Add beyond support

Change-Id: I6802946930f09ad02f090557daef2d79c834d610
Signed-off-by: Hyunsoo Park <hance.park@samsung.com>

Add Google Model codes as internal api

Change-Id: I751448688f7089006aecda8d32c2bc4ffd49208e
Signed-off-by: Hyunsoo Park <hance.park@samsung.com>

test: Fix hard-coded backend type and input layer name

Change-Id: I5448d98bb93d9288d024d367368fcba379cfc2c4
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: Fix target device configuration

Fixed an issue that target device type isn't updated correctly
when binding a backend engine.

This patch makes the target device configuration to be done
before a backend engine binding is completed because actual target device
configuration to the backend engine is done at binding time.

Change-Id: I24b5779f294abdf8944d4ec0339f1c1282e2156b
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: Load mlapi specific ini file correctly

inference_engine_mlapi_backend.ini file is specific to MLAPI backend.
So load this file only in case that a given backend type is MLAPI.

Change-Id: I385d73d8571b506f8ffa6937985436fe8e18f96a
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: Call LoadConfigFile function

This call updates default tensor filter type for MLAPI
from /etc/inference/inference_engine_mlapi_backend.ini file.

Change-Id: I024160aa01470c52dd579b0a84c9e0052dcd5b7f
Signed-off-by: Inki Dae <inki.dae@samsung.com>

Update blazeModel to lite model

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

test: revert the use of stack

Change-Id: Icf69584bbbfbf6aa839dbcba3f02f6522d17a86b
Signed-off-by: Inki Dae <inki.dae@samsung.com>

test: fix stack bug

Change-Id: I6a485fc03a0dee5789918e5f81235a1c54d72964
Signed-off-by: Inki Dae <inki.dae@samsung.com>

Add blazePose test to mv_stream_infer

test command is: mv_stream_infer 5 0 0 90 5 0

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

Modify pose compare

Pose mapping format is changed.
Maximum mapping landmarks are sixteen by mv_inference_human_pose_landmark_e.

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

Fix the number_of_pose

The landmark model outputs 39 landmarks but only 33 are valid.

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

Merge branch 'tizen_devel_gesture' of ssh://review.tizen.org:29418/platform/core/api/mediavision into tizen_devel_gesture

Support blazePose

- Detector by mv_inference_object_detect()
- The detector's output is handled by user_data via callback
- Poselandmark detector by mv_inference_pose_landmark_detect()
- The poselandmark detector's input is preprocesses with the detector's output.
  So, it should be given to mv_inference_pose_landmark_detect() by user_data.

- A mapping file for blazepose is not ready. It will be patched with next commit.

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

Apply mouse press / release action

Change-Id: I75c4220e0a1c5dd6550caa941c7f13c382afecf9
Signed-off-by: Hyunsoo Park <hance.park@samsung.com>

test: fix help messages

Change-Id: I676ba797039fc8cbca948628d24725aab4b96cec
Signed-off-by: Inki Dae <inki.dae@samsung.com>

test: drop dead and commented code for cleanup

Change-Id: I23fbc30065b3f5491752a43feb9dcc9ac8407795
Signed-off-by: Inki Dae <inki.dae@samsung.com>

test: measure total latency correctly

Change-Id: Iaa55d24f5c594d31882849e8f92c500fe5114e68
Signed-off-by: Inki Dae <inki.dae@samsung.com>

test: Drop armnn and mlapi test apps

Change-Id: I96d817a890df6541f928b5eaf5eb5a7e9e3204aa
Signed-off-by: Inki Dae <inki.dae@samsung.com>

test: Use stack for input buffer from Camera source

Change-Id: I404bc0000336a0461a9a631480d448d505f3503d
Signed-off-by: Inki Dae <inki.dae@samsung.com>

test: Integrate stream_infer test apps with one

Change-Id: I9d7b29b9d5d8b5c41bf6439166a58ce5e8a226ba
Signed-off-by: Inki Dae <inki.dae@samsung.com>

