Jihoon Lee [Tue, 15 Jun 2021 06:35:27 +0000 (15:35 +0900)]
[API/Devel] Retouch devel header
**Changes proposed in this PR:**
- Change function sinature to exploit forward declaration
- s/export_to/exportTo
- Change header guard to be unique
**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>
Parichay Kapoor [Wed, 16 Jun 2021 03:17:55 +0000 (12:17 +0900)]
[neuralnet] Neural network bugfix
Two PRs merged have caused main branch to fail.
This patch provides temporary fix.
Proper fix will be done with layer_v2.
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Wed, 16 Jun 2021 03:17:00 +0000 (12:17 +0900)]
[unittests] Update unittests for new gtest version
Remove gtest version check comments and update unittests for the new
gtest versions.
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Thu, 10 Jun 2021 04:21:03 +0000 (13:21 +0900)]
[Loss] Apply requireLabel to feed label
Currently, loss layer is the only one that can accept label which was
prohibiting building a custom loss layer.
**Major Changes**
- Apply requireLabel to feed label
- Last layer is not assumed loss layer when feeding label(still it is in
other places, need to be dealt)
**Minor Changes**
- Feeding multiple label is now possible (spliting label is future work)
- [Fix] Count Trainable includes loss layer
- [Fix] Handle cases When all layer is not trainable for
neuralnet::backwarding
**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>
Parichay Kapoor [Fri, 11 Jun 2021 09:02:56 +0000 (18:02 +0900)]
[layer] Move name to LayerNode
Move layer identifier name to LayerNode.
Now each node in the graph has name as its direct identifier.
This patch also disable unittest_nntrianer_layers for some while
till layer refactoring is taking place. However, other unittests
and apptests must remain enabled to ensure that everything else is
working.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Mon, 14 Jun 2021 02:45:16 +0000 (11:45 +0900)]
[split layer] Initialize class member bugfix
Split layer has class member leading_helper_dim
which remains uninitialized and results in bugs
when setBatchSize() is called before initialization of the layer.
This patch fixes the bug.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Thu, 10 Jun 2021 04:38:00 +0000 (13:38 +0900)]
[layer context] Propose layer context and layer_devel
This patch proposes layer context and layer_devel.
The layer context provides basic necessities for each layer
like inputs, outputs, weights, tensors. Weights are the trainable tensors
required by the layers which can be requested with the context interfaces.
Correspondingly, any other tensor requirements by the layer can also be
requested with the context.
There are two types of context :
1. InitContext - provides context for initialization with input_dimensions
filled. This contains specifications for all the inputs/outputs/weights/tensors
which will be made available during executiong.
2. RunContext - provides allocated weights/inputs/outputs/tensors along with
some of the basic properties of the layer like trainable, etc.
This design allows layer devel to be stateless (derived layer can be stateful with their
own member objects), and behave as layer ops.
Further, this allows dependencies on weights/var_grads as context simplified the interface
with just tensors.
Also, this patch also removes dependency on the manager from layer_devel header.
layer context for each layer is owned by the corresponding LayerNode.
Many of the layer getters like getInputs/Outputs/Weights/etc can now
be moved to LayerNode which does not concern the custom layer writer,
and simplifies the layer operation code as well.
See also #986
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Fri, 11 Jun 2021 10:52:31 +0000 (19:52 +0900)]
[C++17] Bump base version to c++17
This patch bumps baseline version to c++17.
**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>
Parichay Kapoor [Mon, 14 Jun 2021 02:54:32 +0000 (11:54 +0900)]
[deb/pkg] Updates for debian packaging
Update tolerance for sqrt unittest for debian arm64 packaging
increase timeout for appdraw classification app for debian armhf packaging
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Fri, 14 May 2021 10:55:58 +0000 (19:55 +0900)]
[Props] Implement input_layers as a property
This patch implements input_layers as a property with following changes
below
**Changes proposed in this PR:**
- Add ConnectionSpec object for generic purpose
- Add InputSepc as a property
- Add following tests
**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>
Jihoon Lee [Fri, 11 Jun 2021 10:35:39 +0000 (19:35 +0900)]
[Test/models] Add addition test
This patch adds addition test, while add support for reading multiple
inputs and outputs inside `unittest_nntrainer_models.cpp`
**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>
Jihoon Lee [Fri, 11 Jun 2021 10:11:51 +0000 (19:11 +0900)]
[Retouch] unittest_models test cases
**Changes proposed in this PR:**
- Reduce scale of inputs
- Add iterations for recurrent layers (and delete _n) family tcs
- Clean duplicated codes in genModelTests.py
**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>
Parichay Kapoor [Mon, 14 Jun 2021 01:47:45 +0000 (10:47 +0900)]
[deb/pkg] Increase c-api unittest timeout
Increase c-api unittest timeout even further to
pass armhf build on launchpad.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Thu, 10 Jun 2021 08:45:20 +0000 (17:45 +0900)]
[Docs] Update custom documentation for loss layer
This patch updates documentation of custom for loss layer.
