Thank you for attending our webinar on Synthetic Data and Machine Vision in partnership with Zebra Technologies.
Below you can access all post-event resources, including the full replay, downloadable decks from Artificial Pixels and ID Engineering, and written answers to the audience Q&A.
If you’d like to discuss implementation, workflows, or project requirements, direct contact information for each speaker is provided at the bottom of the page.
These resources are reserved exclusively for registrants.
Watch the Webinar recording:
Download the presentation decks:
Artificial Pixels
ID Engineering
Your Questions Answered by the Panelists:
- Do you see a possibility to leverage the existing annotations of defects done for the normal workflow of vision systems in DeepRendering?
Yes, absolutely. Existing annotations can be transferred in any case. Depending on how they are available, they may need to be converted. Internally, we use the Coco format, for example. If annotations are available as pixel-by-pixel segmentations, they can also be used directly.
- Business wise – If a CAD model is available and a great number of annotated pictures are available as well, what is the time horizon one should expect before having full realistic artificial images? And how much to have a digital twin capable of realizing images for more than 1 SKU? (eg. if only the color changes, with the digital twin approach we could change simply the DT and have it render images for this second SKU)
The time required to build a digital twin cannot be answered in a general way, it depends entirely on the specifics and complexity of the problem.
However, once the digital twin is completed, DeepRendering needs only 2–3 days to generate fully realistic synthetic data.
Extending the digital twin to additional SKUs (e.g., when only the color changes) is typically straightforward: you would simply update the parameters of the digital twin, and it can then render images for the new SKU without rebuilding the whole pipeline.
- Is a capability assessment in accordance with VDA Volume 5.3 integrated into your software?
No, our software does not include a built-in capability assessment according to VDA Volume 5.3.
However, we fully support a VDA 5.3-compliant validation process. This means:
We provide full data and model traceability during training and operation.
We can document and validate the use of real and synthetic image data in a way that matches VDA 5.3 requirements.
We deliver capability indicators such as F1 score, MCC, false-positive and false-negative rates, and robustness tests.
We offer an audit-ready process description covering training, validation, deployment and monitoring.
In our Understanding VDA 5.3 is mainly a process validation standard, not a requirement for an internal automated module inside the software.
Have more questions? Email us at editor@mvpromedia.com
Get in touch with the panelists:
Brought to you by:
