From d6ef6cba9dc73737ee66978f8cf9e180b4716626 Mon Sep 17 00:00:00 2001 From: Angelo_Di_Marco Date: Thu, 25 Jun 2026 17:58:49 +0200 Subject: [PATCH 1/3] updated README and metadata.json in _PROJECT_TEMPLATE to account for new fields and the new PR Tool; improved README structure and readability to be more user friendly --- README.md | 161 +++++++++++++++++++++----------- _PROJECT_TEMPLATE/metadata.json | 19 ++-- 2 files changed, 120 insertions(+), 60 deletions(-) diff --git a/README.md b/README.md index 982a0d5ce0..7f61d51385 100644 --- a/README.md +++ b/README.md @@ -1,71 +1,124 @@ # DEEPCRAFT™ Studio Accelerators -This repository contains DEEPCRAFT™ Studio Accelerators - deep learning based projects for various use-cases designed as starting points for building custom applications. These projects contains 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. Download it from [here](https://softwaretools.infineon.com/assets/com.ifx.tb.tool.deepcraftstudio) +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](https://softwaretools.infineon.com/assets/com.ifx.tb.tool.deepcraftstudio) for simplified Edge AI model development. -For commercial use, our standard terms and conditions applies, https://developer.imagimob.com/legal/studio-terms-and-conditions. +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 -## Usage These projects are designed to be used through [DEEPCRAFT™ Studio](https://www.imagimob.com/studio) and should be accessed through that platform. See also Studio's [online documentation](https://developer.imagimob.com/) for more details. -To consider when bringing your project into DEEPCRAFT™ Studio : -- Check the available deep learning algorithms: - - Classification - - Regression - - Object Detection -- Make sure data and labels are in the format that Studio supports: - - Classification and Regression - [more info](https://developer.imagimob.com/deepcraft-studio/data-preparation/data-collection/bring-your-data/bring-your-own-data) - - Object Detection - [more info](https://developer.imagimob.com/deepcraft-studio/data-preparation/data-collection/bring-your-data/bring-your-own-data-object-detection) -- Set up the data preprocessing in Studio: - - Use available Studio layers - [more info](https://developer.imagimob.com/deepcraft-studio/preprocessing) - - Add your custom preprocessing layers - [more info](https://developer.imagimob.com/deepcraft-studio/deployment/custom-layers-functions) -- Use the supported neural networks, layers and functions: - - Classification and Regression - [more info](https://developer.imagimob.com/deepcraft-studio/deployment/supported-layers) - - Object detection - [more info](https://developer.imagimob.com/deepcraft-studio/model-training/training-object-detection) - - -## Contribution +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](https://developer.imagimob.com/deepcraft-studio/data-preparation/data-collection/bring-your-data/bring-your-own-data) + - Object Detection — [more info](https://developer.imagimob.com/deepcraft-studio/data-preparation/data-collection/bring-your-data/bring-your-own-data-object-detection) +3. **Data preprocessing** — configure preprocessing in Studio: + - Use available Studio layers — [more info](https://developer.imagimob.com/deepcraft-studio/preprocessing) + - Add your custom preprocessing layers — [more info](https://developer.imagimob.com/deepcraft-studio/deployment/custom-layers-functions) +4. **Neural networks, layers, and functions** — use only supported building blocks: + - Classification and Regression — [more info](https://developer.imagimob.com/deepcraft-studio/deployment/supported-layers) + - Object Detection — [more info](https://developer.imagimob.com/deepcraft-studio/model-training/training-object-detection) + +## 🤝 Contribution + All users are welcome to submit new models/projects, subject to the Infineon DEEPCRAFT™ Studio Accelerators review process. -## Submission Process -To submit a project, create a pull request with your data and DEEPCRAFT™ Studio project file (.improj) using the automation tool provided below. - -Use the available `_PROJECT_TEMPLATE` to structure your project: -* Rename your accelerator project folder. For instance, the name can contain the use case and the sensor used in the project. Check the project names of the already available projects. Make sure to pick up a name which is not been already used -* Add content to the relevant folders and delete the ones which do not apply to your project. Data folder is mandatory and it will not show up in this repository. Add your custom folder(s) if needed -* Set up the provided project file example or replace it with your own project file -* Add content to the project `README.md` file making sure to include the following information: - - 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 -* Before the submission - - Make sure to remove all the `README.md` files contained in all subfolders of the `_PROJECT_TEMPLATE` if you use it - - Fill in the fields in the `metadata.json` file as follows: - - `title` (max 40 characters): give a title to your project making sure it does not exist already. For instance, use words describing the use case and sensor. Get inspired by the existing ones in DEEPCRAFT™ Studio. - - `description` (max 100 characters): briefly describe your project. Get inspired by the existing ones in DEEPCRAFT™ Studio. - - `algorithm`: choose between **Classification** or **Regression** - - `sensors`: specify the sensor used in you project. Choose from the existing ones in DEEPCRAFT™ Studio: **IMU & Vibration**, **Microphone**, **Capacitive & Inductive Sensing**, **Camera**, etc. - - `domain`: Specify the domain(s) of your project: **Audio**, **Voice**, **Vision**, etc. - - `application`: Specify the application(s) of your project: **Smart Home**, **Smart TV**, **Appliances**, **Wearables**, **Games**, etc. - - `use_case`: Specify the use case(s) of your project: **Object Detection**, **Voice Control**, **Speech Recognition**, etc. - - `kit`: Specify the Infineon's kit(s) your project is compatible with: **PSOC™ 6 Pioneer Kit**, **PSOC™ 6 AI Kit**, **PSOC™ Edge AI Kit**, etc. - - `device`: Specify the device(s) your project is compatible with: **PSOC™ 4**, **PSOC™ 6**, **PSOC™ Edge**, **AURIX**, etc. - - `workflow`: Use **ML Development** - -Once the project is ready, you can download the pull request automation tool or PR tool [pr_tool.zip](https://api.imagimob.com/v1/Data/Object/pr_tool.zip) and run: +### How it works + +1. **Prepare your project** — build your DEEPCRAFT™ Studio accelerator locally. We recommend starting from [`_PROJECT_TEMPLATE`](_PROJECT_TEMPLATE), but you can also bring your own project. Complete your project files, `README.md`, and `metadata.json`. See [Step 1](#step-1--prepare-your-project) and [Step 2](#step-2--prepare-metadatajson) below for details. +2. **Submit your project** — use the [PR tool](https://github.com/Infineon/deepcraft-studio-accelerators-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](#step-3--get-the-pr-tool) and [Step 4](#step-4--run-the-pr-tool-and-submit) 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](https://deepcraft.infineon.com). + +### Submission requirements + +Before opening a pull request, make sure you have the following tools and software: + +- **[DEEPCRAFT™ Studio](https://softwaretools.infineon.com/assets/com.ifx.tb.tool.deepcraftstudio)** — to build and export your accelerator project (`.improj` file and local `Data/` folder) +- **[GitHub account](https://github.com/join)** — required to fork this repository and manage your pull request +- **[PR tool](https://github.com/Infineon/deepcraft-studio-accelerators-pr-tool)** — the latest version from [deepcraft-studio-accelerators-pr-tool](https://github.com/Infineon/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](https://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](#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`](_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`](_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`](_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](https://github.com/Infineon/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](https://github.com/Infineon/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** ```bash -tar -xf \pr_tool.zip +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)** + +```bash +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 -python .\pr_tool.py --path ``` -where `` is the root path of the studio accelerator project. For more information review the tools' `README.md` file. +### Step 4 — Run the PR tool and submit + +From the `pr_tool` folder, run: + +```bash +python .\pr_tool.py --repo accelerators --path +``` + +Replace `` with the root path of your studio accelerator project. For more information, review the tool's [README.md](https://github.com/Infineon/deepcraft-studio-accelerators-pr-tool/blob/main/pr_tool/README.md). -Please be aware that you will need a GitHub Account. When you run the tool using the command shown above it will authenticate using your GitHub account, fork this repository and prepare the pull request. Once ready, it will open the pull request in a window in your browser. Please add the relevant detail requested to complete your pull request which will aid in the review process and then submit. +What happens next: -Please consider that every time you update your project you need to run the Python script `pr_tool.py` as shown above and your existing pull request will be updated. +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. -***NOTE:*** The pipeline will automatically generate the pre-processing, model predictions and train some models based on the default best model selection from DEEPCRAFT™ Studio. If you would not like to have this, then please specify in the pull request if that should not be what should be published. +**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. diff --git a/_PROJECT_TEMPLATE/metadata.json b/_PROJECT_TEMPLATE/metadata.json index d69f3b72ce..34522d6f94 100644 --- a/_PROJECT_TEMPLATE/metadata.json +++ b/_PROJECT_TEMPLATE/metadata.json @@ -1,12 +1,19 @@ { "title": "Project Title MAX 40 characters", "description": "Project description MAX 100 characters", - "algorithm": "Choose between Classification or Regression", + "long_description": "Detailed project description: use case, sensor setup, data collection guidelines, and recommended path to production.", + "algorithm": "Choose between Classification, Regression, or Object Detection", "sensors": ["Specify sensor(s) used in your project: Microphone, IMU, Camera, etc."], "domain": ["Specify domain(s) of your project: Audio, Voice, Vision, etc."], - "application": ["Specify application(s) of your project: Smart Home, Smart TV, Appliances, Wearables, Games, etc."], - "use_case": ["Specify use case(s) of your project: Object Detection, Voice Control, Speech Recognition, etc."], - "kit": ["Specify Infineon kit(s) your project is compatible with: PSoC-6 Pioneer Kit, PSoC-6 AI Kit, PSoC-Edge AI Kit, etc."], - "device": ["Specify device(s) your project is compatible with: PSoC-4, PSoC-6, PSoC-Edge, AURIX, etc."], - "workflow": ["Choose between ML Development and/or ML Deployment"] + "application": ["Specify application(s) of your project: Smart Home, Wearables, Appliances, etc."], + "use_case": ["Specify use case(s) of your project: Speech Recognition, Voice Control, Anomaly Detection, etc."], + "kit": ["Specify Infineon kit(s) your project is compatible with: PSOC\u2122 6 Pioneer Kit, PSOC\u2122 6 AI Kit, PSOC\u2122 Edge AI Kit, etc."], + "device": ["Specify device(s) your project is compatible with: PSOC\u2122 4, PSOC\u2122 6, PSOC\u2122 Edge, AURIX\u2122, etc."], + "type": ["Choose between Model Development and/or Model Deployment"], + "workflow": ["Derived from type: ML Development for Model Development, ML Deployment for Model Deployment"], + "thumbnail_image_id": "Image file name for the project thumbnail. Can be selected using the PR Tool.", + "main_image_id": "Image file name for the main project image — usually the same as thumbnail_image_id. Can be set up using the PR Tool.", + "brand_image_id": "Brand logo file name (e.g. deepcraft.webp for Infineon). Can be set up using the PR Tool.", + "brand_url": "Brand URL (e.g. https://www.infineon.com/design-resources/embedded-software/deepcraft-edge-ai-solutions for Infineon). Can be set up using the PR Tool.", + "links": ["Set up automatically by the PR Tool (Download and GitHub links for your project)"] } From b914c863893de2c9f3b7e7173fed453b575d4646 Mon Sep 17 00:00:00 2001 From: Angelo_Di_Marco Date: Thu, 25 Jun 2026 21:44:23 +0200 Subject: [PATCH 2/3] updated README icons to match Studio units and PSOC repos --- README.md | 37 +++++++++++++++++++++++++++---------- 1 file changed, 27 insertions(+), 10 deletions(-) diff --git a/README.md b/README.md index 7f61d51385..436c09b50a 100644 --- a/README.md +++ b/README.md @@ -1,12 +1,14 @@ # 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](https://softwaretools.infineon.com/assets/com.ifx.tb.tool.deepcraftstudio) 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 +## 🚀 Usage These projects are designed to be used through [DEEPCRAFT™ Studio](https://www.imagimob.com/studio) and should be accessed through that platform. See also Studio's [online documentation](https://developer.imagimob.com/) for more details. @@ -29,10 +31,19 @@ All users are welcome to submit new models/projects, subject to the Infineon DEE ### How it works -1. **Prepare your project** — build your DEEPCRAFT™ Studio accelerator locally. We recommend starting from [`_PROJECT_TEMPLATE`](_PROJECT_TEMPLATE), but you can also bring your own project. Complete your project files, `README.md`, and `metadata.json`. See [Step 1](#step-1--prepare-your-project) and [Step 2](#step-2--prepare-metadatajson) below for details. -2. **Submit your project** — use the [PR tool](https://github.com/Infineon/deepcraft-studio-accelerators-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](#step-3--get-the-pr-tool) and [Step 4](#step-4--run-the-pr-tool-and-submit) 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](https://deepcraft.infineon.com). +1. 📁 **Prepare your project** — build your DEEPCRAFT™ Studio accelerator locally. We recommend starting from [`_PROJECT_TEMPLATE`](_PROJECT_TEMPLATE), but you can also bring your own project. Complete your project files, `README.md`, and `metadata.json`. See [Step 1](#step-1--prepare-your-project) and [Step 2](#step-2--prepare-metadatajson) below for details. + +--- + +2. 📤 **Submit your project** — use the [PR tool](https://github.com/Infineon/deepcraft-studio-accelerators-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](#step-3--get-the-pr-tool) and [Step 4](#step-4--run-the-pr-tool-and-submit) 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](https://deepcraft.infineon.com). ### Submission requirements @@ -45,11 +56,11 @@ Before opening a pull request, make sure you have the following tools and softwa - **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](https://cli.github.com/) only if you need it on other platforms or prefer a system-wide copy -## 📤 Submission Process +## 📝 Submission Process Follow the steps below to prepare and submit your project. For a high-level overview, see [How it works](#how-it-works) in the Contribution section. -### Step 1 — Prepare your project +### 📁 Step 1 — Prepare your project You can bring your own DEEPCRAFT™ Studio project, but we **recommend using [`_PROJECT_TEMPLATE`](_PROJECT_TEMPLATE)** as a starting point — it provides the expected folder layout and files for submission. @@ -67,14 +78,18 @@ If you use 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`](_PROJECT_TEMPLATE) as a reference. -### Step 2 — Prepare `metadata.json` +--- + +### 📋 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`](_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 +--- + +### 🛠️ Step 3 — Get the PR tool Get the pull request automation tool (PR tool) from the [deepcraft-studio-accelerators-pr-tool](https://github.com/Infineon/deepcraft-studio-accelerators-pr-tool) repository. @@ -103,7 +118,9 @@ git sparse-checkout set pr_tool cd pr_tool ``` -### Step 4 — Run the PR tool and submit +--- + +### 🚀 Step 4 — Run the PR tool and submit From the `pr_tool` folder, run: From e12c71fb226dc27c04d67e75700c250b8e25b152 Mon Sep 17 00:00:00 2001 From: Angelo_Di_Marco Date: Fri, 26 Jun 2026 11:15:54 +0200 Subject: [PATCH 3/3] removed from READMEs any reference to Starter Models and duplicated Help/Support sections and minor fixes --- AnomalousVibrationDetection/README.md | 16 +- .../Tools/IMUDataCollectonProject/README.md | 2 +- BabyCryDetection/README.md | 4 +- ChainsawDetection/README.md | 4 +- DrillMaterialDetectionIMU/README.md | 8 +- DrillMaterialDetectionMicrophone/README.md | 8 +- FallDetection/README.md | 4 +- GlassBreakDetection/README.md | 96 +++--- GroundObstacleDetectionTOF/README.md | 10 +- GunshotDetection/README.md | 6 +- HomeSoundsDetection/README.md | 6 +- HumanActivity/README.md | 2 +- KeywordDetector/README.md | 2 +- MotorImbalanceDetection/README.md | 6 +- MovementTypeDetection/README.md | 8 +- README.md | 282 +++++++++--------- SirenDetection/README.md | 2 +- SurfaceDetectionMicrophone/README.md | 10 +- .../Tools/DataCollectionGraphUX/README.md | 2 +- TermiteDetection/README.md | 6 - TouchDetection/README.md | 6 +- VirtualWindingTemperatureSensing/README.md | 6 +- .../Tools/README.md | 4 +- 23 files changed, 244 insertions(+), 256 deletions(-) diff --git a/AnomalousVibrationDetection/README.md b/AnomalousVibrationDetection/README.md index 285201240d..31335b9538 100644 --- a/AnomalousVibrationDetection/README.md +++ b/AnomalousVibrationDetection/README.md @@ -1,17 +1,17 @@ -# Anomaly Detection for Vibrating Machinery - Starter Model Project +# Anomaly Detection for Vibrating Machinery - Studio Accelerator Project This project is designed to work exclusively with DEEPCRAFT™ Studio. Download it from [here](https://softwaretools.infineon.com/assets/com.ifx.tb.tool.deepcraftstudio) ## Use-case description -This starter model aims to provide general guidance on how to develop an **anomaly detection system** for detecting anomalous behavior in machinery based on vibration measurements. +This Accelerator project aims to provide general guidance on how to develop an **anomaly detection system** for detecting anomalous behavior in machinery based on vibration measurements. This project will monitor a simple desktop fan, but the same concept and workflow can be easily ported to any other machinery, whether industrial or consumer. The task is framed as a **classification project**: a type of Supervised Learning where a model learns to classify data into a discrete number of classes. For this project, only two classes will be used: normal functioning and anomalous functioning. You will need to provide both normal and anomalous data for the machinery you want to monitor to build a robust classifier. ### How can I know if this project fits my use case? -You can use this starter project if: +You can use this Accelerator project if: - You need to monitor a machinery whose behavior can be inferred by its vibration; - You have the possibility of collecting both normal functioning data and anomalous functioning data, either from an already faulty machine or by artificially inducing anomalies on a functioning machine. @@ -41,7 +41,7 @@ This project demonstrates how to approach classification-based vibration monitor ## Sensor settings specification -This starter project requires the [PSOC™ 6 AI Evaluation Kit](https://www.infineon.com/cms/en/product/evaluation-boards/cy8ckit-062s2-ai/). This platform is equipped with PSoC™ 6 MCU and IMU sensors. The board is designed for easy prototyping and lets you collect real-life data to easily build a compelling