add new struct to set mutex locked.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Sangjung Woo [Fri, 26 Jul 2019 07:00:49 +0000 (16:00 +0900)]
[C-API] Update the return value of each API
This patch updates the return value of each API by adding
'ML_ERROR_NOT_SUPPORTED' if the device is not supported
'machine_learning.inference'. Moreover, this updates nnstreamer_doc.h
file for doxygen.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Sangjung Woo [Fri, 26 Jul 2019 05:00:54 +0000 (14:00 +0900)]
[C-Api] Check feature before processing API
This patch checks the machine_learning.inference feature before
processing API by calling ml_get_feature_enabled() function. In case of
Unit test, feature checking could be skipped by calling internal API,
ml_set_feature_status().
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
wip
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Jaeyun [Mon, 29 Jul 2019 04:00:01 +0000 (13:00 +0900)]
[Android/Api] add custom-filter callback
1. Add custom-filter in java to execute tensors data in the application.
2. Update sample code to set the custom-filter.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Fri, 26 Jul 2019 11:14:29 +0000 (20:14 +0900)]
[Android/Api] invalid data conversion
1. Fix wrong conversion of byte buffer in native.
2. Add methods to allocate buffer with native byte order.
3. Fix invalid condition to close handle.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Thu, 25 Jul 2019 06:58:21 +0000 (15:58 +0900)]
[Android/Api] refactor code
1. separate native code to each java class.
2. add common functions and definitions.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Hyoung Joo Ahn [Wed, 24 Jul 2019 08:25:59 +0000 (17:25 +0900)]
[Filter/TF] apply c_api at tensor_filter_tensorflow
change the codes of tensorflow filter to use c_api only
Signed-off-by: Hyoung Joo Ahn <hello.ahn@samsung.com>
Wook Song [Fri, 26 Jul 2019 06:44:08 +0000 (15:44 +0900)]
[Tests/Filter/Tensorflow] bugfix: Fix a typo and wrong if-logic
This patch fixes a critical typo and wrong if-logic in order to make the
test cases properly run when the '/etc/nnstreamer.ini' is used.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Fri, 26 Jul 2019 04:49:11 +0000 (13:49 +0900)]
[Tests/Filter/PyTorch] bugfix: Fix a typo and wrong if-logic
In order to make the test cases properly run when the
'/etc/nnstreamer.ini' is used, this patch fixes a critical typo and
wrong if-logic for such cases.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Fri, 26 Jul 2019 04:24:12 +0000 (13:24 +0900)]
[Tests/Filter/Caffer2] bugfix: Rewrite wrong if-logic
Since it is wrong that the if-logic for the test cases using
'/etc/streamer.ini', the tests are always terminated before even
started. This patch fixes this bug.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Fri, 26 Jul 2019 04:17:10 +0000 (13:17 +0900)]
[Tests/Filter/Caffe2] bugfix: Fix a typo
This patch fixes a critical typo in the runTest script. In order to run
test cases using '/etc/nnstreamer.ini', 'ini_path' should be replaced
with 'ini_file'.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Jaeyun [Thu, 25 Jul 2019 07:14:17 +0000 (16:14 +0900)]
[Filter] condition to check model
Prevent error case if filter framework does not need a model file.
Add a flag in nnfw struct and change condition to check model in transform callback.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Thu, 25 Jul 2019 07:11:33 +0000 (16:11 +0900)]
[Test] free string
free strings after getting the pad list in testcase.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Wook Song [Fri, 24 May 2019 04:45:53 +0000 (13:45 +0900)]
[Conf] Handle the case of using the default 'sysconfdir' value of meson
Since the default value of 'sysconfdir' in the Meson build system is
just 'etc' instead of '/etc', NNSTREAMER_CONF_FILE could be set to wrong
path when the default value is used. This patch fixes this issue.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Thu, 25 Jul 2019 02:45:48 +0000 (11:45 +0900)]
[Tests/CustomFilter] Fix a bug in code for checking existence of OpenCV
This patch fixes a minor bug in the code for checking existence of
OpenCV libraries.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Jaeyun [Fri, 19 Jul 2019 09:22:26 +0000 (18:22 +0900)]
[Android/Sample] add sample code to use library
Add sample code to use nnstreamer library.