Also deletes redundant `pow.cpp` that sneaked in.
**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>
Parichay Kapoor [Fri, 11 Jun 2021 09:34:37 +0000 (18:34 +0900)]
[deb] Update lower tolerance bug + increase timeouts
Inverse the check for lower error tolerance for debian build
Also increase timeouts for nntrainer capi/ccapi unittests
as they timeout for armhf architecture on launchpad.
See also #1251
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Thu, 10 Jun 2021 04:42:09 +0000 (13:42 +0900)]
[Trivial] Change return -> retval
When stating value in doxygen it is recommended to use @retval, this
patch reflects the issue
**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>
Parichay Kapoor [Thu, 10 Jun 2021 08:43:56 +0000 (17:43 +0900)]
[interpreter] Updated export_to usage with LayerNode
Updated export_to function usage based on LayerNode
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Mon, 7 Jun 2021 06:52:50 +0000 (15:52 +0900)]
[layerNode] Refactor get/set number of inputs/outputs
Refactor get/set number of inputs/outputs.
This is now handled by layer_node.
Updating the input/output layers updates the count.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Mon, 7 Jun 2021 06:20:16 +0000 (15:20 +0900)]
[layer/node] Move input/output layers to LayerNode
This patch moves input/output layers to LayerNode.
Also create an interface for accessing/setting input/output layers
from in the LayerNode.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Fri, 4 Jun 2021 03:51:10 +0000 (12:51 +0900)]
[api] Added api dependency in pc files
Added API dependecy in pc files for the capi and ccapi.
The dependency is of the capi-ml-common which provides ml-api-common.h header.
Further, ml-api-common.h header from nntrainer repo is also removed.
Updated android build to download ml-api-common.h for nnstreamer daily builds.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
jijoong.moon [Fri, 28 May 2021 02:24:46 +0000 (11:24 +0900)]
[ RNN ] Add Multi-Layer RNN Unit Tests
This commit includes,
. Multi-Layered RNN Unit Tests
. Some Fixes to run
**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>
Jihoon Lee [Thu, 10 Jun 2021 08:30:35 +0000 (17:30 +0900)]
[Trivial] Fix NDK warning -Wbraced-scalar-init
This patch fixes trivial NDK warning below
```
././../nntrainer/layers/lstm.cpp:315:25: warning: braces around scalar initializer [-Wbraced-scalar-init]
Tensor dh_nx = Tensor({derivative_.width()});
^~~~~~~~~~~~~~~~~~~~~
././../nntrainer/layers/lstm.cpp:316:25: warning: braces around scalar initializer [-Wbraced-scalar-init]
Tensor dc_nx = Tensor({derivative_.width()});
^~~~~~~~~~~~~~~~~~~~~
././../nntrainer/layers/lstm.cpp:352:26: warning: braces around scalar initializer [-Wbraced-scalar-init]
hs_prev = Tensor({hs_t.width()});
^~~~~~~~~~~~~~
././../nntrainer/layers/lstm.cpp:354:26: warning: braces around scalar initializer [-Wbraced-scalar-init]
cs_prev = Tensor({cs_t.width()});
```
**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>
Parichay Kapoor [Thu, 10 Jun 2021 08:29:34 +0000 (17:29 +0900)]
[test/model] Cleanup unittest_nntrainer_models
Cleaned up unittest_nntrainer_models.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Tue, 8 Jun 2021 05:30:11 +0000 (14:30 +0900)]
[layer] Support inPlace() as an interface for the layer
Support inPlace() as an interface for the layer.
