Synvex · synthetic defect data
Generate the defects you cannot wait for
Synvex renders photorealistic defect images from a part's CAD, material and a defect taxonomy, each one labeled to the pixel. Rare defects get as many training examples as common ones.
Specification
- Inputs
- STEP, USD or OBJ geometry; material specs; defect taxonomy
- Outputs
- Images with pixel masks, boxes, classes, metadata (COCO and KITTI export)
- Rendering
- NVIDIA Omniverse Replicator, path-traced on DGX/HGX or cloud GPUs
- Randomization
- Lighting, pose, camera, background, texture, defect parameters
What Synvex does
- 01
CAD and materials
Import STEP or USD geometry and material specs, and mark the faces inspection cares about.
Core - 02
Procedural defects
Place scratches, cracks, dents, misalignment, porosity and contamination with controlled size and depth.
Core - 03
Domain randomization
Vary lighting, pose, camera, background and texture so models don't learn the render.
Core - 04
Labels for free
Every frame ships with pixel masks, boxes, classes and full provenance metadata.
Core - 05
Dataset versioning
Diff two runs, pin a dataset to a model, and reproduce any run later.
Core - 06
A growing defect library
Defect classes and realism settings improve with every part and industry (planned).
Planned
What Synvex produces
Illustrative. Every Synvex frame ships with its label, so there is nothing to annotate by hand.
Inside Synvex
Scene
CAD, material and camera intrinsics assemble a scene.
Defects
The taxonomy places defects within your tolerances.
Render
Omniverse Replicator renders on DGX or cloud GPUs.
Export
A versioned dataset goes to Coreon for training.
Example defect-class catalogue
Filter by industry. Classes and label types are examples of the planned starter taxonomy; pilot partners will shape what ships first.
Showing 10 of 10 example defect classes
- Typical parts
- Machined metal, glass, plastics
- Labels
- Masks, boxes, length and depth
- Typical parts
- Sheet metal, housings
- Labels
- Masks, boxes, depth map
- Typical parts
- Coated and painted parts
- Labels
- Masks, boxes
- Typical parts
- Castings, ceramics, solder joints
- Labels
- Masks, boxes, branch topology
- Typical parts
- Castings, welds, molded parts
- Labels
- Masks, pore count
- Typical parts
- Machined and stamped edges
- Labels
- Boxes, edge segments
- Typical parts
- Components, labels, caps
- Labels
- Boxes, offset and angle
- Typical parts
- Fasteners, pins, components
- Labels
- Boxes, presence flags
- Typical parts
- Vials, blisters, food trays
- Labels
- Masks, particle size
- Typical parts
- Blister packs, pouches, lids
- Labels
- Masks, seal-width profile
Specification
Items marked planned are in design with pilot partners.
| Item | Detail |
|---|---|
| Inputs | STEP, USD or OBJ geometry; material specs; defect taxonomy |
| Outputs | Images with pixel masks, boxes, classes, metadata (COCO and KITTI export) |
| Rendering | NVIDIA Omniverse Replicator, path-traced on DGX/HGX or cloud GPUs |
| Randomization | Lighting, pose, camera, background, texture, defect parameters |
| Rare-defect augmentation | World and diffusion models for novel defects (planned) |
| Realism scoring | Sim-to-real gap estimate per dataset (planned) |
Synvex in the Viseena loop
- P01 · GenerateSynvex
Renders labeled defect images from CAD, materials and a defect taxonomy.
- P02 · TrainCoreon
Fine-tunes a shared defect model into a per-SKU adapter, closing the sim-to-real gap.
- P03 · InspectVeridex
Runs the model on edge hardware at the station, at line speed, on every part.
- P04 · ExplainPilora
Turns flagged frames into defect reports, triage and probable root cause.
- P05 · CommissionTwinix
Simulates cameras and lighting in a line twin before anything is mounted.
Uncertain and confirmed real frames from Veridex flow back to Synvex and Coreon. Pilora triggers retraining when it sees drift, and Twinix sends new scenarios to Synvex before a changeover.
About Synvex
How realistic are the renders?
Realistic enough to train on, which is the test that matters. Coreon measures each dataset against real frames, and Synvex settings are tuned until that gap closes.
Can we add our own defect classes?
Yes. Describe the defect, its size range and where it occurs, and it becomes a class in your taxonomy.
Who owns the generated data?
You do. Generated datasets for your parts belong to you.
Try Synvex on one SKU
Pilots start with one part, one defect list and accuracy measured on real frames.