ML product fast. +This Accelerator project requires the [PSOC™ 6 AI Evaluation Kit](https://www.infineon.com/cms/en/product/evaluation-boards/cy8ckit-062s2-ai/). This platform is equipped with PSoC™ 6 MCU and IMU sensors. The board is designed for easy prototyping and lets you collect real-life data to easily build a compelling ML product fast. The desktop fan is optional; you may want to collect data directly from your machinery instead. However, if you want to replicate the project out-of-the-box with a small desktop fan, any inexpensive product similar to the one shown will be suitable: @@ -77,7 +77,7 @@ Once you have completed data collection, you can save the sample in the `Data` f ### A note on data labeling This project uses only one label to frame the task as a binary classification problem. -Note that Deepcraft Studio introduces an "Unlabelled data" class by default, which we will use as "Normal" behavior data. +Note that DEEPCRAFT™ Studio introduces an "Unlabelled data" class by default, which we will use as "Normal" behavior data. The only additional label needed is "anomaly", which represents anomalous data. **Anomaly**: This label indicates that the machinery is operating anomalously. @@ -126,8 +126,8 @@ More in detail, the steps to be followed could look like this: **3. Import your data and train the prototype model** - Import the data you collected in the "Data" tab of the .improj file in Deepcraft Studio. - You are now able to follow the standard Deepcraft Studio steps for processing, training, and deploying your Anomaly Detection model. + Import the data you collected in the "Data" tab of the .improj file in DEEPCRAFT™ Studio. + You are now able to follow the standard DEEPCRAFT™ Studio steps for processing, training, and deploying your Anomaly Detection model. The preprocessor is already set, and some models are already defined for you, which performance is guaranteed to be in real-time on the PSOC6 AI Kit. **4. Deploy and do a real-time test of your prototype model** @@ -138,7 +138,7 @@ More in detail, the steps to be followed could look like this: Last step is to move to the actual final production setup. The production system will likely have the MCU placed on a board inside the machine and the IMU sensor in a specific position, not necessarly the same one of the prorotyping phase. If you can go as close as possible to production conditions during prototyping phase, you will be able to deliver the same model also on the production board with little-to-no additional training or data needed. If this is not the case, you might need to do a new data collection step to allow the model to learn the nuances of the final setup. Follow again steps 2, 3 and 4 also for the production setup to reach a functioning application. -You may also leverage Deepcraft Studio's Transfer Learning features for fine-tuning the prototype model to production data. This could lead to better results and faster go-to-production times, but the usage of Transfer Learning is recommended only to experienced ML users. +You may also leverage DEEPCRAFT™ Studio's Transfer Learning features for fine-tuning the prototype model to production data. This could lead to better results and faster go-to-production times, but the usage of Transfer Learning is recommended only to experienced ML users. **Note:** All subsequent ML system lifetime monitoring procedures must be defined and implemented by you according to you needs, requirements and targets. diff --git a/AnomalousVibrationDetection/Tools/IMUDataCollectonProject/README.md b/AnomalousVibrationDetection/Tools/IMUDataCollectonProject/README.md index efec887508..f87e0b4c4f 100644 --- a/AnomalousVibrationDetection/Tools/IMUDataCollectonProject/README.md +++ b/AnomalousVibrationDetection/Tools/IMUDataCollectonProject/README.md @@ -2,7 +2,7 @@ ## Overview -This starter project shows you how to collect and annotate IMU data live. This can be done directly from your PSOC 6 AI Kit attached over USB-serial. +This Studio project shows you how to collect and annotate IMU data live. This can be done directly from your PSOC 6 AI Kit attached over USB-serial. The graph that you see in the Main.imunit contains input/data source nodes representing the device connected through the serial port. diff --git a/BabyCryDetection/README.md b/BabyCryDetection/README.md index 7806e49eeb..a80e187f4f 100644 --- a/BabyCryDetection/README.md +++ b/BabyCryDetection/README.md @@ -6,7 +6,7 @@ This project is designed to work exclusively with DEEPCRAFT™ Studio. Download ## Overview -This starter project allows you to build a baby cry detector that can be used on any supported Infineon MCU with a microphone. Everything is included to allow you to expand on the project to bring it to production on your own. +This Accelerator project allows you to build a baby cry detector that can be used on any supported Infineon MCU with a microphone. Everything is included to allow you to expand on the project to bring it to production on your own. Below you can find code examples about how to deploy the output of this project to any supported Infineon MCU with a microphone. @@ -28,7 +28,7 @@ The project has the following classes: ## Taking the Project Further -This project is only a starter project and as such some work is needed to further develop this project. Such as including more data from home environments as well as some data You can take the project further in a number of different ways: +This project is only an Accelerator project and as such some work is needed to further develop this project. Such as including more data from home environments as well as some data You can take the project further in a number of different ways: 1. You can add additional classes to the existing ones by adding the relevant data, for example, children talking, children playing, different ages etc. 2. Add your own recorded data to the dataset and see if you can improve the performance of the provided model. diff --git a/ChainsawDetection/README.md b/ChainsawDetection/README.md index d0d44f6139..bae65f3fca 100644 --- a/ChainsawDetection/README.md +++ b/ChainsawDetection/README.md @@ -3,7 +3,7 @@ This project is designed to work exclusively with DEEPCRAFT™ Studio. Download it from [here](https://softwaretools.infineon.com/assets/com.ifx.tb.tool.deepcraftstudio) ## Overview -This is a starter model that classifies if there is an actively cutting chainsaw in the vicinity; chainsaws that are stalling are defined as not cutting. +This is an Accelerator project that classifies if there is an actively cutting chainsaw in the vicinity; chainsaws that are stalling are defined as not cutting. A fully developed model could be used to detect illegal logging or create automatic warning systems. ## Collection of Data @@ -14,7 +14,7 @@ The data was collected at 16000Hz, and the project contains around 700 minutes o After a preliminary evaluation, the model performed very poorly on chainsaw audio played through a speaker, making it harder to demo. As such, additional data was collected by playing chainsaw audio through a variety of speakers to supplement the dataset, after which it significantly improved on sounds played through speakers. ## Adding More Data -Adding more background noise data can be done online or by collecting microphone data of a suitable environment (i.e., forest, construction site, river). This can then be imported into the studio project. +Adding more background noise data can be done online or by collecting microphone data of a suitable environment (i.e., forest, construction site, river). This can then be imported into the Studio project. Adding more chainsaw data ought to be done through a thorough collection. Adding variation of different types of chainsaws, different types of trees, and varying the distance from the microphone is strongly recommended, with the distances being most important. ## Steps to Production diff --git a/DrillMaterialDetectionIMU/README.md b/DrillMaterialDetectionIMU/README.md index f72094bb33..03a29fe7f4 100644 --- a/DrillMaterialDetectionIMU/README.md +++ b/DrillMaterialDetectionIMU/README.md @@ -4,24 +4,24 @@ This project is designed to work exclusively with DEEPCRAFT™ Studio. Download ## Overview -This is a starter model that is capable of classifying the material a power drill is drilling into based on the IMU (6-axis accelerometer and gyroscope) signature and is imagined to be incorporated into smart power tools. +This is an Accelerator project that is capable of classifying the material a power drill is drilling into based on the IMU (6-axis accelerometer and gyroscope) signature and is imagined to be incorporated into smart power tools. A similar model was developed using the audio data instead, if that is more fitting to your use case. It is developed as a proof of concept and is not fully optimized, achieving around a 90% plastic/wood accuracy with a 50% air/none accuracy. This large fault can, however, be mitigated by adding an additional data stream stating whether the drill is on or not. Furthermore, the current project only differentiates between wood, plastic, and air but is easily scalable to include more labels. The preprocessor and model architectures can be used not only for a drill but for any number of motor-based projects where there is a variability in the IMU output based on the desired classification labels. ## Collection of Data -The data was collected from the built-in IMU on the AI-Eval kit (CY8CKIT-062S2-AI) taped to the left top side of a drill and streamed directly into DEEPCRAFT Studio at 50Hz. +The data was collected from the built-in IMU on the AI-Eval kit (CY8CKIT-062S2-AI) taped to the left top side of a drill and streamed directly into DEEPCRAFT™ Studio at 50Hz. The drilling was mainly done straight down into an 11mm thick plank of wood and a 3mm thick piece of acrylic plastic, with around 5% of the data being horizontal drilling. Not