Technology

Understand the car first.
Then change the background.

Most "background replacements" cut out a silhouette and drop in a picture. We first recognize the make, model and real dimensions — and only then build a scene where the car actually stands correctly.

car 0.98wheel 0.95plate 0.91
DINOv3 Car ID · Skoda Octavia · liftback · 0.89length 4 689 · width 1 829 · height 1 470 · wheelbase 2 686 · clearance 155

Four steps to a finished photo

01

Detection and identification

Boxes: car, wheels, plate. DINOv3 Car ID returns the make, model, body type and five dimensions.

02

Segmentation

Masks for the body, glass, wheels and plate. The body itself is never repainted — only the light around it.

03

Scene and light

The background is built for the known dimensions and angle: floor plane, perspective, contact shadow, reflections in the glass.

04

Plate and assembly

Your plate is fitted into the perspective and light. The result is 2048 px, 4:3, with the source’s fine detail intact.

DINOv3 Car ID · Dimensions

Our own model.
Millions of photos.

A fine-tuned DINOv3 encoder with two heads: make-and-model classification over a broad catalog of makes and models, and regression of five dimensions in millimeters. Three checkpoints vote by median, so an error on a rare model doesn’t skew the size.

Millionsof photos in training
5dimensions in mm: length, width, height, wheelbase, clearance
3checkpoints, median of the regression
1 photoneeded on input
HFmodel card open on Hugging Face
Servers

Runs on our servers.
Or on yours.

This is how our cloud runs — about 600 frames an hour. We set up the same system on your own servers, too, if everything needs to stay on your own network.

~600frames an hour on our servers
2options: our servers or yours