DEEPCRAFT™ Studio Accelerators

📖 Overview

This repository contains DEEPCRAFT™ Studio Accelerators — deep learning based projects for various use-cases designed as starting points for building custom applications. These projects contain data and a project file that is ready to be used with DEEPCRAFT™ Studio for simplified Edge AI model development.

This repository is automatically pulled and content is generated in DEEPCRAFT™ Studio. For the best experience, access these models through DEEPCRAFT™ Studio.

For commercial use, our standard terms and conditions apply: https://developer.imagimob.com/legal/studio-terms-and-conditions.

🚀 Usage

These projects are designed to be used through DEEPCRAFT™ Studio and should be accessed through that platform. See also Studio's online documentation for more details.

When bringing your project into DEEPCRAFT™ Studio, consider the following:

1. Supported algorithms — Classification, Regression, and Object Detection
2. Data and labels — ensure they match the format Studio supports:
   • Classification and Regression — more info
   • Object Detection — more info
3. Data preprocessing — configure preprocessing in Studio:
   • Use available Studio layers — more info
   • Add your custom preprocessing layers — more info
4. Neural networks, layers, and functions — use only supported building blocks:
   • Classification and Regression — more info
   • Object Detection — more info

🤝 Contribution

All users are welcome to submit new models/projects, subject to the Infineon DEEPCRAFT™ Studio Accelerators review process.

How it works

1. 📁 Prepare your project — build your DEEPCRAFT™ Studio Accelerator locally. We recommend starting from _PROJECT_TEMPLATE, but you can also bring your own project. Complete your project files, README.md, and metadata.json. See Step 1 and Step 2 below for details.

---

2. 📤 Submit your project — use the PR tool to open a pull request against this repository. The tool validates your project layout and metadata, pushes your files to your fork, and opens the PR in your browser. See Step 3 and Step 4 below.

---

3. 🔍 Review — the Infineon team reviews your pull request. Automated pipelines may run to generate pre-processing, model predictions, and training outputs. Reviewers may request changes — address feedback by updating your project locally and re-running the PR tool to update the same pull request.

---

4. 🌐 Publication — once approved, your pull request is merged into main. The project is then published and becomes available through DEEPCRAFT™ Studio and the DEEPCRAFT™ AI Hub.

Submission requirements

Before opening a pull request, make sure you have the following tools and software:

• DEEPCRAFT™ Studio — to build and export your Accelerator project (.improj file and local Data/ folder)
• GitHub account — required to fork this repository and manage your pull request
• PR tool — the latest version from deepcraft-studio-accelerators-pr-tool; validates your project, pushes files, and opens the pull request
• Python 3.10+ — to run the PR tool (no extra packages required)
• Git — version 2.43 or newer (the PR tool uses it to manage your submission)
• GitHub CLI (gh) (optional) — for authentication; bundled with the PR tool on Windows. Install from cli.github.com only if you need it on other platforms or prefer a system-wide copy

📝 Submission Process

Follow the steps below to prepare and submit your project. For a high-level overview, see How it works in the Contribution section.

📁 Step 1 — Prepare your project

You can bring your own DEEPCRAFT™ Studio project, but we recommend using _PROJECT_TEMPLATE as a starting point — it provides the expected folder layout and files for submission.

If you use the template:

1. Rename your Accelerator project folder. The name can describe the use case and sensor used. Check existing project names and pick one that is not already in use.
2. Add content to the relevant folders and delete those that do not apply. The Data/ folder is mandatory and will not appear in this repository. Add custom folder(s) if needed.
3. Set up the provided project file example or replace it with your own .improj file.
4. Write the project README.md, including:
   • Use-case description
   • Sensor settings, specifications, and data description
   • Guidelines for collecting and expanding the dataset
   • Recommended path to production, including steps to make the model production-ready, with focus on reducing False Positives and/or False Negatives
5. Clean up — remove all README.md files from subfolders of _PROJECT_TEMPLATE if you used the template.

If you bring your own project instead, make sure it includes the required files (README.md, metadata.json, .improj, Data/) and follows the expected layout. Use _PROJECT_TEMPLATE as a reference.

---

📋 Step 2 — Prepare metadata.json

Choose one of the following options:

1. Guided (recommended) — when you run the PR tool (Step 4), it walks you through metadata collection interactively and writes metadata.json for you.
2. Manual — fill in metadata.json yourself using _PROJECT_TEMPLATE/metadata.json as a reference for the required fields and structure. The PR tool will validate your file when you run it.

---

🛠️ Step 3 — Get the PR tool

Get the pull request automation tool (PR tool) from the deepcraft-studio-accelerators-pr-tool repository.

Before submitting any project, make sure you are using the latest version of the PR tool — if you already have a copy, update it first (for example, run git pull in an existing clone, or download/clone the repository again).

You can obtain the tool in one of the following ways:

Option A — Download as ZIP

1. Open deepcraft-studio-accelerators-pr-tool on GitHub.
2. Click Code → Download ZIP, extract the archive, and use the pr_tool folder inside.

Option B — Clone the repository

git clone https://github.com/Infineon/deepcraft-studio-accelerators-pr-tool.git
cd deepcraft-studio-accelerators-pr-tool\pr_tool

Option C — Clone only the pr_tool folder (sparse checkout)

git clone --filter=blob:none --sparse https://github.com/Infineon/deepcraft-studio-accelerators-pr-tool.git
cd deepcraft-studio-accelerators-pr-tool
git sparse-checkout set pr_tool
cd pr_tool

---

🚀 Step 4 — Run the PR tool and submit

From the pr_tool folder, run:

python .\pr_tool.py --repo accelerators --path <project-path>

Replace <project-path> with the root path of your Studio Accelerator project. For more information, review the tool's README.md.

What happens next:

1. You will be prompted to authenticate with your GitHub account (required).
2. The tool forks this repository and prepares the pull request.
3. Your browser opens the pull request page — add the relevant details to aid the review process, then submit.

Updating an existing pull request — every time you change your project, re-run the same command above. Your existing pull request will be updated automatically.

Note: The pipeline will automatically generate pre-processing, model predictions, and train some models based on the default best model selection from DEEPCRAFT™ Studio. If you would not like this, specify in the pull request that it should not be published.