every hole was made all the way through the material in order to have data for such cases. Apart from drilling into air, wood, and plastic, the drill was moved around in order to create an invariance to general movement. There are additional labels for the moment of removing the drill from the material, called wood_out and plastic_out. Further thought towards the necessity of these labels is recommended; it might be better to just train a model that classifies this as regular wood and plastic. ## Adding More Data -In order to add more data-be it similar, from another type of drill, or another material entirely-you simply need to attach an AI-eval kit (purchasable here: https://www.infineon.com/cms/en/product/evaluation-boards/cy8ckit-062s2-ai/) to a drill and stream the IMU data of drilling into DEEPCRAFT Studio and then label it appropriately. +In order to add more data-be it similar, from another type of drill, or another material entirely-you simply need to attach an AI-eval kit (purchasable here: https://www.infineon.com/cms/en/product/evaluation-boards/cy8ckit-062s2-ai/) to a drill and stream the IMU data of drilling into DEEPCRAFT™ Studio and then label it appropriately. It is recommended to have a minimum of 100 seconds of data per label, preferably more. ## Steps to Production -The recommended path to production for this starter model is to identify which materials you want to differentiate from, which drill types you want this to work on, and which conditions your application will be placed in. +The recommended path to production for this Accelerator project is to identify which materials you want to differentiate from, which drill types you want this to work on, and which conditions your application will be placed in. It is unlikely that keeping non-matching existing data worsens the model so long as your application isn't wildly different. After obtaining the appropriate machines and materials, you should perform data collection as outlined above, making sure to collect data including all eventualities you want invariance to. For example, if you want a model that classifies when a drill is entering a new material, you will want to collect data on many different material combinations, using many different types of drills, held by many different people, in many different settings. diff --git a/DrillMaterialDetectionMicrophone/README.md b/DrillMaterialDetectionMicrophone/README.md index 066b50ad5d..044f34e215 100644 --- a/DrillMaterialDetectionMicrophone/README.md +++ b/DrillMaterialDetectionMicrophone/README.md @@ -4,24 +4,24 @@ This project is designed to work exclusively with DEEPCRAFT™ Studio. Download ## Overview -This is a starter model that is capable of classifying the material a power drill is drilling into based on the audio signature and is imagined to be incorporated into smart power tools. +This is an Accelerator project that is capable of classifying the material a power drill is drilling into based on the audio signature and is imagined to be incorporated into smart power tools. A similar model was developed using the IMU data instead, if that is more fitting to your use case. It is developed as a proof of concept and is not fully optimized, achieving around an 85% plastic/wood accuracy. Furthermore, the current project only differentiates between wood, plastic, and air but is easily scalable to include more labels. The preprocessor and model architectures can be used not only for a drill but for any number of motor-based projects where there is a variability in the audio signature based on the desired classification labels. ## Collection of Data -The data was collected from the built-in microphone on the AI-Eval kit (CY8CKIT-062S2-AI) taped to the left top side of a drill and streamed directly into DEEPCRAFT Studio at 16000Hz. +The data was collected from the built-in microphone on the AI-Eval kit (CY8CKIT-062S2-AI) taped to the left top side of a drill and streamed directly into DEEPCRAFT™ Studio at 16000Hz. The drilling was mainly done straight down into an 11mm thick plank of wood and a 3mm thick piece of acrylic plastic, with around 5% of the data being horizontal drilling. Not every hole was made all the way through the material in order to have data for such cases. Apart from drilling into air, wood, and plastic, the drill was moved around in order to create an invariance to general movement. There are additional labels for the moment of removing the drill from the material, called wood_out and plastic_out. Further thought towards the necessity of these labels is recommended; it might be better to just train a model that classifies this as regular wood and plastic. ## Adding More Data -In order to add more data-be it similar, from another type of drill, or another material entirely-you simply need to attach an AI-eval kit (purchasable here: https://www.infineon.com/cms/en/product/evaluation-boards/cy8ckit-062s2-ai/) to a drill and stream the audio data of drilling into DEEPCRAFT Studio and then label it appropriately. +In order to add more data-be it similar, from another type of drill, or another material entirely-you simply need to attach an AI-eval kit (purchasable here: https://www.infineon.com/cms/en/product/evaluation-boards/cy8ckit-062s2-ai/) to a drill and stream the audio data of drilling into DEEPCRAFT™ Studio and then label it appropriately. It is recommended to have a minimum of 100 seconds of data per label, preferably more. ## Steps to Production -The recommended path to production for this starter model is to identify which materials you want to differentiate from, which drill types you want this to work on, and which conditions your application will be placed in. +The recommended path to production for this Accelerator project is to identify which materials you want to differentiate from, which drill types you want this to work on, and which conditions your application will be placed in. It is unlikely that keeping non-matching existing data worsens the model so long as your application isn't wildly different. After obtaining the appropriate machines and materials, you should perform data collection as outlined above, making sure to collect data including all eventualities you want invariance to. For example, if you want a model that classifies when a drill is entering a new material, you will want to collect data on many different material combinations, using many different types of drills, held by many different people, in many different settings. diff --git a/FallDetection/README.md b/FallDetection/README.md index 68b11aa1e6..cd9dcb2f9a 100644 --- a/FallDetection/README.md +++ b/FallDetection/README.md @@ -4,8 +4,8 @@ This project is designed to work exclusively with DEEPCRAFT™ Studio. Download ## Overview -This starter model allows you to build models to detect a fall using an IMU (Interial Mesurement Unit - accelerometer and gyroscope) mounted on the buckle of a belt. -For that, this starter model uses data collected from 2 different IMU: a Bosh IMU and an ST-Microelectronics IMU. Both IMU sensors are set up to collect data at 50 Hz using a +- 8g for the accelerometer scale and +- 500 dps for the gyro scale. +This Accelerator project allows you to build models to detect a fall using an IMU (Interial Mesurement Unit - accelerometer and gyroscope) mounted on the buckle of a belt. +For that, this Accelerator project uses data collected from 2 different IMU: a Bosh IMU and an ST-Microelectronics IMU. Both IMU sensors are set up to collect data at 50 Hz using a +- 8g for the accelerometer scale and +- 500 dps for the gyro scale. This project gives you the infrastructure to allow you to expand on the project by adding other events to detect or adding more data and make the model production ready. diff --git a/GlassBreakDetection/README.md b/GlassBreakDetection/README.md index 0a70dc749b..2525ab26a0 100644 --- a/GlassBreakDetection/README.md +++ b/GlassBreakDetection/README.md @@ -1,48 +1,48 @@ -# Glass Break Detection - -This project is designed to work exclusively with DEEPCRAFT™ Studio. Download it from [here](https://softwaretools.infineon.com/assets/com.ifx.tb.tool.deepcraftstudio) - -## Overview - -This machine learning project focuses on developing a glass break detection model that can accurately identify glass break events by analyzing audio signals. -Such models play a critical role in modern security infrastructures, where they enhance situational awareness and enable rapid response. - -Glass break detection can be deployed across diverse environments: - -- Home security systems: safeguarding families and property against intrusions. - -- Commercial and retail buildings: protecting assets and ensuring customer safety. - -- Vehicles: detecting accidents or break-ins to trigger immediate alerts and safety measures. - -By combining audio signal processing with machine learning the model provides accurate event classification, minimizes false alarms, -and supports intelligent safety and security management across multiple domains. This model enhances the reliability of traditional alarm systems while also serving as a foundation for advanced IoT‑based security solutions that intelligently adapt to different environments and conditions. -## Contents - -`Data` - Folder containing the glass break audio wav files used in this project. - -`Models` - Folder where trained models, their predictions and generated Edge code are saved. - -## Collection of Data -The data for this project was primarily collected from the Freesound site, with careful verification of licensing to ensure proper eligibility for use in model training. Infineon also contributed the data providing samples that strengthen -the reliability and diversity of the training set. All audio recordings in the dataset have a sampling frequency of 48 kHz, ensuring consistency and high-resolution quality for signal analysis and model training. - - -## Adding More Data -The project