Lightbox has developed a robotic capture system and NERF-based processing pipeline that can digitize an object in 3D with pixel-perfect results.
The digital twin creation process
Product is placed in the robotic scanner. It captures 2000 unique images of the product in 5 minutes or less.
Data from the scan is automatically processed in the cloud using our AI-based algorithm.
That's it! The digital twin has been created. This asset can now be leveraged for a variety of product content needs.
The above process does not involve 3D artists at any point. No manual “touch-up” or editing is needed. Because of this, our process is capable of extremely high throughput compared to other 3D pipelines.
About the technology
Our processing algorithm is based on a new discovery in computer graphics called NERF.
However, NERFs on their own cannot achieve photograph-level fidelity. In other words, if you looked at a photograph of an object and a rendering created by a NERF of that same object, it would be obvious which is the photograph and which is the reconstruction.
We’ve spent the last few years working on a variant of NERF with a focus on extremely high fidelity. Our reconstructions look just like photographs. Take a look at the example below.
Notice that our pipeline was able to handle the complex reflections coming off the geode without any issues.
It’s important to note that we have been able to achieve this level of fidelity while maintaining impressively low training times. For example, the model above took 12 hours to train on a single A100 GPU.
For more test captures, please visit www.lbxlabs.com.