This sample includes basic example to run the single-shot and pipeline examples repeatedly.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Wed, 10 Jul 2019 11:31:14 +0000 (20:31 +0900)]
[Android/Api] api base
Update files for android-java api.
- initial commit to add android java api set.
- single-shot / pipeline functions.
- data callback from sink node in pipeline api.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Fri, 19 Jul 2019 06:42:50 +0000 (15:42 +0900)]
[Decoder/Sub] build static lib
change font file in decoder-subplugins to build static library.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
jijoong.moon [Fri, 14 Jun 2019 04:23:35 +0000 (13:23 +0900)]
[ANDROID] Built as Shared Library with tensorflow-lite
In order to build nnstreamer shared libraries including
sub-plugins (decoder, filters), the android.mk file is
restructured. Becuse gstreamer building blocks list for nnstreamer might be
required for static build and android application build, it is moved
to nnstreamer.mk.
If you bulid libnnstreamer.so using this android.mk, gstreamer shared
libries should be prepared. Please refere "Build Gstreamer full source
based Android rootFS with Cerbero" in
https://github.com/nnsuite/nnstreamer-example/tree/master/android.
Once the build is finished successfully, you can find libnnstreamer.so
and gstreamer shared libraries in libs/$TARGET_ARCH/ which is listed
in nnstreamer.mk under the name of "NNSTREAMER_BUILDING_BLOCK_LIST".
you can use these to build your own android application.
New script for tensorflow-lite tensor filter library is added. If
there is not tflite source, "prepare_tflite.sh" will download and
prepare the external packages of tflite. Using these source code,
android.mk make tflite static library and use it to build
libnnstreamer_filter_tensorflow-lite.so (which includes static
tensorflowlite.a for android of course)
If you want to build android application statically, you could just use the
macro defined in nnstreamer.mk.
Build NNStreamer as Shared Library.
- Gstreamer Shared Libraries : Cerbero
: Download and build by yourself
- Tensorflow-lite : If TENSORFLOW_LITE Macro is not defined,
NNStreamer is going to download and configure
--> static lib
- NNStreamer : libnnstreamer.so
- NNStreamer tflite : libnnstreamer_filter_tensorflow-lite.so
- NNStreamer decoder: libnnstreamer_decoder_direct_video.so
libnnstreamer_decoder_bounding_boxes.so
libnnstreamer_decoder_image_labeling.so
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: jijoong.moon <jijoong.moon@samsung.com>
Jaeyun [Mon, 22 Jul 2019 09:38:26 +0000 (18:38 +0900)]
[Sink] remove unnecessary functions
use basesink default and remove unnecessary functions.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Sangjung Woo [Fri, 19 Jul 2019 09:45:14 +0000 (18:45 +0900)]
[C-API] check gstreamer initialization status in ml_single_open()
Before calling gstreamer function, this patch checks its initialization
status by calling gst_init_check() in ml_single_open().