currently includes 369 audio WAV files, forming the foundation for model training. While this provides a solid starting point, additional recordings are needed to broaden coverage. -In particular, expanding the dataset with more diverse glass break samples as well as recordings captured under noisier conditions and varying distances will help improve the model’s robustness and generalization to real-world environments. -Deepcraft Studio provides a streamlined environment for recording glass break audio samples and offers integrated tools for labeling and preprocessing, ensuring that data preparation is both efficient and consistent. While studio recordings offer clean baseline samples, they should be complemented with field recordings -or augmented with synthetic noise to ensure the model performs reliably in real-world scenarios. - -## Steps to Production -The recommended path to production for a glass break detection model begins with clearly defining the environments and glass types you want the -system to handle such as residential windows, retail storefronts, or automotive glass. Once the scope is set, data collection should be performed -across diverse scenarios ensuring coverage of different glass materials, break mechanisms, microphone placements, and acoustic conditions so the -model learns to remain robust under varying circumstances. To further increase variability, data augmentation techniques such as adding background -noise, pitch shifting or time stretching can be applied. All recordings should be standardized to a consistent sampling frequency such as 48 kHz and -preprocessed to remove noise and extract meaningful features. It is also essential to maintain a strict separation between Train, Validation, -and Test sets, with the Test set containing unseen data that reflects diverse scenarios. Negative data, such as recordings of similar but non‑glass -sounds, should be included to reduce false alarms and strengthen reliability. After training, the model must be -evaluated in real‑world conditions to identify weaknesses. If performance drops in specific environments, adding representative data from those -contexts is often the most effective way to improve robustness. This iterative cycle of data expansion, parameter tuning, and scenario testing -ensures the final model is production‑ready, delivering accurate detection and seamless integration into security systems. +# Glass Break Detection + +This project is designed to work exclusively with DEEPCRAFT™ Studio. Download it from [here](https://softwaretools.infineon.com/assets/com.ifx.tb.tool.deepcraftstudio) + +## Overview + +This machine learning project focuses on developing a glass break detection model that can accurately identify glass break events by analyzing audio signals. +Such models play a critical role in modern security infrastructures, where they enhance situational awareness and enable rapid response. + +Glass break detection can be deployed across diverse environments: + +- Home security systems: safeguarding families and property against intrusions. + +- Commercial and retail buildings: protecting assets and ensuring customer safety. + +- Vehicles: detecting accidents or break-ins to trigger immediate alerts and safety measures. + +By combining audio signal processing with machine learning the model provides accurate event classification, minimizes false alarms, +and supports intelligent safety and security management across multiple domains. This model enhances the reliability of traditional alarm systems while also serving as a foundation for advanced IoT‑based security solutions that intelligently adapt to different environments and conditions. +## Contents + +`Data` - Folder containing the glass break audio wav files used in this project. + +`Models` - Folder where trained models, their predictions and generated Edge code are saved. + +## Collection of Data +The data for this project was primarily collected from the Freesound site, with careful verification of licensing to ensure proper eligibility for use in model training. Infineon also contributed the data providing samples that strengthen +the reliability and diversity of the training set. All audio recordings in the dataset have a sampling frequency of 48 kHz, ensuring consistency and high-resolution quality for signal analysis and model training. + + +## Adding More Data +The project currently includes 369 audio WAV files, forming the foundation for model training. While this provides a solid starting point, additional recordings are needed to broaden coverage. +In particular, expanding the dataset with more diverse glass break samples as well as recordings captured under noisier conditions and varying distances will help improve the model’s robustness and generalization to real-world environments. +DEEPCRAFT™ Studio provides a streamlined environment for recording glass break audio samples and offers integrated tools for labeling and preprocessing, ensuring that data preparation is both efficient and consistent. While studio recordings offer clean baseline samples, they should be complemented with field recordings +or augmented with synthetic noise to ensure the model performs reliably in real-world scenarios. + +## Steps to Production +The recommended path to production for a glass break detection model begins with clearly defining the environments and glass types you want the +system to handle such as residential windows, retail storefronts, or automotive glass. Once the scope is set, data collection should be performed +across diverse scenarios ensuring coverage of different glass materials, break mechanisms, microphone placements, and acoustic conditions so the +model learns to remain robust under varying circumstances. To further increase variability, data augmentation techniques such as adding background +noise, pitch shifting or time stretching can be applied. All recordings should be standardized to a consistent sampling frequency such as 48 kHz and +preprocessed to remove noise and extract meaningful features. It is also essential to maintain a strict separation between Train, Validation, +and Test sets, with the Test set containing unseen data that reflects diverse scenarios. Negative data, such as recordings of similar but non‑glass +sounds, should be included to reduce false alarms and strengthen reliability. After training, the model must be +evaluated in real‑world conditions to identify weaknesses. If performance drops in specific environments, adding representative data from those +contexts is often the most effective way to improve robustness. This iterative cycle of data expansion, parameter tuning, and scenario testing +ensures the final model is production‑ready, delivering accurate detection and seamless integration into security systems. diff --git a/GroundObstacleDetectionTOF/README.md b/GroundObstacleDetectionTOF/README.md index 5ee0dfa6ba..eed03fb204 100644 --- a/GroundObstacleDetectionTOF/README.md +++ b/GroundObstacleDetectionTOF/README.md @@ -33,11 +33,11 @@ Connect the TOF camera by USB(3.1 up) and also make sure you get the royale SDK ## Steps to get started: Model Labeling and Training -### Label the Data and Train a Model in Deepcraft Studio +### Label the Data and Train a Model in DEEPCRAFT™ Studio Detailed in this file [YOLO_README.md](Resources/YOLO_README.md) -### Label the Data in Roboflow and Train a Model in Deepcraft Studio +### Label the Data in Roboflow and Train a Model in DEEPCRAFT™ Studio * Upload the folder of image to Roboflow [https://roboflow.com/](https://roboflow.com/) * Create a project and do annotation @@ -99,13 +99,7 @@ To bring this project and its trained models to production, follow these main st ![Demo GIF](Resources/tof_gif.gif) -## Getting Started - -Please visit [developer.imagimob.com](https://developer.imagimob.com), where you can read about Imagimob Studio and go through step-by-step tutorials to get you quickly started. - -## Help & Support -If you need support or if you want to know how to deploy the model on to the device, please submit a ticket on the Infineon [community forum ](https://community.infineon.com/t5/Imagimob/bd-p/Imagimob/page/1) Imagimob Studio page. ## Getting Started Please visit [developer.imagimob.com](https://developer.imagimob.com), where you can read about Imagimob Studio and go through step-by-step tutorials to get you quickly started. diff --git a/GunshotDetection/README.md b/GunshotDetection/README.md index c16dec5b12..7f71038da8 100644 --- a/GunshotDetection/README.md +++ b/GunshotDetection/README.md @@ -4,7 +4,7 @@ This project is designed to work exclusively with DEEPCRAFT™ Studio. Download ## Overview -This is a starter model for detecting gunshots in a noisy environment, quite far along in its development. The model includes strong invariance to many different background noises with around 1 hour of microphone data. +This is an Accelerator project for detecting gunshots in a noisy environment, quite far along in its development. The model includes strong invariance to many different background noises with around 1 hour of microphone data. It is presented here with the purpose of being supplemented with a significant amount of gunshot data, without having to worry as much about recording and adding background noise data. ## Collection of Data @@ -12,12 +12,12 @@ The data was collected by downloading Creative Commons licensed audio files from ## Adding More Data In order to add more data, you need to upload 48000 Hz audio files with appropriate labels. This could be done either by finding more data online or recording audio using a microphone, such as the AI Evaluation Kit (https://www.infineon.com/cms/en/product/evaluation-boards/cy8ckit-062s2-ai/). -Regardless of how the data is obtained, it can be labelled in DEEPCRAFT Studio. +Regardless of how the data is obtained, it can be labelled in DEEPCRAFT™ Studio. The primary goal should be to add more