If returns true, manager can set its inputs/outputs to be
the same memory location, but not necessarily.
Corresponding graph code is also updated.
Resolves #867
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Tue, 8 Jun 2021 06:07:11 +0000 (15:07 +0900)]
[Layer] Add requireLabel() function
Add a simple function `Layer::requireLayer()`.
This function is used to detect if a certain layer requires a layer.
**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>
Parichay Kapoor [Wed, 2 Jun 2021 05:34:14 +0000 (14:34 +0900)]
[APP] Add tensorflow based product rating
Added tensorflow based product rating.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Wed, 2 Jun 2021 07:19:59 +0000 (16:19 +0900)]
[SplitLayer] Support flexible dimension
This patch adds support for the flexible dimension in split layer.
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Cc: Parichay Kapoor <pk.kapoor@samsung.com>
Signed-off-by: Jihoon Lee <jhoon.it.lee@samsung.com>
Jihoon Lee [Mon, 10 May 2021 02:53:39 +0000 (11:53 +0900)]
[Permute] Add permute layer forward/backward
This patch implements forward/backward with a following test
**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>
jijoong.moon [Fri, 21 May 2021 02:53:36 +0000 (11:53 +0900)]
[ RNN ] Add unittest case & return_sequence
This PR includes.
. unittest of rnn layer
. support of return sequence of rnn layer
**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>
Jihoon Lee [Wed, 26 May 2021 12:15:44 +0000 (21:15 +0900)]
[Recorder] Add multiout equivalent in transLayer
When getting gradient from `gradientTape` it adds all derivatives of the
dependent layer. While this behavior is true, we need to measure to what
extent each layer contributed to the layer.
This patch adds a multiout layer which consists of some stublayer that
seperates derivatives one by one.
**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>
jijoong.moon [Thu, 27 May 2021 11:20:32 +0000 (20:20 +0900)]
[ LSTM ] Add Multi-Layerd LSTM Unittest
This commit includes,
. Multi-Layerd LSTM Unit test.
**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>
Jihoon Lee [Thu, 6 May 2021 12:22:57 +0000 (21:22 +0900)]
[Permute] Implement peripheral functions
This patch implements initialize, export and setProperty of permute
layer
**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>
Parichay Kapoor [Fri, 21 May 2021 09:17:58 +0000 (18:17 +0900)]
[layer] Move distribute to LayerNode
move distribute to LayerNode from layer
Further, the distribute layer is now supposed to be used
as a wrapper of a layer maintained by the LayerNode,
and must not be maintained externally.
Rather, use getDistribute() to check if the node is distributed
and treat the layer as a regular layer.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Tue, 4 May 2021 01:50:09 +0000 (10:50 +0900)]
[Layer] Add permute layer scaffolding
This patch adds permute layer scaffolding
**Minor Fix**
- Add error handling for transpose tensordim
**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>
jijoong.moon [Thu, 27 May 2021 04:43:41 +0000 (13:43 +0900)]
[ LSTM ] Add More Unit Test Cases
This Commit inclues,
. Additional unittest
- with return sequneces
- with multiple batch size
- with multiple batch + return sequences
- with multiple batch + return sequences + multi-iteration
. Fixed initialization bug during training
. Remove check for tensor dot product to support multi-dimension
tensor.
**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>
jijoong.moon [Mon, 24 May 2021 05:54:44 +0000 (14:54 +0900)]
[ LSTM ] Add Unit test for LSTM Layer
This commit includes,
. unit test case for lstm layer
. fixup of potential bug : tanh prime calculation
.
**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>
jijoong.moon [Tue, 18 May 2021 01:25:52 +0000 (10:25 +0900)]
[ LSTM ] Add return_sequence
This commit includes,
. impelementation of return_sequence. If it is true, then it
generate all the output of time iteration and if it false, then
only last output is available.
