Age | Commit message (Collapse) | Author |
|
Signed-off-by: Yi Zhao <yi.zhao@windriver.com>
|
|
Signed-off-by: Yi Zhao <yi.zhao@windriver.com>
|
|
Signed-off-by: Yi Zhao <yi.zhao@windriver.com>
|
|
Signed-off-by: Yi Zhao <yi.zhao@windriver.com>
|
|
Signed-off-by: Yi Zhao <yi.zhao@windriver.com>
|
|
Signed-off-by: Yi Zhao <yi.zhao@windriver.com>
|
|
Signed-off-by: Yi Zhao <yi.zhao@windriver.com>
|
|
Signed-off-by: Yi Zhao <yi.zhao@windriver.com>
|
|
Signed-off-by: Yi Zhao <yi.zhao@windriver.com>
|
|
Signed-off-by: Yi Zhao <yi.zhao@windriver.com>
|
|
Signed-off-by: Yi Zhao <yi.zhao@windriver.com>
|
|
add nmt into meta-tensorflow, support vi-en tranlation
Signed-off-by: Changqing Li <changqing.li@windriver.com>
|
|
- To enable modular op registration support, which non-monolithic build
defines `framework_shared_object=true'
- The non-monolithic build of TensorFlow does not support
`-Wl,--as-needed' and explicitly set `-Wl,--no-as-needed'
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
It is still based on MIT license, but the files should include
a Wind River copyright.
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
- Set option by modify setup.py
- Do not override do_compile to call `setup.py build_ext', it will
missing options at do_install time which causing host contamination
...
Found webp/demux.h
Found library webpmux at /usr/lib/libwebpmux.so
Found library webpdemux at /usr/lib/libwebpdemux.so
/usr/include/pthread.h:682:6: error: 'regparm' is not valid on this platform
...
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
The supportetd MACHINE in meta-yocto-bsp is genericx86-64
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
The demo does not train a model from scratch, it retrains from an existed model.
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Require hardware:
- genericx86-64 bsp;
- usb webcam;
Require softwre:
- OpenCV
- TensrFlow/TensorFlow Lite
- The usb webcam kernel driver
Provide function:
- collect_face: start camera to collect and label human
face images for retraining.
- recognize_face: start camera and inference through tensorflow lite
model to recognize human face
The opencv code refers:
https://docs.opencv.org/3.4.5/d7/d8b/tutorial_py_face_detection.html
- Save avi video from usb webcam
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
The demo recipe requires extra layers which is not necessary for
meta-tensorflow, so create a new sub layer meta-demo
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
On host:
- Retrain model mobilenet_0.50_224
- Convert it to Tersorflow Lite model
On target:
- Tersorflow inference label_image.py
- Tersorflow Lite inference label_image_lite.py which refers:
https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/examples/python/label_image.py
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
While building with '-march=nehalem' or '-msse4.2', it failed with
'NEON_2_SSE.h' file not found.
Refer https://github.com/tensorflow/tensorflow/issues/23440, define
marco TF_LITE_DISABLE_X86_NEON to workaround it.
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
The var-PYTHON_SITEPACKAGES_DIR is used for target, correct with native.
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Since [b1a6aa3 fix `Failed to extract install_base_key'
while running bazel], revert [a81b50e build tensorflow-native
and tensorflow in order]
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
1. The problem is the binary is edited after by setting the rpath. Which might
mess up with zip data, you might want to run zip -A on Bazel. It stills needs
to extract the install key to verify it does not need to extract all the
remaining parts. Also Bazel contains other C++ binary and you are going to
run in the rpath problem all the time.
https://github.com/bazelbuild/bazel/issues/2978
2. Since binary bazel in recipe-sysroot-native is a hardlink, while multiple
recipes invoke one bazel at the same time, there is probably a race error
...
FATAL: Failed to open Bazel as a zip file: (error: 5): Input/output error
...
Copy and run a duplicated bazel could workaround it
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Previously model does not accurate, refer upstream:
https://www.tensorflow.org/lite/models/image_classification/overview
update to latest modle.
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Refers:
https://github.com/tensorflow/tensorflow/blob/master/tensorflow/lite/examples/label_image/label_image.md
Add a pre-trained model `Mobilenet v1 224' for lable_image
https://storage.googleapis.com/download.tensorflow.org/models/tflite/mobilenet_v1_224_android_quant_2017_11_08.zip
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Refers:
https://github.com/tensorflow/tensorflow/tree/master/tensorflow/examples/label_image
Add a pre-trained model `Inception v3' for lable_image
https://cloud.google.com/tpu/docs/inception-v3-advanced
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
It caused undefined reference neon_tensor_utils failure
...
|tensor_utils.cc:(.text._ZN6tflite12tensor_utils18ReductionSumVectorEPKfPfii+0x0):
undefined reference to `tflite::tensor_utils::NeonReductionSumVector(float const*,
float*, int, int)'
...
Define Yocto specific cpu `armeabi' for lite label_image
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Add required bsp layer, build beagleboard, raspberrypi,
genericx86 (including atom), genericx86-64, and
imx6sxsabresd successfully.
So remove beagleboard, raspberrypi, atom from future plan.
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Add tensorflow-native to tensorflow's DEPENDS, actually tensorflow
does not require tensorflow-native, but to avoid do_compile at
the same time. Bazel build system does not support parallel build
very well (very slowly).
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Build fails looking for `execinfo.h` when building against musl
|In file included from ./tensorflow/core/platform/stacktrace.h:26,
| from tensorflow/core/platform/stacktrace_handler.cc:34:
|./tensorflow/core/platform/default/stacktrace.h:27:10: fatal error:
execinfo.h: No such file or directory
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
The idea comes from upstream arm compiler which `Build from
source for the Raspberry Pi'
$ ls <TensorFlow source code>/third_party/toolchains/cpus/arm/
arm_compiler_configure.bzl BUILD CROSSTOOL.tpl
https://www.tensorflow.org/install/source_rpi
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|
|
The build steps refers README of https://github.com/tensorflow/estimator
Signed-off-by: Hongxu Jia <hongxu.jia@windriver.com>
|