Neural Scratch
3D inspection and capture of a product from images, without relying on a complete CAD model.
What it solves
It turns images, views or scans of a product into a dense, navigable 3D model, to inspect it and reason about its form, structure and physical behaviour.
The technique is neural 3D reconstruction: AI pipelines that infer, reconstruct and structure geometry from multimodal references, with radiance fields and implicit neural geometry.
Fig. 1
Reconstruction of a vehicle from 14 views. The cameras, in red, supply the images; the geometry is refined until complete.
IllustrativeInnovation
- Reconstructs from images, views or scans, without a complete CAD model
- Radiance fields and implicit neural geometry
- Dense, navigable spatial representations
- Geometry usable for reasoning about form, structure and physical behaviour
How to collaborate
- Joint project
- Adaptation to each domain
Fields of application
- Damage inspection and maintenance
- Digitising existing products
- Starting geometry for simulation
- 3D references for design
Interested in this technology?
Write to us at weare@deepdesignsystems.com