Bora Bicer, CPO of the Turkey-based No. 1 general marketplace, spoke at AutosBuzz about advanced search functionality consumers can use simply by taking a picture of a car.

How does it work?

The Sahibinden auto app improves the process of searching for a car by allowing consumers to take a picture of almost any part of a car, get it recognised immediately and then search for buying options for that make and model.

“We built a machine-learning model that pre-processes massive amounts of car image data. We developed the model with millions of photos and tested it with a different dataset. We then deploy an AI prediction service and all aspects provide continuous improvement.”

Users can take a photo from any angle, the photos don’t need to be perfect and the app recognizes the type of car from both interior and exterior photos.

Bicer showed an example live in front of the audience using a toy car that was immediately recognized and takes the user to a search function for the exact make and model of car they have showed an interest in.

“Our model can do this just from the tyres, at night, from almost any type of image, even ones hidden by tree foliage or other obstacles.”

“The model has given us 94% accuracy on defining the car, up from 52%,” said Bicer. “For the make, model, year and type it takes just 0.3 seconds and the accuracy has increased from 91% to 98%.”

“So you can take a picture of the car and then search for that car on our platform, all done on the app via smartphone in a matter of seconds.”

“We want to introduce it to other marketplaces.”

Print Friendly, PDF & Email

Related Articles