New AI app lets you decorate your home before the first brush is stroked
Posted: 24 April 2017 | By Michael Garwood
AI startup DigitalBridge has created an app, which is designed to help people visualise how their home would look with a new lick of paint or some new furniture before making a decision they might later regret.
The idea for the app came from David Levine and his wife, who wanted to redecorate the dining room in their Manchester (UK) home, but couldn’t picture how any of the designs would look on their walls.
The technology uses GPU-accelerated machine learning and computer vision to let people visualise how wallpaper, a coat of paint, new furniture and other home decor would look in their own rooms.
“I figured I wasn’t the only person who couldn’t imagine what wallpaper would look like in a room,” Levine said.
Save money, time and stress
In an independent survey conducted for DigitalBridge, about a third of shoppers said they’d delayed or cancelled decorating projects because they couldn’t picture how items would look in their own homes. Most said they were afraid of making the wrong choice.
“You can’t undo paint on a wall. You can’t undo carpet. You can’t undo wallpaper,” Levine said. “We’re that undo button.”
You can’t yet use DigitalBridge to try out the chartreuse sofa or purple paint that’s caught your eye. In the next few months, the company will roll out its machine learning home decor tool with U.K. department store John Lewis and other European retailers who are incorporating it into their websites. It’s planning to expand to U.S. retailers by the end of the year.
Machine learning for home decor
To get started with DigitalBridge, do-it-yourself decorators upload a photo of the room they plan to redo. Machine learning and computer vision detect, pixel by pixel, the room’s lighting and where walls, furniture and other items are in relation to each other. That’s so the virtual room matches the lighting and spatial characteristics of real life.
“If you’re trying to select wallpaper, we want to make sure it’s placed behind the furniture rather than on top of the furniture. If you have light shining in the window, we want to replicate that,” Levine said.
Levine trained his machine learning algorithm on GeForce GTX TITAN X GPUs using thousands of pictures of rooms. Then, he deployed it using NVIDIA Tesla K80 GPUs with the CUDA parallel computing platform in the Amazon cloud. Now DigitalBridge is integrating deep learning algorithms to improve accuracy, using the cuDNN library and the Caffe deep learning framework.
“We could not do anything like what we’re doing without GPUs, and we can’t do any deep learning without GPUs,” he concluded.