Merge pull request #106526 from shonohs/add_cvs_domain

Add a documentation about domain concept in Custom Vision Service.

Author: garycentric

articles/cognitive-services/Custom-Vision-Service/select-domain.md

@@ -0,0 +1,68 @@
+---
+title: "Select a domain for a Custom Vision project - Computer Vision"
+titleSuffix: Azure Cognitive Services
+description: This article will show you how to select a domain for your project in the Custom Vision Service.
+services: cognitive-services
+author: shonohs
+manager: nitinme
+ms.service: cognitive-services
+ms.subservice: custom-vision
+ms.topic: conceptual
+ms.date: 03/06/2020
+ms.author: shono
+---
+
+# Select a domain for a Custom Vision project
+
+From the settings blade for your Custom Vision project, you can select a domain for your project. Choose the domain that is closest to your scenario.
+
+## Image Classification
+
+|Domain|Purpose|
+|---|---|
+|__Generic__| Optimized for a broad range of image classification tasks. If none of the other domains are appropriate, or you're unsure of which domain to choose, select the Generic domain.|
+|__Food__|Optimized for photographs of dishes as you would see them on a restaurant menu. If you want to classify photographs of individual fruits or vegetables, use the Food domain.|
+|__Landmarks__|Optimized for recognizable landmarks, both natural and artificial. This domain works best when the landmark is clearly visible in the photograph. This domain works even if the landmark is slightly obstructed by people in front of it.|
+|__Retail__|Optimized for images that are found in a shopping catalog or shopping website. If you want high precision classifying between dresses, pants, and shirts, use this domain.|
+|__Compact domains__| Optimized for the constraints of real-time classification on edge devices.|
+
+## Object Detection
+
+|Domain|Purpose|
+|---|---|
+|__General__| Optimized for a broad range of object detection tasks. If none of the other domains are appropriate, or you are unsure of which domain to choose, select the Generic domain.|
+|__Logo__|Optimized for finding brand logos in images.|
+|__Products on shelves__|Optimized for detecting and classifying products on shelves.|
+|__Compact domains__| Optimized for the constraints of real-time object detection on edge devices.|
+
+## Compact domains
+
+The models generated by compact domains can be exported to run locally. Model performance varies by selected domain. In the table below, we report the model size and inference time on Intel Desktop CPU and NVidia GPU \[1\]. 
+
+> [!NOTE]
+> These numbers don't include preprocessing and postprocessing time.
+
+|Task|Domain|Model Size|CPU inference time|GPU inference time|
+|---|---|---|---|---|
+|Classification|General (compact)|5 MB|13 ms|5 ms|
+|Object Detection|General (compact)|45 MB|35 ms|5 ms|
+|Object Detection|General (compact) [S1]|14 MB|27 ms|7 ms|
+
+## VAIDK (Vision AI Dev Kit)
+
+When a compact domain is selected an extra option "Export Capabilities" is provided allowing for distinguishing between "Basic Platforms" and "Vision AI Dev Kit".
+
+Under _Export Capabilities_ the two options are:
+
+- Basic platforms (Tensorflow, CoreML, ONNX, etc.)
+- Vision AI Dev Kit.
+
+When _Vision AI Dev Kit_ is selected the _Generic_, _Landmarks_, and _Retail_ but not the _Food_ compact domains are available for Image Classification while both _General (compact)_ and _General (compact) [S1]_ are available for object detection.
+
+>[!NOTE]
+>__General (compact)__ domain for Object Detection requires special postprocessing logic. For the detail, please see an example script in the exported zip package. If you need a model without the postprocessing logic, use __General (compact) [S1]__.
+
+>[!IMPORTANT]
+>There is no guarantee that the exported models give the exactly same result as the prediction API on the cloud. Slight difference in the running platform or the preprocessing implementation can cause larger difference in the model outputs. For the detail of the preprocessing logic, please see [this document](python-tutorial.md).
+
+\[1\] Intel Xeon E5-2690 CPU and NVIDIA Tesla M60

articles/cognitive-services/Custom-Vision-Service/toc.yml

@@ -57,6 +57,8 @@
     href: test-your-model.md
   - name: Improve your classifier
     href: getting-started-improving-your-classifier.md
+  - name: Select a domain
+    href: select-domain.md
   - name: Use the prediction API
     href: use-prediction-api.md
   - name: Use Smart Labeler