Sometimes it’s the simple ideas that make the best products. It’s also the simple ideas that can be the most challenging to implement. So it was when ecommerce marketplace startup Bonanza set out to build an automated tool for removing the background from user-uploaded images; a tool that, one year of painstaking development later, would become Background Burner.
Images with a white background drive 10 to 15 percent greater sales conversion than those without. It’s a well-known industry reality and the reason that Amazon mandates all images uploaded to its marketplace have white background. Apparently, there’s something about the way that sort of clean presentation appeals to our eye and makes us more attracted to the product being displayed. It’s also faster and easier to scroll through a page of listings.
It’s understandable why Bonanza was motivated to create such a tool. The company sees sees hundreds of millions of images uploaded to its servers yearly, most of which are plain ugly, CEO Bill Harding says. A double digit conversion lift in the difficult ecommerce marketplace space is some serious motivation.
The real question is, why didn’t a tool like this exist already?
“When I went to talk to my old image processing professor, he told me flat out that ‘It can’t be done – there’s simply too much variety in the foreground and background patterns of images to do this effectively,’” Harding says. “If I’d known how difficult it would be when we started, I probably would never have done it. It took us a year to get it right.”
The closest anyone’s come to automating background removal tools are those by Adobe Photoshop and Microsoft Word which can remove solid color backgrounds, according to Harding. Other companies offer a service through which they’ll remove complex backgrounds manually for $1 per image, a process that takes 24 hours and surely doesn’t scale. But no one has made it possible to automatically detect the core item within an image – say a person, a clothing item, or a consumer electronic device – and then delete everything else.
As for the reason that no one else had built this previously, Harding believes that it takes a special set of circumstances to make the time investment worthwhile. First, Bonanza is profitable, which allows the company to take on small side projects and make secondary product bets that aren’t core to its mission. Not a lot of companies have that luxury. Second, the company has an intrinsic motivation in the potential sales conversion lift to encourage it to stick with the development through the many, many roadblocks it encounters. It wasn’t simply about building a product to sell, it was just as much about improving its core business.
Bonanza has no image processing or computer vision domain expertise on its dev team, Harding concedes, so they simply approached the problem from a common sense perspective. “We asked ourselves, what does our brain do when looking at a photo, how does it know what’s the focus of the image and what’s not?” he says. “It’s served us pretty well thus far.”
Background Burner has been live in various iterations on Bonanza’s site since late 2012 and the company has fielded almost daily requests from other developers and companies wanting to use it. One company, for example, wanted to use it to process criminal mugshots to allow for better facial recognition when searching multiple databases.
Today, Bonanza is launching a productized version of Background Burner to the public via an open API. For now, the product is entirely free. But eventually, Harding says, he expects to make the first 100 photos processed per day are free and charge a nominal fee of one to two cents for each image beyond that point – he’s still mulling final pricing.
“We’re hoping to take an Android-like approach, to get the product into people’s hands, see what they do with it, and figure out monetization later,” he says.
Bonanza is already in talks with Amazon, Etsy, and other ecommerce marketplaces about potential partnerships or integrations.
Currently, Backround Burner is about 75 percent accurate on its first pass, according to Harding, up from the 30 percent success rate the company saw with its first ever attempt at the software. The eventual solution involved a combination of more than 100 image manipulation techniques, starting in OpenCV and supplementing with additional algorithms, a Ruby layer on top, and a host of feedback loops, Harding explains in a blog post. Harding believes that with continued development, the company can get that accuracy number above 95 percent.
The company has yet to put a full time developer, let alone development team on the product, but instead has borrowed dev cycles from the core Bonanza team – just five developers among its 15 total employees. Again, the profitability of the core marketplace product afforded the young company this luxury, even if some of its investors weren’t thrilled at the time. But now that Background Burner is a commercial product, the plan is to build out a small, dedicated team.
“If we can get to the point where we’re seeing hundreds of thousands of images processed daily, that would justify us putting some full-time resources toward this,” Harding says. “Building a marketplace is really, really hard and takes dedicated focus. So if this takes off, we need to light a fire under it and dedicate resources to the thing that’s growing the fastest.”
In it blog post announcing Background Burner, Harding writes:
We’ve been using and improving the burner on Bonanza for years now, so we think it’s a battle-tested platform upon which to build an interesting app. We’re hoping that somebody will build these apps (plus others) and grow prosperous:
• A tool that can take a picture of an item, recognize what it is, and post it to Bonanza from mobile (without typing)
• A tool that burns away background from Facebook/Twitter pictures and replaces with something funnier/campier
• A mobile app that lets people burn backgrounds and put the photo subjects into greeting cards
• A mobile app that can use the burner to recognize a product at a store (using its color and shape properties)
• Customizing avatars or graphics in video games
• Make graphic design (brochure layout, Powerpoints, etc.) easier
• Google Glass object identification
Harding isn’t quite ready to ignore Bonanza, however. The company remains profitable and grew revenue 100 percent over the last six months, after several years of 50 percent annual growth. “We’re finally starting to see network effects come into play,” he says.
Image background removal is hardly a sexy topic, Harding admits. But there’s every possibility that it ends up being a more valuable business than Bonanza’s core marketplace. The company is already building out its patent portfolio around the technology and Harding and his investors are considering whether to spin Background Burner out into its own company, or at least a standalone team within Bonanza.
The next step, should things go well with the Background Burner commercial rollout, will be to build a complementary image matching engine, Harding says. “That would be the jelly to the peanut butter of what we’ve built.”
Background Burner is a great example of what’s so powerful about startup-grade innovation. Rather than accept that background removal technology wasn’t possible, a handful of smart engineers simply put their heads down and built it. They didn’t know at the time that this would be a big commercial hit, just that it might help Bonanza squeeze a few margin points out of its existing business. That and it was a cool problem to tackle intellectually.
It remains to be seen what becomes of Background Burner the business, but Harding and his team already view the initiative as a success. Now it’s time to see what the market thinks.