From: Inki Dae <inki.dae@samsung.com>
Date: Thu, 11 Jun 2020 07:29:14 +0000 (+0900)
Subject: Set model path according to MLAPI backend
X-Git-Tag: submit/tizen/20200626.050805~3
X-Git-Url: http://review.tizen.org/git/?a=commitdiff_plain;h=c08e5496c150d2c667cb4caf9bce0f0d02dcff6e;p=platform%2Fcore%2Fmultimedia%2Finference-engine-mlapi.git

Set model path according to MLAPI backend

NNFW - in-house NN Runtime - needs NNPackage type of package
which is a directory containing a model file and its meta file.
For more details, you can refer to
https://github.com/Samsung/ONE/tree/master/nnpackage/examples/one_op_in_tflite

ML Single API framework of NNStreamer receives a full path of a given model file
from user - in our case, inference-engine-mlapi backend - and find metadata
in the directory that the a given model file is located.

So inference-engine-mlapi backend should pass a full path of
the a given model file to ML Single API framework.

Change-Id: I6bdd871d5b683dbd6e60fce0f6dbd052985cd514
Signed-off-by: Inki Dae <inki.dae@samsung.com>
---

diff --git a/src/inference_engine_mlapi.cpp b/src/inference_engine_mlapi.cpp
index d6d4706..ddb6507 100644
--- a/src/inference_engine_mlapi.cpp
+++ b/src/inference_engine_mlapi.cpp
@@ -103,11 +103,7 @@ namespace MLAPIImpl
 	{
 		LOGI("ENTER");
 
-		// ML Single API of MLAPI requires model_paths rule like below,
-		// "so library file path,nb model file path" or vise versa.
-		std::string model_str(model_paths[0] + "," + model_paths[1]);
-
-		LOGI("Model name = %s", model_str.c_str());
+		std::string model_str("");
 
 		// TODO. Set NNFW backend type and HW type properly.
 
@@ -128,9 +124,14 @@ namespace MLAPIImpl
 				LOGE("Invalid target device type.");
 				return INFERENCE_ENGINE_ERROR_NOT_SUPPORTED;
 			}
+
+			// ML Single API of MLAPI requires model_paths rule like below,
+			// "so library file path,nb model file path" or vise versa.
+			model_str += model_paths[0] + "," + model_paths[1];
 			break;
 		case INFERENCE_BACKEND_NNFW:
 			nnfw_type = ML_NNFW_TYPE_NNFW;
+
 			if (mTargetDevice == INFERENCE_TARGET_CPU) {
 				nnfw_hw = ML_NNFW_HW_CPU_NEON;
 				LOGI("Target device is NEON.");
@@ -141,6 +142,8 @@ namespace MLAPIImpl
 				LOGE("Invalid inference target device type.");
 				return INFERENCE_ENGINE_ERROR_INVALID_PARAMETER;
 			}
+
+			model_str += model_paths[0];
 			LOGI("NNFW tensor filter will be used.");
 			break;
 		// TODO.
@@ -149,6 +152,11 @@ namespace MLAPIImpl
 			return INFERENCE_ENGINE_ERROR_INVALID_PARAMETER;
 		}
 
+		LOGI("Model name = %s", model_str.c_str());
+
+		// TODO. create ml_tensor_info for input and output tensor and pass
+		//		 them as parameters of ml_single_open function.
+
 		int ret = ml_single_open(&mSingle, model_str.c_str(), NULL, NULL,
 								 nnfw_type, nnfw_hw);
 		if (ret != ML_ERROR_NONE) {