Added documentation

Documentation~/index.md

@@ -0,0 +1,39 @@
+# Barracuda Starter Kit
+
+This is a [Barracuda](https://github.com/Unity-Technologies/barracuda-release) demo library containing usage examples. All examples are also compatible with mobile platforms (iOS and Android).
+
+Each example is placed in a separate folder with a demo scene and all the required resources.
+
+## Static Image Recognition
+
+![MobileNetV2](images/mobilenet.png)
+
+MobileNetV2 [pytorch.org](https://pytorch.org/hub/pytorch_vision_mobilenet_v2) example running in Barracuda on a set of static images. 
+
+To use this demo, open and run the `01-StaticImageRecognition-MobileNetV2` scene. Select the image and the top matching class will be selected for it, outputting the class name and accuracy percentage. 
+
+This model was converted to `onnx` file using Python and `torch.onnx` from the `pytorch_vision_mobilenet_v2.ipynb` Google Colab. Softmax layer was added to the output layer to normalize resulting accuracy values. The following script was added to the PyTorch MobileNet demo source code.
+```Python
+import torch
+import torch.onnx
+#...
+#...
+model = torch.nn.Sequential(model, torch.nn.Softmax(1))
+torch_onnx.export(model, input_batch, "mobilenet_v2.onnx", verbose=False)
+```
+
+This model has 1000 classes and should be used with a square `224x224` image. Demo project preprocesses the image to fit the target image dimensions, but the best accuracy is reached when the height and width ratio equals one. 
+
+Images used for this demo are created by Aurimas Petrovas and are under Unity license.
+
+## Face Tracking 
+
+![MobileNetV2](images/blazeface.png)
+
+[MediaPipe BlazeFace](https://sites.google.com/view/perception-cv4arvr/blazeface) model. Simplified real time camera face tracking Barracuda demo [project](https://github.com/keijiro/BlazeFaceBarracuda) originally created by Keijiro Takahashi.
+
+To use this demo, open and run the `02-FaceTracking-BlazeFace` scene. Point the web camera at a face and it will add a bounding box around the face with accuracy percentage and facial feature points. 
+
+The BlazeFace ONNX model was originally converted by PINTO0309 (Katsuya Hyodo).
+Please refer to [his model zoo](https://github.com/PINTO0309/PINTO_model_zoo) for more information. Keijiro Takahashi converted his ONNX model into a Barracuda-compatible form. Please
+check [this Colab notebook]( https://colab.research.google.com/drive/1O1KDIVsmYyYDqEqv7hEqofsHMCa49xaZ?usp=sharing) for the details of the conversion process.