Fix Seg. fault problem without parameters

Change-Id: I948d6b440f021cf46b137a3aeede0d57f60163f9
Signed-off-by: Inki Dae <inki.dae@samsung.com>

Use stack for camera source

Change-Id: I830bea1df77b6183f328f58de25904056ef01d1b
Signed-off-by: Inki Dae <inki.dae@samsung.com>

Add common util and use stack

Change-Id: I274b8774133c286442a5b74e6a1194e84de1055e
Signed-off-by: Inki Dae <inki.dae@samsung.com>

correction heuristic algorithm

Change-Id: I259bddb9525075d377b1992b9fc3f1523d21a6cf

Fix tensor information for DLDT

Change-Id: I4752fb271094b8c79481933b2a192a90f3ffd18a
Signed-off-by: Kwang Son <k.son@samsung.com>

Add DLDT(Openvino) backend support

Change-Id: Ib77f33843268e00a8bbc3ea56a234a98c5b5ba89
Signed-off-by: Kwang Son <k.son@samsung.com>

test: update stream_infer_armnn

Change-Id: If3f429b53b576d72cee20f9ffc040b85f796deed
Signed-off-by: Inki Dae <inki.dae@samsung.com>

Merge mediavision which is used internally

Change-Id: I17b2fbc41745eec5b722ff249488c8e3afbef7c6
Signed-off-by: Hyunsoo Park <hance.park@samsung.com>

[testsuite] add hand gesture new model

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

Merge branch 'tizen_devel_gesture' of ssh://review.tizen.org:29418/platform/core/api/mediavision into tizen_devel_gesture

Change-Id: I3a548d0a493dcc9746ba6d25a65961522e20e0e3

[WIP #1] support human pose with a camera

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

mv_inference: Caculate tensor size correctly

Change-Id: I74273668500ed259544e0fc2977d1351756de81b
Signed-off-by: Inki Dae <inki.dae@samsung.com>

[WIP #1] support human pose with a camera

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

Merge branch 'tizen_devel_pose' into tizen_devel_gesture

[testsuite] add additional hand model test config

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

[Bug] fix bug to apply roi when it is given

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

[WIP #1] fix bug to apply roi if it exist

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

[Bug] fix bug to access invalid memory

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

Draw human pose with a video file

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

[WIP #8] support label when a model's output has a label

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

[Bug] fix bug when a video file given

A video file should be given with numbers when num-buffers option is enabled.

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

[WIP #7] add hand gesture code handling 1x42 output dims

Previously, output is one tensor of the heatmaps with shape(1xHxWxC)
or 2 tensors (gesture with scalar and coordinate with (1xHxW)).
But, it couldn't handle one tensor of coordinate with shape(1xH or 1xW).
This patch temporally handle it.

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

[WIP #6] human pose work around code

Human pose and hand pose are working with camera stream

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

[WIP #5] fix stream_infer testsuites to run handgesture with pose api on ACR

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

[WIP #4] add hand models

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

[WIP #3] add missed hand detection test

While adding hand detection bugs area detected.
1. Need to fix that the output of pose estimation should be between 0 and 1.
Then, an user has to rescale it.
2. Need to define hand pose

3. Need to distinguish human and hand pose

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

[WIP #2] temporally distinguish hand and pose model with the number of outputs

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

[WIP #1] sync to tizen_devel brach; up to pose acr

Merge pose landmark detection() API to tizen_devel_gesture branch.

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

test: Add hand gesture test for ARMNN

Change-Id: I09b9627412a70762720b34f13e2801291a9f4dbb
Signed-off-by: Inki Dae <inki.dae@samsung.com>

test/stream_infer: Add inference test app on MLAPI

This patch adds an application which tests an inference
on On-device Neural Engine via ML Single API of NNStreamer.

With this patch, two applications will be generated,
 - mv_stream_infer_tflite which is for using TFLITE backend.
 - mv_stream_infer_mlapi which is for using ML Single API backend.

