Automotive

Laan Labs builds computer vision and 3D capture software for automakers, dealers, online marketplaces, insurers and aftermarket companies. A phone recognizes a vehicle, estimates its position and orientation in real time, guides the person holding it to the right shots, and records the car as photographs, a 3D scan or a photorealistic splat. For a German automaker, Laan Labs built an app that guides car sellers to consistent, high-quality photographs using a custom-trained pose estimation network.

Cars are hard subjects for computer vision. They are large, glossy and reflective, they look different in every light and setting, and the people photographing them are sellers, owners and technicians, not photographers. Laan Labs has worked on the problem from both ends: the models that understand a vehicle in a camera frame, and the synthetic training data that makes those models affordable to train, with more than 10,000 labeled images rendered instead of photographed and annotated by hand.

The same capture serves the workshop. A scan of an engine bay, a wheel arch or an interior gives a builder or parts maker real geometry to design against, and a dated 3D record of a vehicle shows what condition it was in at handover, at return or before a repair.

How Laan Labs helps

Guided vehicle photography

A capture flow that recognizes the car, shows the seller where to stand for each standard exterior and interior shot, and checks framing, lighting and focus before the photo is accepted, so listings look consistent whoever took them.

Real-time vehicle pose estimation

Neural networks that estimate the position and orientation of a car from a phone camera, on the device and in real time, as described in real-time 3D car pose estimation trained on synthetic data.

Synthetic training data

Photorealistic renders of vehicles across models, colors, lighting, weather and backgrounds, with pose, outline and keypoint labels computed from the 3D scene, replacing most of the photography and manual annotation a vision model otherwise needs.

AR on and around the vehicle

Graphics, features and accessories anchored to a real car at true scale on a phone, tablet or headset, for product explanation, configuration, owner guidance and service instructions.

Condition and damage records

A walk-around capture at check-in, lease return, rental handover or first notice of loss, with each photograph and finding tied to its place on the vehicle. The same approach serves insurance claims.

Scans for fitment and custom parts

LiDAR and photo scans of engine bays, wheel arches, interiors and body panels, exported as meshes for CAD, so brackets, body kits, wraps and interior parts are designed against the real vehicle and not a drawing of it.

Photorealistic 3D for listings and showrooms

Gaussian splat captures of a vehicle, inside and out, that reproduce paint, glass and chrome better than a conventional 3D scan and open in a browser from the listing page.

Where to start

Most automotive projects start with one step in one workflow, such as listing photographs for a marketplace, return inspections for a fleet, or scans for one family of aftermarket parts. Laan Labs builds a working prototype in weeks and tests it on real vehicles in real conditions, outdoors and in poor light, because that is where vehicle vision systems succeed or fail.

The result can be delivered as a white-label app, as components inside an existing app, or as trained models and the data pipeline behind them. See computer vision and AI development and 3D capture and technology licensing.

Laan Labs products in this field

• 3D Splat App — Photorealistic Gaussian splat captures on Mac
• RadianceView — Scans, splats and photographs as one condition record

Common questions