gunshot audio files, since the existing dataset is lacking in that regard. The practical and legal concerns for collecting gunshot data are unfortunately unavoidable and outside of our expertise to advise on; however, they are likely necessary to produce a production-ready model. ## Steps to Production -The recommended path to production begins with identifying the use case and tailoring this starter model to that need. +The recommended path to production begins with identifying the use case and tailoring this Accelerator project to that need. This involves identifying the intended environment since a model trained on indoor ballistics will have a different performance than one trained in a field or in the woods. The most important step is adding relevant gunshot data to the model, after which the details of how the model can be tested need to be tackled. Due to the nature of this model, there is a large gap between creating a proof of concept and implementing an actual production-ready model. When testing the model, it is possible that gunshot sounds played through a speaker will not be detected, but real gunshots will. This is desired for most applications but an annoyance in testing and demoing of the model. diff --git a/HomeSoundsDetection/README.md b/HomeSoundsDetection/README.md index be482e522a..abcab5326c 100644 --- a/HomeSoundsDetection/README.md +++ b/HomeSoundsDetection/README.md @@ -3,16 +3,16 @@ This project is designed to work exclusively with DEEPCRAFT™ Studio. Download it from [here](https://softwaretools.infineon.com/assets/com.ifx.tb.tool.deepcraftstudio) ## Overview -This starter model is capable of detecting a number of audio signatures common to a home setting. +This Accelerator project is capable of detecting a number of audio signatures common to a home setting. It currently has 3 labels: 'cough', 'baby cry', and 'water tap', but it can easily be modified to add more. -The starter model contains 550 minutes of data, most of it being unlabelled background noise. +The Accelerator project contains 550 minutes of data, most of it being unlabelled background noise. ## Collection of Data The data was collected by downloading Creative Commons licensed audio files from freesounds.org. ## Adding More Data In order to add more data, you need to upload 16000 Hz audio files with appropriate labels. This could be done either by finding more data online or recording audio using a microphone, such as the AI Evaluation Kit (https://www.infineon.com/cms/en/product/evaluation-boards/cy8ckit-062s2-ai/). -Regardless of how the data is obtained, it can be labelled in DEEPCRAFT Studio. +Regardless of how the data is obtained, it can be labelled in DEEPCRAFT™ Studio. ## Steps to Production The first step toward production is identifying which home sounds you want your model to detect. If you wish to focus on human sounds, for example, 'water tap' might be removed. Another thing that is strongly recommended is to use the augmentation functionality to improve model performance. This model originally had augmented data but it was removed to make the download smaller, you can perform this through the data tab of the project file (.improj) diff --git a/HumanActivity/README.md b/HumanActivity/README.md index d6a93195a5..b2316d631b 100644 --- a/HumanActivity/README.md +++ b/HumanActivity/README.md @@ -4,7 +4,7 @@ This project is designed to work exclusively with DEEPCRAFT™ Studio. Download ## Overview -This starter model allows you to build a human activity detector that can be used on any supported Infineon MCU (or other MCUs) with a BMI160 IMU or another IMU. You can use this project as a starting point to develop a production ready model intended for deployment in wristed wearables. +This Accelerator project allows you to build a human activity detector that can be used on any supported Infineon MCU (or other MCUs) with a BMI160 IMU or another IMU. You can use this project as a starting point to develop a production ready model intended for deployment in wristed wearables. Below you can find code examples about how to deploy the output of this project to any supported Infineon MCU with a BMI160 IMU. diff --git a/KeywordDetector/README.md b/KeywordDetector/README.md index 1d8ac9c7c5..f272d33de1 100644 --- a/KeywordDetector/README.md +++ b/KeywordDetector/README.md @@ -12,7 +12,7 @@ The original dataset can be downloaded from [here](http://download.tensorflow.or The dataset differs from the original dataset in that the single wave files are concatenated to longer time series by stitching them together with 0.1 seconds of silence in between to prevent that multiple words are in one input time window of the model. For the transformation, you need to modify the paths in the Python script (locatied at Tools/prepare_dataset.py). -You can add more data by recording with your preferred recorder app of the OS, Deepcraft Studio's [Graph UX](https://developer.imagimob.com/data-preparation/data-collection/collect-data-using-graph-ux) with either your Computer mic or an MCU with [Imagimob Streaming protocol](https://github.com/Infineon/mtb-example-imagimob-streaming-protocol). +You can add more data by recording with your preferred recorder app of the OS, DEEPCRAFT™ Studio's [Graph UX](https://developer.imagimob.com/data-preparation/data-collection/collect-data-using-graph-ux) with either your Computer mic or an MCU with [Imagimob Streaming protocol](https://github.com/Infineon/mtb-example-imagimob-streaming-protocol). ## Taking this model to production diff --git a/MotorImbalanceDetection/README.md b/MotorImbalanceDetection/README.md index 9aaa2c13cb..4260266894 100644 --- a/MotorImbalanceDetection/README.md +++ b/MotorImbalanceDetection/README.md @@ -67,7 +67,7 @@ To prepare the PSOC C3 Motor Control Kit (KIT_PSC3M5_MC1) for data collection: ![Connection Diagram 1](./Resources/image1.png) -To start streaming data to Deepcraft studio first, ensure that the driver for the USB to UART converter (CP210x) device is installed on the +To start streaming data to DEEPCRAFT™ Studio first, ensure that the driver for the USB to UART converter (CP210x) device is installed on the computer which will be used for data collection in Studio. The USB to UART converter supports 15kHz data streaming to Studio. The converter used in the project is available at: https://www.reichelt.com/se/en/shop/product/developer_boards_-_usb_type-a_to_uart_cp2102-266051?country=se&CCTYPE=private&LANGUAGE=en. @@ -88,7 +88,7 @@ https://developer.imagimob.com/getting-started/infineon-ai-evaluation-kit#stream #### Option 1: Predefined GraphUX Project -Launch Deepcraft Studio. Open the Main.imunit file by double-clicking it. Click on the Serial Capture block and adjust the COM port +Launch DEEPCRAFT™ Studio. Open the Main.imunit file by double-clicking it. Click on the Serial Capture block and adjust the COM port settings. Select the correct COM port using the configuration shown in the figure below. ![Connection Diagram 4](./Resources/image4.png) @@ -98,7 +98,7 @@ If the sensor and sample rate are not detected: + Try changing the stop bit setting. + Right-click anywhere in the workspace and click Start to begin data capture. -#### Option 2: Create the project from scratch by following the guidelines provided in the Deepcraft Studio documentation. +#### Option 2: Create the project from scratch by following the guidelines provided in the DEEPCRAFT™ Studio documentation. The figure below shows the recording window, where the blue waveform represents current and the green waveform indicates motor speed. The data is labeled as 'open loop' and 'balanced'. diff --git a/MovementTypeDetection/README.md b/MovementTypeDetection/README.md index 2b86c2738b..071ddbbdd8 100644 --- a/MovementTypeDetection/README.md +++ b/MovementTypeDetection/README.md @@ -4,16 +4,16 @@ This project is designed to work exclusively with DEEPCRAFT™ Studio. Download ## Overview -This is a simple starter model capable of differentiating between 3 different movement types: circle, shaking and stationary based on the IMU (6-axis accelerometer and gyroscope) of the AI Evaluation Kit. Note that in this project, stationary is unlabelled. This project serves as a code example but can also be adapted and expanded if you have an interesting application. As is, the model performs well when differentiating between the clear movement types, but has some errors in edge cases where shaking and circle motions are similar (and indeed with no clear ground truth). +This is a simple Accelerator project capable of differentiating between 3 different movement types: circle, shaking and stationary based on the IMU (6-axis accelerometer and gyroscope) of the AI Evaluation Kit. Note that in this project, stationary is unlabelled. This project serves as a code example but can also be adapted and expanded if you have an interesting application. As is, the model performs well when differentiating between the clear movement types, but has some errors in edge cases where shaking and circle motions are similar (and indeed with no clear ground truth). ## Collection of Data -The data was collected from the built-in IMU on the AI-Eval kit (CY8CKIT-062S2-AI) held in hand and streamed into DEEPCRAFT Studio at 50Hz. The data collection was performed separately by two people. Considerations were taken to vary the frequencies of the shaking and circling motions, as well as incorporating a large number of transitions between the 3 states. It was ensured that the model had plenty of transition data by 5 second alternating: stationary, shaking, circle, shaking, stationary, shaking, stationary, circle. This ensures that all combinations of transitions are covered. Furthermore, for the circles the radius and direction of travel were also varied. +The data was collected from the built-in IMU on the AI-Eval kit (CY8CKIT-062S2-AI) held in hand and streamed into DEEPCRAFT™ Studio at 50Hz. The data collection was