Skip to content

Latest commit

 

History

History
311 lines (256 loc) · 10.2 KB

EdgeApplicationSDK_FuncSpec_DevelopmentContainer.adoc

File metadata and controls

311 lines (256 loc) · 10.2 KB
Raw

Edge Application SDK
Development Container
Functional Specifications

TOC

1. Change history

Date What/Why

2022/11/16

Initial draft.

2023/01/30

Directory structure change. Added or modified features provided by the SDK. Updated the PDF build environment.

2023/05/26

Fixed the notation of tool names and parentheses.
Some environments do not render AsciiDoc’s Mermaid diagrams, so modified to rendered image references.
Added alternate text to images.

2. Terms/Abbreviations

Terms/Abbreviations Meaning

Dev Container

A Docker container with a software development environment available in GitHub Codespaces and VS Code. This SDK is provided based on Dev Container

"Cloud App"

AI application running in Cloud with data processed by "Edge Application" as input

4. Expected use case

  • Reduce the effort by using an environment that already contains the components needed for development

  • Use an environment that doesn’t depend on other environments

  • Teams use the same environment

  • Understand an overview of the entire workflow for AI application development by trying it out with sample code

  • Develop smoothly even without knowledge of AI application development

5. Functional overview/Algorithm

Functional overview

  • Provides a container environment for developing AI applications

    • The container environment can be used in the following ways:

      • Using Codespaces

        • There are two types of UI: Browser and VS Code desktop

      • Build a container environment on your Local PC and use it from VS Code

    • The container environment includes:

Note
 Specific details of each function contained in a container are described in the functional specifications of each function, not in this document.

  • Users can get information needed to develop AI applications

    • Users can view documentation for each workflow in AI application development

    • Users can view functional specifications

Others Exclusive conditions / specifications

  • No Docker images are included

  • No build environment for firmware of edge AI devices is included

  • Provides reference links for the sample "Cloud App"

AI application development workflows and features to provide

flowchart TD;
    %% definition
    classDef object fill:#FFE699, stroke:#FFD700
    classDef external_service fill:#BFBFBF, stroke:#6b8e23, stroke-dasharray: 10 2
    style legend fill:#FFFFFF, stroke:#000000

    %% impl
    subgraph legend["Legend"]
        process(Processing/User behavior)
    end
Loading 
flowchart TB
    id0((Start))
    id1(Project initial processing)
    id2(Prepare dataset)
    id3(Create an AI model)
    id4(Quantize an AI model)
    id5(Develop post-processing)
    id6("Deploy an AI model and post-processing")
    id7(Evaluation)
    id8(((Finish)))
    id0 -->id1
    id1 -->id2
    id2 -->id3
    id3 -->id4
    id4 -->id5
    id5 -->id6
    id6 -->id7
    id7 -->id8
Loading

Features provided by the SDK

Workflow Deliverables (documents) Deliverables (runtime environment, sample)

Project initial processing

  • "Console for AITRIOS" procedure

-

Prepare dataset

  • Procedure for annotating using CVAT

  • "Console for AITRIOS" procedure

  • Environment where CVAT can be used

  • Notebook to download images from COCO

  • Notebook importing and exporting datasets from CVAT

  • Notebook that splits datasets for train/validate

Create an AI model

  • Procedure for transfer learning an AI model (Image Classification)

  • "Console for AITRIOS" procedure

  • Sample notebook for transfer learning an AI model (Image Classification)

Quantize an AI model

  • Procedure for quantizing a user-created AI model (Image Classification) using "MCT"

  • Procedure for evaluating the accuracy of a user-created AI model (Image Classification) before and after quantization

  • Environment for quantization using "MCT"

  • AI model evaluation environment

  • Sample notebook quantizing an AI model (Image Classification)

  • Sample notebook evaluating the following types of AI models (Image Classification)

    • Keras

    • TFLite

    • TFLite (quantized)

