To carry out the analyses, FitYourOutfit needs an image of the person in the profile. Subjects take the pictures themselves and must wear light-tight clothing, underwear, or a leotard. The position of the person’s body is crucial. For example, a stray elbow will interfere with the profile and the result. To boost success, the company created an app that approves or rejects images the end-users took.

Once the photo is taken, the process must strip out the background and convert the image to a single color. The distribution of pixels in the image over seven zones is fed into a patented, algorithm-based pixel sizing and counting method, allowing for body composition analysis.

The advanced technology then calculates the body's size and shape, body composition, fat mass, free fat mass, abdominal mass, cardiovascular risk measurement, and other information. This is then sent to the end customer’s care provider.

Lately, online clothes retailers have discovered the service because it helps increase the accuracy of fittings and reduce the number of returns.

Two years ago, software engineer, CTO, and Partner at Pixelcando, Paolo Cecchetti, created an in-house background removal solution. On the back of this, the company’s customers reached the tens of thousands. However, the cost of hardware investment and system training increased in line with their growth. At some point, they would also have to hire a pricy background removal expert. They needed fixed or, ideally, falling costs per sale.

Additionally (and despite the app) Pixelcando received too many complaints about images from which the background was not properly removed. The AI tended to leave artefacts in the images and would produce two body profiles instead of one if there was a mirror in the picture. Customers only pay for successful outcomes and quality was paramount.

“We don’t want complaints, we want to form a top-grade service!”, says Talluri.

But Cecchetti’s solution was tested, accurate, and reliable.

“It’s not like sales images, because if you have different results in terms of a number of pixels, you have different values in estimates”, says Talluri.

And that can lead to inconsistent results—and unhappy clients.

Another factor. The first image end-users send to care professionals would probably mark the start of weight loss or muscle gain. So, the input from which Pixelcando’s AI works must be consistent each time.

“You have to compare apples with apples,” says Talluri.

As soon as Cecchetti discovered remove.bg, he hoped it might solve all their issues.

“I thought, ‘Why don't we focus on what we are going to do? And why don’t we try to use something from a provider that works better than ours?’” he says. "There is no need to reinvent the wheel."

The first step was to submit to remove.bg the images that had failed in the in-house system. The results were promising.

However, Talluri and Cecchetti needed to ensure that the system met Pixelcando’s requirement for minimal output variability. An intense period of testing followed.

“We want to ensure the results were consistent with our needs. As long as the profile is consistent, we have no trouble,” says Cecchetti.

“If you lose diameter, you are not necessarily losing fat. So we had to protect our reputation, first and foremost. This is more than just how many pictures we are analyzing,” explains Talluri.

remove. bg passed the tests, and Pixelcando now uses it to process thousands of images monthly. There’s a stretch goal of “thousands per minute.”

Cecchetti says the in-house system produced around 10% “reject” results. He estimates that remove.bg has probably cut that by 8%. The issues with mirrors and artifacts have largely been solved, and the company believes they remove.bg is nine times as successful as their solution.

“Another way of putting this is that remove.bg seems to have a 98-99% success rate,” Cecchetti says.

To improve the results, Pixelcando sends those 2% of images to remove.bg for examination and to improve the algorithm.

“You can have a very fast checking loop!” says Talluri.

While the partnership is only a month old, early results are promising: they used to get complaints from customers once every two or three days. Since the change, they haven’t had a single grumble.

remove.bg comes onboard at an exciting time for Pixelcando. FitYourOutfit is producing double the amount of body composition work carried out year-on-year. Outreach to clothes retailers through their third-party contractors is underway.

Instead of rising technology costs, remove.bg delivers results that become more affordable per image as volume increases.

“It’s a win-win for both,” says Cecchetti.

“We are experiencing exponential growth right now,” concludes Talluri. We hope that remove.bg is the right choice for us, not only now but also in the future.”