What does Wizart technology do?

Wizart builds digital showrooms that allows our clients to imagine how a given room is going to look like with specific wall coverings and floorings applied (e.g. helping to visualise how a particular wallpaper is going to look in one’s home after a renovation).

...

This solution can be packaged as a mobile app and/or a web module, which, in turn, can be integrated into a e-store or online catalogue to help website visitors make the right, confident, and quick choice, thus increasing visitors to buyers conversion rates and boosting sales.

How does it work?

Behind the sleek interface is a neural network - an algorithm that is built upon machine learning from a set of data through a process that mimics the way the human brain operates. We have fed it tens of thousands of images of various interiors and help it recognise and separate various objects, such as walls, floors, furniture and other elements of interior.

...

To render the end result of computer vision visualisation as close to photorealistic as we possible can, we use several such algorithms and neural networks that recognise rooms' geometry, shadows, perspective, and scale.

How precise is the algorithm?

The greater the variety of interiors and objects fed to the neural network, the better the precision with which it recognises them. Every day we feed it more and more photographs, specifically focusing on the image types most commonly taken and uploaded online by users. Consequently, our algorithms' precision improves every day and our users are getting increasingly impressive photorealistic results in most cases.

Our stated goal lies in “95% precision for the 90% of the most common interiors”. In other words, 9 out of 10 uploaded photos should produce a result photorealistic enough for a customer to be able to decide whether they would want to buy it or not. However, just like the human eye, the algorithm occasionally makes mistakes and current level of computer vision technology (at Wizart or elsewhere) cannot guarantee 100% precision.

Why is that?

Just like the human vision, computer vision has its limits. For example, if there are objects on the photo never before seen by the neural network - e.g. an anteater, extraterrestrial life’s statue or a medieval stained glass window - it may not separate them from the walls correctly. In some photos, even a human cannot be sure if they see an actual cat in a room or a painting of a cat on the floor. By the same token, by merely looking at a picture, you may not always be able to say how big a painting on the wall is, or distinguish a brick wall from a brick wallpaper on said wall. Computer vision may experience the same difficulties.

In what cases can one expect a subpar result?

• If you photograph an interior atypical for an average customer (e.g. inside the emperor’s room in the Hermitage, a train car, a theatre, a designer interior with unusual solutions or round walls, an exhibition booth with no ceiling etc), you may encounter the problem with Wizart not recognising the walls or recognising them incorrectly.

• If the wall is not flat, e.g. brick, leather or vine (green wall), Wizart may not recognise the wall as suitable for applying finishing materials or confuse wall coverings' texture for shadows.

• If there are unusual objects in the photo (anteaters, abstract art etc), they or their margins may be incorrectly cut out.

• Some pictures may be incorrectly scaled and the applied wallpaper pattern may seem smaller or bigger than in reality (only 10% of results are currently affected, but still).

• If your interior features a complex geometry - many corners, protrusions, alcoves, layered ceiling etc - parts of it may be lost or incorrectly mapped out.

• Sometimes the algorithm, just like a human, is unable to distinguish a flat wooden surface from a similarly coloured wall and offers to wallpaper it.

• Dimly lit rooms and bad angles may sometimes produce an unordinary or controversial result.

What do I do if my interior is “atypical”?

Most cases with atypical interiors may be resolved by making the room brighter or photograph it from a different angle.

...

You can also make use of a “report unsatisfactory result” button and we will include it into our quality control processes and into the next batch of photos for neural network to learn from. We continuously replenish our neural network’s data sets and the results submitted by you will help our neural network to learn how to work with your specific interior. The more users use our product and report subpar results the quicker we can improve the overall quality, especially by focusing on the most common issues end users encounter.

How long will it take to integrate Wizart with my systems?

It usually takes up to 10 days to fully deliver Wizart solution into your systems, depending on complexity of integration and your unique needs. For example, if you only want to add a button launching Wizart showroom on your website and/or have your own white label iOS app using Wizart, it won’t take us long to deliver. However, if you want to make your product information management system to seamlessly work with ours or do the SDK/API integration, it may take a bit longer. All clients are different and have unique needs - drop us a line and we will give you a more precise estimate.