DEEPCRAFT™ Starter Models

The DEEPCRAFT™ Starter Models are designed to kickstart your Edge AI journey. The DEEPCRAFT™ Starter Models are deep learning-based projects that cover various use cases and serve as starting points for building custom applications. The DEEPCRAFT™ Starter Models are open-source and include all the necessary datasets, preprocessing steps, model architectures, and instructions to help you develop production-ready Edge AI models.

You can download the DEEPCRAFT™ Starter Models from the DEEPCRAFT™ Studio and start fine-tuning them to suit your specific needs. The DEEPCRAFT™ Studio offers 3000 minutes of compute time per month, free for development, evaluation, and testing purposes. This provides a valuable opportunity to gain hands-on experience in creating and deploying machine learning models from start to finish.

You can use the Starter Model as:

• a starting point and fine-tune the model as per your requirement
• an inspiration and collect your own data to build a similar project

How do I get started with Starter Models?

1. Open DEEPCRAFT™ Studio.

2. Click New Project in the welcome screen. The welcome screen appears, when you open the Studio for the first time.
   OR
   Go to File and select New Project. The New Project window appears.

3. Navigate to Starter Models> Classification and select the starter model that meets your requirement.

   
   
   

4. In New Project Name, enter the name of the project.

5. In Location, specify the location where you want to create the workspace and the project directory.

6. Select Download project data checkbox to download the project data for that particular project. The project directory is downloaded to the workspace in Studio.

After you download the models, you can start collecting data to fine-tune the model as per your needs.Refer to Data Collection to know how to collect data in DEEPCRAFT™ Studio and Data Preparation to understand the different steps involved after collecting the data in Studio.

Choose your Starter Model

Based on the sensor type utilized in the project, we offer the following Starter Models: IMU and Vibration, Microphone and Capacitive and Inductive Sensing

IMU and Vibration

1. Human activity detection: 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 wrist-worn wearables.

2. Fall detection (Belt-mounted): This starter model allows you to build models to detect a fall using an IMU (accelerometer and gyroscope) mounted on the buckle of a belt. For that, this Starter Model uses data collected from two different IMUs: a Bosh IMU and an STMicroelectronics IMU. Both sensors are set up to collect data at 50 Hz using a +- 8g for the accelerometer scale and +- 500 dps for the gyro scale.

3. Anomalous vibration detection: 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 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.

4. Drill material detection IMU: This starter model is capable of classifying the material a power drill is drilling into based on the IMU (6 axis accelerometer and gyroscope) signature. It is designed to be incorporated into smart power tools. It differentiates between wood, plastic and air.

Microphone

1. Siren Detection: This machine learning project contains everything you need to develop and deploy your very own siren detection model. Bundled with the project is an already trained model, and instructions for how to deploy it to the Infineon AURIX™ TC375 Lite Kit Board and KITA2G Audio Shield Board, which is ideally suitable for automotive and industrial applications.

2. Baby Cry detection: This machine learning project contains everything you need to develop and deploy your very own baby cry detection model. Bundled with the project is an already trained model, and instructions for how to deploy it on the Infineon PSOC™ 6, which is ideally suitable for consumer applications.

3. Drill Material Detection: This starter model is capable of classifying the material a power drill is drilling into based on the audio signature. It is designed to be incorporated into smart power tools. It is developed as a proof of concept and is not fully optimized, achieving around a 85% plastic/wood accuracy. It differentiates between wood, plastic and air.

4. Gunshot Detection: This starter model detects gunshots in a noisy environment. The model includes strong invariance to many different background noises, and has around one hour of microphone data. A limitation with this model is difficulty in testing.

5. Keyword Detection: This machine learning project contains everything to get started with keyword detection. Bundled with the project is a trained model, the Google Speech commands dataset, and the guide for how to download and prepare the dataset. You also get some hints on how to take the model to production.

6. Chainsaw detection: This starter model classifies if there is a chainsaw actively cutting material in the vicinity. Chainsaws that are stalling are defined as not cutting. A fully developed model based on this Starter Model could be used to detect illegal logging or to create automatic warning systems.

7. Home Sounds Detection: This starter model is capable of detecting a number of audio signatures common to a home setting. It currently has three labels; 'cough', 'baby cry', and 'water tap'. This Starter Model can easily be modified to add more labels. This contains 550 minutes of data, most of which is unlabeled background noise.

Capacitive and Inductive Sensing

1. Touch detection (CAPSENSE™): This starter model allows you to build a touch detection model that can be used on any supported Infineon MCU with CAPSENSE™. This starter model gives you the infrastructure you need to expand on the project or to mimic it and create your own project based on the available/included data and tools.