doc/API-reference: elaborate the document.

Add some explanations to the topic lists.

Signed-off-by: MyungJoo Ham <myungjoo.ham@samsung.com>

diff --git a/Documentation/API-reference.md b/Documentation/API-reference.md
index 1cc18dd..17938a4 100644 (file)
--- a/Documentation/API-reference.md
+++ b/Documentation/API-reference.md
@@ -2,10 +2,34 @@
 title: API reference
 ...
 
+
 ### [NNStreamer Doxygen book](http://ci.nnstreamer.ai/nnstreamer/html/index.html)
 
-### [Tizen native API](https://docs.tizen.org/application/native/guides/machine-learning/overview/)
+[NNStreamer Doxygen book](http://ci.nnstreamer.ai/nnstreamer/html/index.html) provides documents generated by doxygen comments of nnstreamer's source code.
+If you are investigating the internals of nnstreamer, this document might be helpful.
+
+### [Tizen Machine-Learning Native API reference](https://docs.tizen.org/application/native/guides/machine-learning/overview/)
+
+NNStreamer is acting as the library to execute inferences in Tizen; in other words, NNStreamer is the implementation of the Tizen's MachineLearning.Inference APIs.
+[This link](https://docs.tizen.org/application/native/guides/machine-learning/overview/) provides native APIs (C-APIs) for Tizen Machine Learning.
+In this document, the "inference.single" and "inference.pipeline" are implemented by NNStreamer.
+MachineLearning.Inference.Single allows using subplugins of ```tensor_filter``` to run the given neural network model, invoking the model directly from the application with a single-shot input data.
+MachineLearning.Inference.Pipeline allows describing and executing GStreamer pipelines including NNStreamer elements.
+Note that MachineLearning.Training and MachineLearning.Service are implemented by other packages in this organization.
+
+If you are writing Tizen applications in C/C++ along with neural network inferences, you are supposed to use Tizen Machine-Learning.Inference Native APIs.
+
+### [Tizen Machine-Learning .NET API reference](https://docs.tizen.org/application/dotnet/guides/machine-learning/overview/)
+
+If you are writing Tizen applications in .NET along with neural network inferences, you are supposed to use Tizen Machine-Learning.Inference.Singleshot/Pipeline .NET APIs.
+The subdomains have the same roles as in Native APIs.
+
+### [Tizen Machine-Learning Web API reference](https://docs.tizen.org/application/web/guides/machine-learning/overview/)
+
+If you are writing Tizen web applications along with neural network inferences, you are supposed to use Tizen Machine-Learning.Inference.Single/Pipeline Web APIs.
+The subdomains have the same roles as in Native APIs.
 
-### [Tizen .NET API](https://docs.tizen.org/application/dotnet/guides/machine-learning/overview/)
+### [Android Java API reference](http://ci.nnstreamer.ai/nnstreamer/ci/daily-build/build_result/latest/java_html/index.html)
 
-### [Android Java API](http://ci.nnstreamer.ai/nnstreamer/ci/daily-build/build_result/latest/java_html/index.html)
+We provide Java APIs for Android applications.
+Note that this supports SingleShot/Pipeline subdomains and do not include training or ML-service capabilities.