. add 'return_sequence' keyward
Resolves:
**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>
jijoong.moon [Tue, 18 May 2021 01:24:54 +0000 (10:24 +0900)]
[ LSTM ] LSTM Unit Test Cases
This PR includes,
. LSTM Unittest Case
. python code in genModels
Resolves:
**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>
jijoong.moon [Thu, 13 May 2021 04:03:49 +0000 (13:03 +0900)]
[ LSTM ] Fix implementation & add backwarding testcase
This commits includes:
. fix implementation ( apply actvation )
. add one unittest to check functionality of backwarding
**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>
Jihoon Lee [Fri, 30 Apr 2021 10:34:06 +0000 (19:34 +0900)]
[Exporter] Implement saving fc node
**Changes proposed in this PR:**
- This patch add saving fullyconnected layer with node_exporter
- add dedeicated buffer to `TfNode`
**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>
Parichay Kapoor [Tue, 8 Jun 2021 06:11:54 +0000 (15:11 +0900)]
[app] reduce Logistic regression application epochs
Reduce logistic regression application epochs.
This application is to check working than verify accuracy and can
run with fewer epochs.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Wed, 26 May 2021 12:09:29 +0000 (21:09 +0900)]
[Recorder] Add Input map generation
There was a bug that layer input is directly from the previous layer.
If the model is not sequential, this assumption is not true, making
Addition test fail.
This patch build `model.recoder__input_maps` to correctly locate
the input tensor with some minor clean ups
**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>
Parichay Kapoor [Thu, 20 May 2021 05:22:08 +0000 (14:22 +0900)]
[meson] Enable tflite-interpreter by default
Enable tflite-interpreter by default as tflite layer is
already enabled.
Move flatc dependency out to the main meson.build
Also add proper path for the generated tf_schema
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Tue, 18 May 2021 05:44:09 +0000 (14:44 +0900)]
[layer] Move flatten property to layerNode
Move flatten property out to layerNode
also prepare for moving other properties out to layerNode
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Thu, 29 Apr 2021 06:31:33 +0000 (15:31 +0900)]
[Refactor] Delegate tfNode creation to exporter
This patch enables delegating exporting logic to exporter.
Note that this is supposed to be 1:1 mapping, if it has to be m:n
mapping, it should be directly handled by the interpreter, which
actually can see the node.
Also s/`save_result`/`saveResult` and `get_result`/`getResult`
**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>
[Android] Add tflite interpreter option
This patch add tflite interpreter option to ndk build
**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>
hyeonseok lee [Tue, 1 Jun 2021 04:52:09 +0000 (13:52 +0900)]
[Fix] Initialize member variable
- Initialize member variable sorted
- Catch exception when create the tensor
resolves: 1231330, 1143556
Signed-off-by: hyeonseok lee <hs89.lee@samsung.com>
Parichay Kapoor [Fri, 4 Jun 2021 05:54:46 +0000 (14:54 +0900)]
[deb/pkg] Reduce error tolerance for release build
Reduce error tolerance for release builds to succeed build on launchpad and tizen builds.
The default error tolerance is only reduced for all release builds for all tests that use defined tolerance.
However, the CI will run at original tolerance for all the builds and will ensure that
values are matched with original precision.
The tolerance has to be reduced because different os
have different versions and implementations of the blas
which results in minor difference in the results of the float
operations. However, when the difference is accumulated over
a range of operations, the unittests fail to match the
exact values.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Wed, 28 Apr 2021 10:46:58 +0000 (19:46 +0900)]
[Tflite] Separate `tfopnode`
This patch separates tfopnode to a file.
tfOpNode will be directly built from the exporter if it doesn't need to
be fused.
**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>
Parichay Kapoor [Wed, 2 Jun 2021 12:44:31 +0000 (21:44 +0900)]
[tflite] Update dependency from tflite to tf2lite
Update dependency from tensorflow-lite to tensorflow2-lite.
This is done for Tizen and Debian build. It has already been done for Android.
Tensorflow2-lite header is dependent on flatbuffer, so flatbuffers-devel
is additionally added as a build-dependency.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Fri, 21 May 2021 00:55:10 +0000 (09:55 +0900)]
[pkg/spec] Update spec packaging for tflite layer
Update spec packaging for tflite layer with tflite dependency
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Thu, 3 Jun 2021 05:15:27 +0000 (14:15 +0900)]
[test] Remove run_unittests
Remove run_unittests
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Thu, 3 Jun 2021 05:13:13 +0000 (14:13 +0900)]
[deb/pkg] Minor update to deb packaging
Update debian packaging:
- update the unittests to use meson tests
- update dh_missing to fail than give warnings
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Thu, 3 Jun 2021 03:34:47 +0000 (12:34 +0900)]
[unittest] Update unittest tolerance for debian build
Update unittest tolerance to be 1e-5 for debian build to match
across platforms (bionic, focal) and architectures.
