We are excited to introduce a powerful new integration with Edge Impulse: Vision Datasets by syntheticAIdata. This tool allows users to generate high-quality, fully labeled synthetic images that simulate defects in manufactured products, perfect for training computer vision models without the need for real-world data collection or manual labeling. The platform offers 3D-rendered datasets of common industrial components and defects that can be imported directly into Edge Impulse, and they plan to release more over the coming months, with a focus on manufacturing.
In testing, I built a model using only synthetic images of plastic bottle caps — and it successfully detected both correct and faulty caps on real-world bottles. This demonstrates how syntheticAIdata’s datasets can help jumpstart defect detection models in real manufacturing scenarios, even before capturing a single image in the factory.
Using only synthetic data, my model was able to detect correct and incorrect real-world bottle caps
This ability to simulate and model defective parts with 3D simulation and output labeled data with in-built support for Edge Impulse ingestion via the Dashboard accessible API key makes it an easy tool to get started with.
Read their launch blog for more details on their datasets, or follow our tutorial to get started today.
Currently Available Datasets:
- Power Sockets and Switches (EU)
Synthetic dataset of EU-standard sockets and switches in diverse installation conditions and angles, designed for smart inspection and validation. - Plastic Bottle Caps
Images of correctly and incorrectly placed plastic bottle caps, ideal for training models in defect detection and quality control. - Laptop Disassembly
A collaborative dataset with Danish Technological Institute (DTI) featuring realistic simulations of laptop disassembly for robotics, remanufacturing, and sustainability-focused automation. - Screws, Nuts, Bolts, and Washers
High-resolution synthetic images of screws, bolts, nuts, and washers arranged in varied configurations to support computer vision tasks in inspection and automation.
Upload to Edge Impulse
With a couple of clicks, you can now load hundreds to thousands of perfectly labeled synthetic images straight into any Edge Impulse project. No cameras, light‑boxes, or labeling sessions required.
In summary
It's impressive to see how well the rendered images appear, being both physically and mechanically accurate, and how quickly users can fully label and resize a dataset and add it to a project. Their tool looks very promising, and I'm excited to see what other datasets get released in their manufacturing suite.
By joining and creating a curated vision dataset with syntheticAIdata, you will gain access to high-quality, labeled images ideal for simulating and detecting defects cost-effectively.
To get started, you can recreate the Bottle Defects project that I created in this blog or on our dedicated tutorial.
If you want to learn more about synthetic data, please see our other blogs and tutorials, and please share your thoughts in the comments below or on our Discord server.