performed separately by two people. Considerations were taken to vary the frequencies of the shaking and circling motions, as well as incorporating a large number of transitions between the 3 states. It was ensured that the model had plenty of transition data by 5 second alternating: stationary, shaking, circle, shaking, stationary, shaking, stationary, circle. This ensures that all combinations of transitions are covered. Furthermore, for the circles the radius and direction of travel were also varied. ## Adding More Data Adding more data for the existing gesutres or adding another gesture is simple with the [AI-eval kit](https://www.infineon.com/cms/en/product/evaluation-boards/cy8ckit-062s2-ai/). First, you need to flash and configure the [Imagimob Streaming Protocol Firmware](https://github.com/Infineon/mtb-example-imagimob-streaming-protocol/blob/master/README.md) on your AI Kit. -Connecting this to DEEPCRAFT Studio allows you to stream the data directly, and then labelling it. See below for the setup in GraphUX. +Connecting this to DEEPCRAFT™ Studio allows you to stream the data directly, and then labelling it. See below for the setup in GraphUX. ![](Resources/imgs/GraphUX.png) @@ -28,7 +28,7 @@ It is recommended to have a minimum of 100 seconds of data per label, preferably More detailed instructions on collecting data can be found [here](https://developer.imagimob.com/data-preparation/data-collection). ## Steps to Production -The recommended path to production for this starter model is to identify what motions you want to identify. If your application requires a class for no distinct motion, go through the existing data and create labels for 'Stationary' to replace the existing unlabelled data. This choice should be given some thought. Collect data for your application as outlined above, in a situation that is as representative as your final use case as possible. Next you should think about what motions you want to ignore, and incorporate negative data for those. +The recommended path to production for this Accelerator project is to identify what motions you want to identify. If your application requires a class for no distinct motion, go through the existing data and create labels for 'Stationary' to replace the existing unlabelled data. This choice should be given some thought. Collect data for your application as outlined above, in a situation that is as representative as your final use case as possible. Next you should think about what motions you want to ignore, and incorporate negative data for those. When all the data collection is done, you might want to add preprocessing steps - currently the model only has a sliding window. If your motions have a set expected frequency then this could be leveraged. For the sliding window size, you should have considerations for the inference time requirements of your model, a longer window results in a longer inference time. After evaluating the model, you might realize that your model performs poorly in certain situations; there are no set solutions for this, but adding representative data could help. ## Getting Started diff --git a/README.md b/README.md index 436c09b50a..b19c615652 100644 --- a/README.md +++ b/README.md @@ -1,141 +1,141 @@ -# 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](https://softwaretools.infineon.com/assets/com.ifx.tb.tool.deepcraftstudio) 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](https://www.imagimob.com/studio) and should be accessed through that platform. See also Studio's [online documentation](https://developer.imagimob.com/) 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](https://developer.imagimob.com/deepcraft-studio/data-preparation/data-collection/bring-your-data/bring-your-own-data) - - Object Detection — [more info](https://developer.imagimob.com/deepcraft-studio/data-preparation/data-collection/bring-your-data/bring-your-own-data-object-detection) -3. **Data preprocessing** — configure preprocessing in Studio: - - Use available Studio layers — [more info](https://developer.imagimob.com/deepcraft-studio/preprocessing) - - Add your custom preprocessing layers — [more info](https://developer.imagimob.com/deepcraft-studio/deployment/custom-layers-functions) -4. **Neural networks, layers, and functions** — use only supported building blocks: - - Classification and Regression — [more info](https://developer.imagimob.com/deepcraft-studio/deployment/supported-layers) - - Object Detection — [more info](https://developer.imagimob.com/deepcraft-studio/model-training/training-object-detection) - -## 🤝 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`](_PROJECT_TEMPLATE), but you can also bring your own project. Complete your project files, `README.md`, and `metadata.json`. See [Step 1](#step-1--prepare-your-project) and [Step 2](#step-2--prepare-metadatajson) below for details. - ---- - -2. 📤 **Submit your project** — use the [PR tool](https://github.com/Infineon/deepcraft-studio-accelerators-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](#step-3--get-the-pr-tool) and [Step 4](#step-4--run-the-pr-tool-and-submit) 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](https://deepcraft.infineon.com). - -### Submission requirements - -Before opening a pull request, make sure you have the following tools and software: - -- **[DEEPCRAFT™ Studio](https://softwaretools.infineon.com/assets/com.ifx.tb.tool.deepcraftstudio)** — to build and export your accelerator project (`.improj` file and local `Data/` folder) -- **[GitHub account](https://github.com/join)** — required to fork this repository and manage your pull request -- **[PR tool](https://github.com/Infineon/deepcraft-studio-accelerators-pr-tool)** — the latest version from [deepcraft-studio-accelerators-pr-tool](https://github.com/Infineon/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](https://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](#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`](_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`](_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`](_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](https://github.com/Infineon/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](https://github.com/Infineon/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** - -```bash -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)** - -```bash -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: - -```bash -python .\pr_tool.py --repo accelerators --path -``` - -Replace `` with the root path of your studio accelerator project. For more information, review the tool's [README.md](https://github.com/Infineon/deepcraft-studio-accelerators-pr-tool/blob/main/pr_tool/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. +# 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](https://softwaretools.infineon.com/assets/com.ifx.tb.tool.deepcraftstudio) 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](https://www.imagimob.com/studio) and should be accessed through that platform. See also Studio's [online documentation](https://developer.imagimob.com/) 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](https://developer.imagimob.com/deepcraft-studio/data-preparation/data-collection/bring-your-data/bring-your-own-data) + - Object Detection — [more info](https://developer.imagimob.com/deepcraft-studio/data-preparation/data-collection/bring-your-data/bring-your-own-data-object-detection) +3. **Data preprocessing** — configure preprocessing in Studio: + - Use available Studio layers — [more info](https://developer.imagimob.com/deepcraft-studio/preprocessing) + - Add your custom preprocessing layers — [more info](https://developer.imagimob.com/deepcraft-studio/deployment/custom-layers-functions) +4. **Neural networks, layers, and functions** — use only supported building blocks: + - Classification and Regression — [more info](https://developer.imagimob.com/deepcraft-studio/deployment/supported-layers) + - Object Detection — [more info](https://developer.imagimob.com/deepcraft-studio/model-training/training-object-detection) + +## 🤝 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`](_PROJECT_TEMPLATE), but you can also bring your own project. Complete your project files, `README.md`, and `metadata.json`. See [Step 1](#step-1--prepare-your-project) and [Step 2](#step-2--prepare-metadatajson) below for details. + +--- + +2. 📤 **Submit your project** — use the [PR tool](https://github.com/Infineon/deepcraft-studio-accelerators-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](#step-3--get-the-pr-tool) and [Step 4](#step-4--run-the-pr-tool-and-submit) 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](https://deepcraft.infineon.com). + +### Submission requirements + +Before opening a pull request, make sure you have the following tools and software: + +- **[DEEPCRAFT™ Studio](https://softwaretools.infineon.com/assets/com.ifx.tb.tool.deepcraftstudio)** — to build and export your Accelerator project (`.improj` file and local `Data/` folder) +- **[GitHub account](https://github.com/join)** — required to fork this repository and manage your pull request +- **[PR tool](https://github.com/Infineon/deepcraft-studio-accelerators-pr-tool)** — the latest version from [deepcraft-studio-accelerators-pr-tool](https://github.com/Infineon/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](https://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](#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`](_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`](_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`](_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](https://github.com/Infineon/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](https://github.com/Infineon/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** + +```bash +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)** + +```bash +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: + +```bash +python .