Similar issues have occured for i586 tizen as well.
So, this patch lowers the tolerance for all ccapi and capi test.
This issue occurs to os and library versioning issues.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Thu, 3 Jun 2021 01:59:44 +0000 (10:59 +0900)]
[deb/pkg] Add nnstreamer filter and update testing
Add nnstreamer filter in the packaging
Also update the unittests to use the meson tests
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
jijoong.moon [Wed, 2 Jun 2021 13:15:52 +0000 (22:15 +0900)]
[ Release ] NNTrainer 0.2.0 Release
NNTrainer v0.2.0 is released.
**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, 2 Jun 2021 12:51:52 +0000 (21:51 +0900)]
[deb/pkg] Update debian install filenames
Update debian install filenames for the capi and ccapi.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Wed, 2 Jun 2021 06:42:00 +0000 (15:42 +0900)]
[Product Ratings] Fix inconsistent batch
This patch fixes inconcistent batch size between source file and model
configuration.
Also found a bug #1238, which will be dealt seperately.
**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>
Parichay Kapoor [Wed, 2 Jun 2021 04:50:16 +0000 (13:50 +0900)]
[debian] Refactor debian packaging
Refactor debian package to make separate packages for capi and ccapi, along with their devels.
Other bug fixes/improvements are:
- Update the names of the packages in pkg-config files for the capi and ccapi
- Update the installed files based on the requirement by the devel headers
(this might be even expanded). This will be refactored again once layer_internal and neuralnet
cleanup is performed.
- Update the included headers
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Tue, 1 Jun 2021 12:11:36 +0000 (21:11 +0900)]
[APP] Change embeding to product rating
This patch changes embedding application to product rating application
which is more practical use-case
**Changes proposed in this PR:**
- Update Readme
- Change directory and app name embedding->productratings
- tune model and sample dataset
**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>
Jihoon Lee [Mon, 19 Apr 2021 11:52:39 +0000 (20:52 +0900)]
[tflite] implement build operations in subgraph
This patch implements build operation in the subgraph
**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>
Parichay Kapoor [Tue, 1 Jun 2021 04:47:17 +0000 (13:47 +0900)]
[context/optimizer] Enable plugged optimizer
Enable plugged optimizer with the app_context
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Tue, 1 Jun 2021 02:30:01 +0000 (11:30 +0900)]
[optimizer] Update optimizer setProperty interface
Update optimizer setProperty interface based on the requirement
from the custom optimizer.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Thu, 27 May 2021 07:44:40 +0000 (16:44 +0900)]
[Example] Concat example
This patch adds concat example with embedding
The structure is
(input)
/ \
(embed1) (embed2)
\ /
(concat)
|
(flatten)
|
(fc)
**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>
Jihoon Lee [Tue, 1 Jun 2021 04:51:21 +0000 (13:51 +0900)]
[Fix] Prevent creating multiout for split
This patch make an exception for split layer to not create multiout
layer. While fixing trivial build error of sign compare
**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>
Parichay Kapoor [Mon, 31 May 2021 09:44:43 +0000 (18:44 +0900)]
[split/test] Added unittests for split layer
Added unittests for split layer covering initialization,
forwarding and backwarding. Corresponding bug fixes are
also added.
This has revealed an issue with concat layer #1226
which will be resolved soon.
