[FEATURE-REQUEST] Document Enhancement Request

[xiaolul]

When reading the document, we feel the following aspects are not clear enough. Some points mentioned below are not easy to find:

• An overview page that shows people e2e structure of IoT edge modules running on Azure Percept, including relationships between different modules, providing links to different resources.
• A well-structured documentation that gives instructions to developers and machine learning engineers (or data scientists) on how to add custom modules.
• A clear documentation that shows how to compile azureeyemodule end-to-end on Windows and Linux OS. Current document provided in github is not clear in the following aspects:
  • Differences between native compile and non-native compile?
  • When to use a native compile method and when to use non-native compile method?
  • how to revert (or fixed) if any steps go wrong? (reset?)
• The documentation doesn't point out modules to be stopped if re-deploy custom azureeyemodule is needed -- we need to stop both azureeyemodule and IoT agent
• The document doesn't tell the readers that rtsp:3000 is working when you customize it using a native way.
• The document doesn't state that the readers can check 8554 using a VLC player
• The document should clearly state expected drivers for external devices, for example, whether a driver for external monitor is expected.
• The document should be revisited and updated frequently based on the OpenVINO update, and check if the recommended approaches of converting models to blob are still valid.
• The document should state that it is possible to convert model on device using the built-in code provided by azureeyemodule, along with the required model files and format. Currently, when reading the code, we understand it is possible to do so, but no instructions or sample model files are provided.
• It would be nice if the document provided a reference link to OpenCV G-API
• It would be nice if the document shows the structure of the model zip file required by azureeyemodule.
• It would be nice if the hardware spec could also provide the GFLOPs besides TOPs. Or provide latency benchmark on the default models provided in azureeyemodule repo.