Change-Id: I11870f55e9ae1ae405063b630447dcec66f27740
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: Change in-house runtime backend name

Official name of NNFW is ONE(On-device Neural Engine)
so use it instead of NNFW.

Change-Id: Iefbab8ed475be0037fb3458842093d59fad418d8
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: Fix tensor data type

Output tensor type of Inception v3 model converted to Vivante NPU
is UINT16.

Change-Id: I7e08093f1a6dfc06a6e6a121f430a14e5d262ca9
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: Change MLAPI backend string to "mlapi"

Change-Id: I1038bdc002c489abb1c51e323660135accc67905
Signed-off-by: Inki Dae <inki.dae@samsung.com>

test: Change a backend type from VIVANTE to MLAPI

Change-Id: Ib4fa271a8ba5511d2bc7166ded760f58d09a5ea3
Signed-off-by: Inki Dae <inki.dae@samsung.com>

Consider tensor filter plugin type for NNFW

This patch adds a new backend type for NNFW support via NNStreamer.

MediaVision will support Vivante NPU and NNFW(in-house NN Runtime)
through ML Single API of NNStreamer.

For this, this patch also adds a new member, backend_type, in
inference_engine_config, which is used to let ML Single API of
NNStreamer know which tensor filter - Vivante NPU or NNFW - should be used.

Change-Id: I532a44b16155dda0ab1ebc68dfc208645944995c
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: Consider float16 type of tensor.

This patch adds float tensor buffer support.
float tensor buffer should be allocated with short type
and release it properly according to the tensor buffer ownership.

Change-Id: I0e9ff667b6273c4ad762d8026f256da88f3076a4
Signed-off-by: Inki Dae <inki.dae@samsung.com>

test: Add Vivante NPU test case

Change-Id: Ifa54daac72b9d34a15c199d5c8fa6a2cbbea7374
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: Add Vivante NPU type

This patch adds a new type for Vivante NPU support.

On MediaVision, Vivante NPU will be controlled by
inference-engine-mlapi backend which interfaces with
ML Single API of NNStreamer.

Ps. Vivante NPU needs two binaries for the inference.
    One is Vivante NPU specific model file(xxx.nb), and
other is so library which implements graph creation API
specific to Vivante NPU.

Change-Id: If54d55f52091682fe9957408258133e6efdba17a
Signed-off-by: Inki Dae <inki.dae@samsung.com>

test: Add NNFW backend test case

Change-Id: I453ebe467da4765614acf00965f84efc3a81370b
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: Do not check a config file in case of NNFW backend

NNFW backend doesn't need a config file so skip checking this file.

Change-Id: Ia3f6194410edd1df460350a9aeb6d0aee7a14703
Signed-off-by: Inki Dae <inki.dae@samsung.com>

[testsuite/stream_infer] Support hand gesture model and add stream_infer

Hand gesture model consists of two models. One is a model that gets an image
as an input and gives a hand segmentation and heatmaps as an output. Another is
a model that gets filtered heatmaps, raw tensor buffer, as an input and gives
coordinate and a gesture as an output.

To support that, apis are added:
mv_source_fill_by_tensor_buffer()
 v_inference_hand_detect()
mv_inference_pose_estimation_detect()

To test the models,
stream_infer testsuite is added.

Change-Id: Id150dd893c229e2a207e099b46f8d53b029e291f
Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

mv_inference: Add uint64 output tensor support

Change-Id: Ieda8df82ae27cf0120dd8a463bbed7aa436f900d
Signed-off-by: Inki Dae <inki.dae@samsung.com>

test: Add hand gesture model test for tflite

Change-Id: Iee0df1201dc1acc0aa527004bc2e605bee3f67fb
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: Consider uint32 tensor type

Change-Id: I99968eaa190d62a801523d4300019b4ff6a40803
Signed-off-by: Inki Dae <inki.dae@samsung.com>

test: Add hand gesture model test for ARMNN

Change-Id: Id0e5adcbaf952ecda44ef300957fd8a3d416b4ac
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: Add hand gesture support

Change-Id: I9ff7a93eb73527a1c1288daaa914a5e229537264
Signed-off-by: Inki Dae <inki.dae@samsung.com>

Revert "Revert "mv_inference: Add Post Estimation feature support""

This reverts commit ddfddc6f6a0a1730558f0874ae3d44a7cbaf5d00.