Also add split layer to android build
Resolves #1208
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Fri, 21 May 2021 11:07:23 +0000 (20:07 +0900)]
[split layer] Add forwarding and backwarding
Add forwarding and backwarding for the split layer
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Fri, 21 May 2021 11:04:52 +0000 (20:04 +0900)]
[split layer] Added split layer scaffolding
Added split layer scaffolding with its properties and initialization
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Thu, 27 May 2021 05:22:06 +0000 (14:22 +0900)]
[App] Make embedding golden data optional
This patch makes embedding golden data optional while changing return
status
**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>
jijoong.moon [Thu, 13 May 2021 04:03:49 +0000 (13:03 +0900)]
[ LSTM ] Fix implementation & add backwarding testcase
This commits includes:
. fix implementation ( apply actvation )
. add one unittest to check functionality of backwarding
**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>
jijoong.moon [Tue, 11 May 2021 04:18:03 +0000 (13:18 +0900)]
[ LSTM ] Backwarding Implementation
This commit includes
. calDerivative & calGradient of LSTM Layer
. Modification of Forwarding of LSTM Layer
**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>
Jihoon Lee [Fri, 30 Apr 2021 04:06:35 +0000 (13:06 +0900)]
[VarGrad] Add clone Transpose
This patch adds clone transpose to `var_grad`, this function clones the
current `var_grad` but transpose and save
**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>
Jihoon Lee [Fri, 16 Apr 2021 08:09:08 +0000 (17:09 +0900)]
[Tflite] Add model input, output
This patch add model to track input and output
**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>
Jihoon Lee [Thu, 27 May 2021 07:35:07 +0000 (16:35 +0900)]
[Fix/graph] fix index after removing node
**Changes proposed in this PR:**
- After removing last node, adj node after the target node becomes not
correct, this patch fixes the issue
**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>
Jihoon Lee [Thu, 20 May 2021 04:18:41 +0000 (13:18 +0900)]
[Test/Model] Prepare addition model
This patch add addition model test
**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>
Jihoon Lee [Mon, 24 May 2021 11:21:25 +0000 (20:21 +0900)]
[Fix] Output layer setup correctly
There was a bug that output layer is not set correctly because of the
timing issue and overlapping checker issue.
This patch sets output layer correctly.
**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>
Jihoon Lee [Mon, 24 May 2021 10:14:33 +0000 (19:14 +0900)]
[LayerNode] Add print function
This patch adds simple print function to be used for debugging
Note that, input_layers are from `layer_internal` for now, but it's
private so it is commented out
**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>
Jihoon Lee [Tue, 18 May 2021 12:18:37 +0000 (21:18 +0900)]
[Test/models] prepare to support non-linear models
**Changes proposed in this PR:**
- Make `transLayer` from proxy to a legit layer
- Add output map so that in each layer can locate it's output from model
- Support non-linear output generation
- Add support for multi input, multi output
**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>
Parichay Kapoor [Mon, 17 May 2021 10:09:53 +0000 (19:09 +0900)]
[graph] Add non-const iterators
Add non-const iterators for the graph.
Most of the locations use const-iterators.
This will be updated when layer_node is updated to use its
properties than layer.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Mon, 17 May 2021 06:50:52 +0000 (15:50 +0900)]
[graph] Graph cleanup getNodes()
Graph remove support for getNodes() but rather
use iterators to iterator over the graph.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Mon, 17 May 2021 05:19:58 +0000 (14:19 +0900)]
[graph] Cleanup graph usage
Cleanup graph usage
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Mon, 17 May 2021 05:07:43 +0000 (14:07 +0900)]
[graph] Migrate to graph core
Migrate neural network independent and generic sections of the graph
to graph core. This is not yet complete and will be done over a few commits.
This will cleanup NetworkGraph class and will allow optimizations to
be done on the GraphCore/NetworkGraph class easily, and simplify
the classes as well.
Now, graph related structures can move out from layer to layer node.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
hyeonseok lee [Tue, 25 May 2021 09:19:56 +0000 (18:19 +0900)]
[Addition] fix forwarding
Correct initialization of output
**Self evaluation:**
1. Build test: [X]Passed [ ]Failed [ ]Skipped
2. Run test: [X]Passed [ ]Failed [ ]Skipped
Signed-off-by: hyeonseok lee <hs89.lee@samsung.com>
Jihoon Lee [Wed, 14 Apr 2021 11:02:18 +0000 (20:02 +0900)]
[Tflite] Fix tf shape building
This patch fixes tf shape building, by adding `eff_dim_flag` and
`dyn_dim_flag`.
`eff_dim_flag` checks which dimension are actually used so that to be
squeezed.