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Wook Song [Mon, 22 Jul 2019 05:48:59 +0000 (14:48 +0900)]
[Tests/Filter/TFLite] bugfix: Fix wrong if-logic
This patch fixes wrong if-logic that results in termination of the test
without running the gstTest code when /etc/streamer.ini is used.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Mon, 22 Jul 2019 05:37:41 +0000 (14:37 +0900)]
[Tests/Filter/TFLite] bugfix: Correct a critical typo
This patch corrects a critical typo by replacing 'ini_path' with
'ini_file'. Note that the 'ini_file' variable is required to run the
test case when '/etc/nnstreamer.ini' is used.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Mon, 22 Jul 2019 05:12:10 +0000 (14:12 +0900)]
[Tests/Decoder/ImageLabeling] bugfix: Fix wrong if-logic
This patch fixes wrong if-logic that leads to always skipping the
gstTest code line when /etc/streamer.ini is used.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Wook Song [Mon, 22 Jul 2019 02:26:25 +0000 (11:26 +0900)]
[Tests/Decoder/ImageLabeling] bugfix: Correct critical typo
When the unit tests for image labeling decoder is run using
/etc/nnstreamer.ini, it would be stuck because of a critical typo in
the variable indicating /etc/nnstreamer. This patch fixes this issue.
Signed-off-by: Wook Song <wook16.song@samsung.com>
Hyoung Joo Ahn [Fri, 12 Jul 2019 07:54:57 +0000 (16:54 +0900)]
[Filter/Tensorflow] apply a new constructor for TFBuffer
the different constructor is applied according to the version of TF
Signed-off-by: Hyoung Joo Ahn <hello.ahn@samsung.com>
Sangjung Woo [Fri, 19 Jul 2019 09:08:56 +0000 (18:08 +0900)]
[Tizen/Api] add smack label for tensor filter
Tensor filters such as libnnstreamer_filter_tensorflow-lite.so does not
have a smack label, smack deny occurs when accessing it . This patch fixes
that bug.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
MyungJoo Ham [Thu, 18 Jul 2019 10:40:11 +0000 (19:40 +0900)]
[Dist/Tizen] Disable tensorflow (non-lite) for Tizen
With the Tensorflow update from 1.09 to 1.13 @ Tizen,
we have decided to drop tensorflow (non-lite).
Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>
Sangjung Woo [Thu, 18 Jul 2019 10:20:51 +0000 (19:20 +0900)]
[C-Api] fix memory leak bug
Since developers can reduce the number of Tensors by calling
ml_tensors_info_set_count(), it causes the memory leak when invoking
ml_tensors_info_free() and ml_tensors_data_destroy(). This patch fixes
that bug by checking all possible elements (i.e. ML_TENSOR_SIZE_LIMIT).
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Jaeyun [Wed, 17 Jul 2019 06:56:30 +0000 (15:56 +0900)]
[C-Api] fix error to run pipeline
1. return valid error code when opening single-shot model.
2. When called destroy() in pipeline, remove all callbacks before changing pipeline state.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Jaeyun [Mon, 15 Jul 2019 04:58:50 +0000 (13:58 +0900)]
[C-Api] free element handle
Add code to free pipeline element and hash-table.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Hyoung Joo Ahn [Fri, 12 Jul 2019 06:09:45 +0000 (15:09 +0900)]
[Filter/TF] fix build conflict with TF version 1.13.1
overide a specific method at the specific version of TF
Signed-off-by: Hyoung Joo Ahn <hello.ahn@samsung.com>
Sangjung Woo [Thu, 11 Jul 2019 05:34:37 +0000 (14:34 +0900)]
[C-Api] Add NNFW enum for future use
This patch newly adds the ML_NNFW_TYPE_NNFW enumeration for future use
of NNFW.
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>
Jaeyun [Wed, 10 Jul 2019 08:23:33 +0000 (17:23 +0900)]
[C-Api/Single] validate input data
Add code to validate input data in single-invoke function.
Signed-off-by: Jaeyun Jung <jy1210.jung@samsung.com>
Sangjung Woo [Thu, 11 Jul 2019 02:07:21 +0000 (11:07 +0900)]
[C-API] Update doxygen for Tizen API
This patch updates the feature name of Tizen API and its related
documentation. Detailed information is as below.
* NNStreamer -> Single / Pipeline without NNStreamer
* Newly add the 2 feature keys (machine_learning &
machine_learning.inference)
Signed-off-by: Sangjung Woo <sangjung.woo@samsung.com>