\pr_tool.py --repo accelerators --path +``` + +Replace `` with the root path of your Studio Accelerator project. For more information, review the tool's [README.md](https://github.com/Infineon/deepcraft-studio-accelerators-pr-tool/blob/main/pr_tool/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. diff --git a/SirenDetection/README.md b/SirenDetection/README.md index a81f7be92a..9be6e6f23c 100644 --- a/SirenDetection/README.md +++ b/SirenDetection/README.md @@ -32,7 +32,7 @@ by adding data recorded with that device. You can also use augmentation to mix d ## Deploying on AURIX TC345 -In our developer pages we have a complete guide, including a code example for deploying this starter project on the Infineon TC375 Lite Kit FreeRTOS + Audio Shield Board. +In our developer pages we have a complete guide, including a code example for deploying this Accelerator project on the Infineon TC375 Lite Kit FreeRTOS + Audio Shield Board. Read more at https://developer.imagimob.com/getting-started/infineon-aurix-and-imagimob-studio. ## Deploying on PSOC and other MCU diff --git a/SurfaceDetectionMicrophone/README.md b/SurfaceDetectionMicrophone/README.md index 4b5b6e30dc..de17f8b238 100644 --- a/SurfaceDetectionMicrophone/README.md +++ b/SurfaceDetectionMicrophone/README.md @@ -1,10 +1,10 @@ -# Surface Detection for Vacuum Cleaners - Starter Model Project +# Surface Detection for Vacuum Cleaners - Studio Accelerator Project This project is designed to work exclusively with DEEPCRAFT™ Studio. Download it from [here](https://softwaretools.infineon.com/assets/com.ifx.tb.tool.deepcraftstudio) ## Use-case description -This starter model offers a framework to develop a **surface detection project** for recognizing different surfaces (Floor, Carpet, or Air) based on sound patterns generated by a vacuum cleaner. The goal is to enable classification of these surfaces by analyzing the vacuum cleaner's audio signals as it operates. +This Accelerator project offers a framework to develop a **surface detection project** for recognizing different surfaces (Floor, Carpet, or Air) based on sound patterns generated by a vacuum cleaner. The goal is to enable classification of these surfaces by analyzing the vacuum cleaner's audio signals as it operates. The project is designed for use with a simple and inexpensive vacuum cleaner, but the methodology can be adapted to other models or machinery with similar sound-based differentiation characteristics. @@ -23,7 +23,7 @@ If you cannot gather sufficient data for all three surface types, or if the audi ### How can this project ease my go-to-production journey? -This starter project provides a streamlined approach to build a surface detection machine learning model. By using this as your starting point, you’ll gain access to: +This Accelerator project provides a streamlined approach to build a surface detection machine learning model. By using this as your starting point, you’ll gain access to: - A pre-configured framework to perform surface classification based on audio signals. - Built-in preprocessing and data windowing settings. @@ -79,14 +79,14 @@ This project uses three labels corresponding to the three surface types: ## Recommended path to production -Bringing this Starter Model to a production level requires a well-planned data collection strategy to capture sufficient variety in the data, enabling ML models to learn the underlying characteristics that distinguish different surfaces. +Bringing this Accelerator project to a production level requires a well-planned data collection strategy to capture sufficient variety in the data, enabling ML models to learn the underlying characteristics that distinguish different surfaces. ### Data Collection Strategy This is a general guideline on how to lead the data collection process. **Environmental Diversity:** Collect data across various environments and rooms with different acoustic properties (e.g., small rooms, large open spaces, rooms with different echo characteristics). Background noise variations are crucial for robust model performance in real-world settings. -Hint: you can leverage DEEPCRAFT Studio's Data Augmentation feature to introduce reverberation, volume variations and background noise in your data. +Hint: you can leverage DEEPCRAFT™ Studio's Data Augmentation feature to introduce reverberation, volume variations and background noise in your data. If you use Data Augmentation, apply it only to training data. **Device Variability:** Include recordings from vacuum cleaners with different levels of wear and tear, as motor sounds change significantly over a device's lifetime. The fullness level of the dust tank also affects the acoustic signature and should be varied during data collection. diff --git a/SurfaceDetectionMicrophone/Tools/DataCollectionGraphUX/README.md b/SurfaceDetectionMicrophone/Tools/DataCollectionGraphUX/README.md index 6d7785c82f..8982455b12 100644 --- a/SurfaceDetectionMicrophone/Tools/DataCollectionGraphUX/README.md +++ b/SurfaceDetectionMicrophone/Tools/DataCollectionGraphUX/README.md @@ -2,7 +2,7 @@ ## Overview -This is an empty starter project containing an empty canvas for you to start building ML graphs in. +This is an empty Studio project containing an empty canvas for you to start building ML graphs in. Get started by opening the Main.imunit file from the Solution Explorer. diff --git a/TermiteDetection/README.md b/TermiteDetection/README.md index 23f38de247..de974f16d9 100644 --- a/TermiteDetection/README.md +++ b/TermiteDetection/README.md @@ -95,13 +95,7 @@ month = { may }, note = { visited on 2025-10-09 }, } -## Getting Started - -Please visit [developer.imagimob.com](https://developer.imagimob.com), where you can read about Imagimob Studio and go through step-by-step tutorials to get you quickly started. -## Help & Support - -If you need support or if you want to know how to deploy the model on to the device, please submit a ticket on the Infineon [community forum ](https://community.infineon.com/t5/Imagimob/bd-p/Imagimob/page/1) Imagimob Studio page. ## Getting Started Please visit [developer.imagimob.com](https://developer.imagimob.com), where you can read about Imagimob Studio and go through step-by-step tutorials to get you quickly started. diff --git a/TouchDetection/README.md b/TouchDetection/README.md index 937de89631..8a0a857f15 100644 --- a/TouchDetection/README.md +++ b/TouchDetection/README.md @@ -4,11 +4,11 @@ This project is designed to work exclusively with DEEPCRAFT™ Studio. Download ## Overview -This starter project allows you to build a touch detection that can be used on any supported Infineon MCU with a capsense. +This Accelerator project allows you to build a touch detection that can be used on any supported Infineon MCU with a capsense. -This starter project gives you the infrastructure to allow you to expand on the project or to mimic it and create your own project based on the available/included data and tools. +This Accelerator project gives you the infrastructure to allow you to expand on the project or to mimic it and create your own project based on the available/included data and tools. -The starter project is intended to be a demonstration of how you could build a model using Imagimob AI for a device with a capsense. +The Accelerator project is intended to be a demonstration of how you could build a model using Imagimob AI for a device with a capsense. Below you can find code examples about how to deploy the output of this project to any supported Infineon MCU with a capsense. diff --git a/VirtualWindingTemperatureSensing/README.md b/VirtualWindingTemperatureSensing/README.md index f273ea881b..8bee40843d 100644 --- a/VirtualWindingTemperatureSensing/README.md +++ b/VirtualWindingTemperatureSensing/README.md @@ -67,7 +67,7 @@ The dataset consists of multiple measurement sessions collected from TLE995x mot **Files:** - 4 measurement sessions (.mat files) -- Multiple training/validation sets ready for DEEPCRAFT Studio +- Multiple training/validation sets ready for DEEPCRAFT™ Studio - Format: CSV files with data.csv (inputs) and label.csv (targets) pairs @@ -125,7 +125,7 @@ To expand the dataset with new measurements: 5. **Dataset Organization** - Processed data is automatically organized in `Data/processed/` subfolders - - Ready for import into DEEPCRAFT Studio + - Ready for import into DEEPCRAFT™ Studio - Training/validation split can be configured in Studio ### Important Measurement Scenarios @@ -150,7 +150,7 @@ To ensure robust model performance, collect data covering: - Record data across **full motor lifecycle** (new motor vs. aged motor) - Add measurements from **different application scenarios** (continuous operation, intermittent use, high-duty cycles) -**DEEPCRAFT Studio Features:** +**DEEPCRAFT™ Studio Features:** - Use Data Augmentation capabilities if applicable to sensor data - Leverage Studio to visualize multiple datasets diff --git a/VirtualWindingTemperatureSensing/Tools/README.md b/VirtualWindingTemperatureSensing/Tools/README.md index 47f77a119b..fbc2bbda0e 100644 --- a/VirtualWindingTemperatureSensing/Tools/README.md +++ b/VirtualWindingTemperatureSensing/Tools/README.md @@ -63,7 +63,7 @@ python Tools\scripts\4_generate_training_set.py | **3** | `3_split_data_label.py` | **Split data/label** | Creates subfolders with data.csv & label.csv | | **4** | `4_generate_training_set.py` | **Training set** | Splits into 100 folders for training | -**Output:** Organized subfolders with `data.csv` (input features) and `label.csv` (target values) ready for DEEPCRAFT Studio. +**Output:** Organized subfolders with `data.csv` (input features) and `label.csv` (target values) ready for DEEPCRAFT™ Studio. --- @@ -119,7 +119,7 @@ python Tools\scripts\4_generate_training_set.py └─────────────────────────────────────────────────────────┘ │ ▼ - Ready for DEEPCRAFT Studio + Ready for DEEPCRAFT™ Studio ``` ---