Here is some toy example which is distinguish simple images.
The Mobile ssd V2 tensor flow lite model is used for the feature extractor and Nearest Neighbor is used for the classifier. All the training and testing is done on the Galaxy S8.
-![image](https://github.com/nnstreamer/nntrainer/blob/master/doc/08b09a80-ef29-11e9-8303-475fd75f4b83.png)
+![image](https://github.com/nnstreamer/nntrainer/blob/master/docs/images/08b09a80-ef29-11e9-8303-475fd75f4b83.png)
Happy(^^), sad(TT), soso(ㅡㅡ) classes are used and prepare 5 images for the training and two images for the test set each as below.
-![image](https://github.com/nnstreamer/nntrainer/blob/master/doc/a73cfb80-ef29-11e9-9ae9-0d6531538eaf.png)
+![image](https://github.com/nnstreamer/nntrainer/blob/master/docs/images/a73cfb80-ef29-11e9-9ae9-0d6531538eaf.png)
After remove the fully connected layer of mobile ssd v2, 128 features are extracted. The features from first training set data is below.
-![image](https://github.com/nnstreamer/nntrainer/blob/master/doc/0997fb00-ef2e-11e9-90a3-51c27bf4013f.png)
+![image](https://github.com/nnstreamer/nntrainer/blob/master/docs/images/0997fb00-ef2e-11e9-90a3-51c27bf4013f.png)
Simple euclidean distance is calculated and the result is quite good. All the test set is collected.
-![image](https://github.com/nnstreamer/nntrainer/blob/master/doc/87103b00-ef2f-11e9-9c1a-83da0faafb63.png)
+![image](https://github.com/nnstreamer/nntrainer/blob/master/docs/images/87103b00-ef2f-11e9-9c1a-83da0faafb63.png)
Due to the simplicity of this toy example, all the test results are collect.
There are two more random pictures which little bit differ from right image. As you can see, it is little bit hard to tell which class it is. First image could be classified as "happy" but the red zone is across with sad and the variance is quite small. Second image is more confused. Cause the smallest distance is all over the classes.
May be should be define the threshold which is not implemented yet.
-![image](https://github.com/nnstreamer/nntrainer/blob/master/doc/33552000-ef36-11e9-88f6-ea6a35ccdf6b.png)
+![image](https://github.com/nnstreamer/nntrainer/blob/master/docs/images/33552000-ef36-11e9-88f6-ea6a35ccdf6b.png)
The results is below.
<p align=center>
-<img src =https://github.com/nnstreamer/nntrainer/blob/master/doc/de916e80-0b9f-11ea-9950-5c40d2bef8e4.gif width=300 >
-<img src =https://github.com/nnstreamer/nntrainer/blob/master/doc/d2f17800-0b9e-11ea-8060-edfeacd6c71e.gif width=300 >
+<img src =https://github.com/nnstreamer/nntrainer/blob/master/docs/images/de916e80-0b9f-11ea-9950-5c40d2bef8e4.gif width=300 >
+<img src =https://github.com/nnstreamer/nntrainer/blob/master/docs/images/d2f17800-0b9e-11ea-8060-edfeacd6c71e.gif width=300 >
</p>
The configuration of the example is below,
<p align = "center">
-<img src="https://github.com/nnstreamer/nntrainer/blob/master/doc/02a7ee80-f0ce-11e9-97b8-bcc19b7eb222.png" width="400" height="250" > </p>
+<img src="https://github.com/nnstreamer/nntrainer/blob/master/docs/images/02a7ee80-f0ce-11e9-97b8-bcc19b7eb222.png" width="400" height="250" > </p>
Training set and test set are below
<p align = "center">
-<img src="https://github.com/nnstreamer/nntrainer/blob/master/doc/7944ec00-f0ce-11e9-87af-aea730bcd0f5.png" >
+<img src="https://github.com/nnstreamer/nntrainer/blob/master/docs/images/7944ec00-f0ce-11e9-87af-aea730bcd0f5.png" >
</p>
After Iterating 300 times, the change of L2 Norm of the Loss function is below.
<p align = "center">
-<img src="https://github.com/nnstreamer/nntrainer/blob/master/doc/d42b1300-f0cf-11e9-9b6f-6db30def4684.png" width="500" height="300">
+<img src="https://github.com/nnstreamer/nntrainer/blob/master/docs/images/d42b1300-f0cf-11e9-9b6f-6db30def4684.png" width="500" height="300">
</p>
and the test results for the 8 test cases are below. Step function is used to make more clear.
As you can see, the test result is ok.
<p align ="center">
-<img src="https://github.com/nnstreamer/nntrainer/blob/master/doc/16555400-f0d2-11e9-959b-f61935fefd5a.png" width ="500" height="180">
+<img src="https://github.com/nnstreamer/nntrainer/blob/master/docs/images/16555400-f0d2-11e9-959b-f61935fefd5a.png" width ="500" height="180">
</p>