Change-Id: I9f3300434f51de7c9eab6084dd9812ab8b209785

Revert "test: drop pose estimation test"

This reverts commit 72b730ed61c69b4a4957b2173776023e71218b13.

Change-Id: I8a14aaaf9f0cb1505a0cb7f33323524e71b9230b

Add capi-media-vision-common.pc

capi-media-vision-common.pc was not exist

Change-Id: Ida62970d0dc22c0f20fec4b2d708c956a59bbab0
Signed-off-by: Kwang Son <k.son@samsung.com>

Merge "Revert "Missing pc file in common-devel"" into tizen

Revert "Missing pc file in common-devel"

This reverts commit 511f7c502f13f558b810db83de337bb73c9576e9.

Change-Id: I819969a3ffde681b19344e076522830a9897ce65

Merge "Missing pc file in common-devel" into tizen

Missing pc file in common-devel

add pc file

Change-Id: I3acc73a5bcf0eabaaf1bd9f991e0df2e2581c368
Signed-off-by: Kwang Son <k.son@samsung.com>

Delete redundant pc.in files

pc.in file is template and not modified.
cmake CONFIGURE_FILE copy to own pc files, so pc.in is enough with one file.

Change-Id: I78cbd414ca55e9875a7e95ea552bcb1536246d91
Signed-off-by: Kwang Son <k.son@samsung.com>

mv_inference: handle an exception case to class creation correctly

Change-Id: Iafaecb467c149479eabdc218eab2e3223c6a374c
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: clean up class member initialization

Change-Id: Ic8626debc2149fdc4b09e38b2d004e3728cf7e7a
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: use basic constructor function

InferenceEngineCommon class doesn't provide parameter
required constructor functions anymore so use basic
constructor function.

Change-Id: I58be39c2a7e70e362b375d54548e142847787092
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: use initializer list

This patch just cleans up a member initialization of Inference
class by using initializer list.

Change-Id: Ib9dc6470fdfef271101316a34f8f18caa29d6f2c
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: drop deprecated internal API

This patch drops deprecated internal API below,
 - mv_inference_configure_input_node_name_open
 - mv_inference_configure_output_node_names_open
 - and relevant callback functions.

No users to use these API anymore.

Change-Id: Id77b44c8ce781ee3b8fed2e44c8c6311180de2db
Signed-off-by: Inki Dae <inki.dae@samsung.com>

Fix UNINIT.CTOR svace issue

Change-Id: I6363096e866de0b2b214235e57d2af069d0fa61f
Signed-off-by: Hyunsoo Park <hance.park@samsung.com>

mv_inference: consider backward compatibility

This patch guarantees backward compatibility of Mediavision C-API.

In new inference engine framework, inference target device types
can be combined together. So to keep backward compatibility with
Tizen 5.5, it adds a new definition, MV_INFERENCE_TARGET_DEVICE_TYPE
and check if Tizen app uses old type of inference target device
or not by checking default values of MV_INFERENCE_TARGET_TYPE
and MV_INFERENCE_TARGET_DEVICE_TYPE.

Change-Id: I31b2fee05ac74265569f3cdac3875879d349dc5d
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: unbind only if a backend instance exists

Change-Id: I622cb586594ba8513423ffdd22ff51588553461f
Signed-off-by: Inki Dae <inki.dae@samsung.com>

Sync mv_inference.c code

capi-media-vision library is refered by dcm-service so
make sure to keep same code as mv_inference framework
which uses a new inference engine framework interfaces.

Change-Id: Ic73ba6b237c04a226cbb3ee9b0276fbfeabac2a5
Signed-off-by: Inki Dae <inki.dae@samsung.com>

fix build error on aarch64

Use %u instead of %zu.