Develop post-processing

  • Procedure to implement and debug post-processing and build it to a Wasm file

  • Environment to build post-processing to a Wasm file

  • Sample code for post-processing (C, C++)

  • Environment to debug post-processing code

Import an AI model and post-processing into "Console for AITRIOS"

  • Procedure to import AI models and post-processing into "Console for AITRIOS" using notebooks

  • "Console for AITRIOS" procedure

  • Notebook to import AI models and post-processing into "Console for AITRIOS"

Deploy an AI model and post-processing to edge AI devices

  • Procedure to deploy AI models and post-processing to edge AI devices using notebooks

  • "Console for AITRIOS" procedure

  • Notebook to deploy AI models and post-processing to edge AI devices

Evaluation

  • "Console for AITRIOS" procedure

-
Other features Deliverables (documents) Deliverables (runtime environment, sample)

Version control

  • Version control examples

-

Directory structure for the container

/tutorials
	/_common
	/1_initialize
	/2_prepare_dataset
	/3_prepare_model
	/4_prepare_application
	/5_evaluate
/docs/development-docs
/.devcontainer
/README.md

6. User interface specifications

Prerequisite

Start container

Start the development environment by the following procedure.

  • Codespaces (Browser)

    1. Press the [Create codespace on <branch name>] from the [Codespaces] tab of the [Code] in the SDK repository

  • Codespaces (VS Code desktop)

    1. Press the [Create codespace on <branch name>] from the [Codespaces] tab of the [Code] in the SDK repository

    2. After creating Codespace, press the [Codespaces] in the bottom left of the Codespace browser

    3. Select the [Open with VS Code] from the drop-down list

  • Local PC

    1. Access the SDK repository from GitHub, clone the SDK repository to your environment, and open it in VS Code

    2. Press the [><] mark at the bottom left of VS Code, or press the "Ctrl + Shift + P" to open the command palette and click the [Reopen in Container]

Note

To interrupt the container during startup, follow the procedure:

  • Codespaces (Browser)

    • Press the [×] button in your browser

  • Codespaces (VS Code desktop) or using VS Code on a Local PC

    • Press the VS Code’s [×] button
Note

To check container startup progress, follow the procedure:

  • Codespaces (Browser)

    • Press the [View logs] when it appears in the Codespaces browser

  • Codespaces (VS Code desktop) or using VS Code on a Local PC

    • Press the [Starting Dev Container (show log) ] from notification in bottom right of VS Code screen

Get information needed to develop AI applications

The following documents are available:

  • Procedure for each workflow of AI application development (README)

    1. Jump from the link in the repository top README.md to the README.md in the tutorials directory of the Directory structure for the container

    2. Jump from the link in the README.md in the tutorials directory to the README.md under each feature directory such as 1_initialize

  • Functional specifications

    1. Jump from the link in the repository top README.md to the functional specifications

7. Target performances/Impact on performances

  • Usability

    • When the SDK environment is built, the container is available for developing AI applications without any additional installation steps

    • Users must be able to navigate the container environment with the VS Code UI

8. Assumption/Restriction

  • Features provided by the SDK may not work properly depending on the specs of Codespaces or Local PC

    • For Codespaces, a Machine Type of 4-core or higher is recommended

9. Remarks

  • No error codes and messages are defined in the SDK

  • Does not specify the UI response time on container startup, as it is affected by the user’s network environment for Codespaces and the user’s Docker operating environment for Local PC

    • However, both Codespaces and Local PC have a proven UI response within 10 seconds on startup

      • Performance was measured under the following conditions:

        • Codespaces: Select Machine Type 4-core

        • Local PC: Start on a machine with the following specs:

Item Description

CPU

Intel® Core™ i7-8665U CPU @ 1.90GHz 2.11 GHz

RAM

16.0 GB

OS

Windows 10 version 21H2

WSL2

Ubuntu-20.04

10. Unconfirmed items

  • None