`dyn_dim_flag` checks which dimension should be set to -1 (meaning it is
not fixed)
With those flag set, our tensor dim is able to convert to the dimension
representation that supports `-1` and not fixed size
Those fixes are apply to fc layer for now
**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>
Jihoon Lee [Mon, 24 May 2021 07:35:37 +0000 (16:35 +0900)]
[Fix] Add missing header
This patch adds missing header for logger when `enable_logging=false`
**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>
Jihoon Lee [Fri, 14 May 2021 09:13:14 +0000 (18:13 +0900)]
[Props] Add Name validator and unittest
**Changes proposed in this PR:**
- Add name validator
- Add source file for `common_properties`
- Change name of `unittest_properties.cpp` ->
`unittest_base_properties.cpp`
- Add `unittest_common_properties.cpp`
**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>
Jihoon Lee [Mon, 10 May 2021 06:33:39 +0000 (15:33 +0900)]
[Conv2d/Test] Add more test
This patch adds test suggested at #1042
Resolves #1054
**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>
Jihoon Lee [Mon, 10 May 2021 03:57:22 +0000 (12:57 +0900)]
[Conv2d] Clean up conv2d relates
This patch handles issue left at #1042
**Changes proposed in this PR:**
- [x] move commented code upward
- [x] remove channel argument
- [x] Add conv negative test
- [ ] Add more test
See also: #1042, #1054
**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>
Jihoon Lee [Wed, 14 Apr 2021 08:21:16 +0000 (17:21 +0900)]
[Tflite] Implement `BuildTensors`
This patch add `BuildTensors` which build tensors into *.tflite 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>
Parichay Kapoor [Fri, 21 May 2021 06:11:46 +0000 (15:11 +0900)]
[fc] Make bias derivative in-place
Make bias derivative in-place for fc layer
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Fri, 21 May 2021 06:01:14 +0000 (15:01 +0900)]
[tensor] Update tensor sum
Update tensor sum to support in-place sum when done over
multiple dimensions.
Also update arguments order to have output tensor as a latter
argument.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Wed, 14 Apr 2021 06:27:26 +0000 (15:27 +0900)]
[tflite] Implement `buildBuffers`
This patch implements buildBuffers, now our generated tflite file has
buffers attached.
**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>
Parichay Kapoor [Tue, 18 May 2021 03:41:39 +0000 (12:41 +0900)]
[android] Fix android build
Fix android build error for file not found
add tensorflow new version to .gitignore
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Tue, 18 May 2021 01:08:56 +0000 (10:08 +0900)]
[time_dist] setProperty for time distribution layer
Added setProperty for time distribution layer, which passes
the properties to its underlying layer.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Tue, 18 May 2021 01:15:21 +0000 (10:15 +0900)]
[model] Model inference should not catch throw
Model inference should not catch throw, and must throw errors
so that the callers can catch it and detect errors.
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Parichay Kapoor [Wed, 12 May 2021 04:39:33 +0000 (13:39 +0900)]
[graph] Add graph core class
Add graph core class
This is mainly separating the existing implementation of core part
of the graph from NetworkGraph.
This patch does not contain new implementation but mostly contain
borrowed implementation from existing graph.
See Also #1176
**Self evaluation:**
1. Build test: [x]Passed [ ]Failed [ ]Skipped
2. Run test: [x]Passed [ ]Failed [ ]Skipped
Signed-off-by: Parichay Kapoor <pk.kapoor@samsung.com>
Jihoon Lee [Mon, 3 May 2021 08:37:49 +0000 (17:37 +0900)]
[Props] Add vector/array handler
This patch add vector property handler.
vectored property have sementics that number of size is not defined
arrayed property have sementics that number of size must match
**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>
Jihoon Lee [Tue, 18 May 2021 07:22:49 +0000 (16:22 +0900)]
[Svace] Fix dangling reference issue
This patch fixes dangling reference caused from Sorted.pop_back();
tracking id: 1229992
**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>
Jihoon Lee [Wed, 14 Apr 2021 04:47:12 +0000 (13:47 +0900)]
[Tflite] Implement serializing opcode
This patch implements implementing serialize opcode by
`buildOperatorCodes`
**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>
Jihoon Lee [Tue, 13 Apr 2021 13:16:11 +0000 (22:16 +0900)]
[TfLite] Implement TfOpIdxMap
This patch implments TfOpIdxMap
Please note that graph::initialize has been extracted for modularity and
testability within this patch.
**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>
Jihoon Lee [Tue, 13 Apr 2021 09:45:51 +0000 (18:45 +0900)]
[Tflite] Implement TfOpNode
This patch implements TfOpNode which will be used to deserialize layer
node to a mere chunk of informations. Please note that those methods are
tentative and subject to change wile implementing other methods
**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>