Change-Id: Idc7588935ddc1562caa0eaabadf9c29783c26864
Signed-off-by: Inki Dae <inki.dae@samsung.com>

fix build error on aarch64

On aarch64, size_t is 64bit data so make sure to use correct
format specifier.

Change-Id: I4e22b46ada58ea520900fcd6271d7582a31347e9
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: Clean up ABI header file

Responsed to ACR-1545 review comments.
- just correct header description formats.

Change-Id: Ie3fabe70177cdbd65d9d0f730190b0a6e0ee3604
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: clean up ABI header files

This patch cleans up the ABI header files,
mv_inference.h and mv_inference_type.h according to
ACR(ACR-1545) review comments below,
- Add a description to each constant of mv_inference_data_type_e
- Change constant names of mv_inference_target_device_e.
- Clean up some descriptions.
- Revive @since_tizen 5.5 tag
- Fixed description of #define MV_INFERENCE_TARGET_TYPE

Change-Id: Icd893710a04a550205c64a99f355ebf48825a584
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: clean up mv_inference_type header

Change-Id: I4eacc53f2b6d96b4f55b77a642b58cbcbe49ace3
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: change enumeration values to new ones

some enumeration values of inference-engine-interface have been
updated so change them.

Change-Id: Iea466768e7ffe4089dc63938c8c9be921628d654
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: keep backward compatibility

This patch keeps backward compatibility with Tizen 5.5
by adding new mv_inference_target_device_e enumeration
instead of modifying existing one.

Change-Id: I40df1dacc6b9a74ba438d0767d84ff1f25133584
Signed-off-by: Inki Dae <inki.dae@samsung.com>

test: drop pose estimation test

Drop pose estimation test code because pose
estimation isn't not a feature targeted on
M1 release but M2.

This code will be moved to tizen_devel branch.

Change-Id: Ifaab03fed0f510e7c1a854c3c5dfe9696a1dc90c
Signed-off-by: Inki Dae <inki.dae@samsung.com>

Revert "mv_inference: Add Post Estimation feature support"

This feature will be added to tizen 6.0 m2 so revert it
from tizen branch and move it to tizen_devel branch.

This reverts commit 68d2d47e71302f563f24555ce1cce4ca10d908a2.

Change-Id: Iec73032301d724ba549560564c529b26e46515dd

mv_inference: drop data types not supported

Mediavision doesn't support float16 and uint16/32bit model
so drop these data types. They can be added later with actual
use case.

Change-Id: Iebe7511326f45c39b09a4429f8de361ca8fe354c
Signed-off-by: Inki Dae <inki.dae@samsung.com>

Change type of 'tensor_info.shape' to size_t

Signed-off-by: Hyunsoo Park <hance.park@samsung.com>

inference testsuite: add opencv with caffe model

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

mv_inference: add post-process step to GetObjectDetectionResults()

To handle model data (SSD) which doesn't have post-process layer,
add the post process step.
SSD generally provides an one output with 1x1xNx7 shape and the last dimension
includes 'image id, class id, confidence, coordinates(left-top (x,y),
and right-bottom(x,y))'

Signed-off-by: Tae-Young Chung <ty83.chung@samsung.com>

Revert "mv_inference: Get layer propery only in required case"

This reverts commit 756a6d9c297b7c756a1090ded0287795b5f12665.

test: Add input tensor data type support

This patch passes user-given input data type to
Inference framework so that each inference engine backend
can know the input tensor data type before kicking up inference.

Change-Id: I762d7f76444d5b6656b019501321ca3c90c6919a
Signed-off-by: Inki Dae <inki.dae@samsung.com>

mv_inference: Add inference data type support

This patch adds inference input data type support for
Mediavision. With this, we can use user-given input tensor
data type for OpenCV backend.

Change-Id: I681f203f4cbe9b3cc7b59a746cfab650009f65c1
Signed-off-by: Inki Dae <inki.dae@samsung.com>