autoRetouch: Shifting online retailers from manual image editing to automated image processing
About autoRetouch
Based in Germany and serving fashion image producers around the world, from retailers to editing services, autoRetouch is a revolutionary, end-to-end automated image processing platform. With its powerful machine learning models, autoRetouch enables its clients to easily define and apply edits such as background removal, reducing costs and cutting delivery times from days to minutes.
Tell us your challenge. We're here to help.
Contact usautoRetouch enables its clients to enhance their online stores and product images cost-effectively, by training and running ML-powered image processing models on Google Cloud and easily scalable GPUs.
Google Cloud results
- Up to 90% lower image retouching costs for end users through ML model training and processing on high-performance GPUs
- Empowers developers with more time to work on new features by reducing operational tasks to a minimum
- Quickly delivers final images to end users by keeping processing times low whether one image is uploaded or thousands
- Controls costs for the platform by providing easy scaling when computational load isn’t needed and per-second billing
Scales in seconds so images are ready in minutes
In the move to ecommerce, fashion retailers and marketers no longer rely on styling mannequins and arranging shop displays to catch their customer's eye. For customers browsing online, what matters is seeing a selection of high-quality images of the product they're looking at, reassuring them that what they see on the screen accurately reflects what they are purchasing. According to recent studies, that means images that are bright, with a uniform background, and high contrast between the background and foreground.
But for retailers with hundreds of product lines and colorways, getting the perfect image for every product by undertaking edits such as background removal and skin retouch is a time-consuming process. That's where autoRetouch comes in. With its end-to-end image processing platform, it is revolutionizing the editing process, making it more time- and cost-effective. For its users, the benefit is not only a quicker turnaround but a cost-per-image that is up to 10 times lower, enabling them to display a greater number of more consistent images on their sites.
"The product images that you see on a retailer's website are the result of background removal, skin retouching, cropping, and other post-production techniques," explains Alex Ciorapciu, CEO and co-founder of autoRetouch. "When done manually, it takes days to get images ready to upload. The idea of training an AI to handle that kind of unstructured data was unthinkable. But with the power of cloud technology and machine learning, that's what we've done." With autoRetouch, customers can simply set the parameters for their image edits on the web application, preview, and then bulk process multiple images simultaneously. The platform also offers the ability to export Adobe PSD files, where finishing touches can be applied, if needed.
"Our machine learning models are extremely power intensive, and we need the biggest, most powerful GPUs we can get. With Google Cloud, we've been able to access alpha programs such as early GPUs and get the technical support we need. As a partner, it's been there from the start."
—Alex Ciorapciu, CEO and co-founder, autoRetouchTo train and run the models for its custom-developed machine learning algorithms, autoRetouch requires powerful CPU and GPU processors that need to be available on demand. "Because our customers are based all around the world, the platform has to be available around the clock and to scale from processing three images to 100,000 images, in seconds," says Alex.
After building its initial prototype based on a Google open source project, autoRetouch decided to continue its journey with Google Cloud. It wanted a partner that understood its desire to push the boundaries and could provide personalized support for its goals.
"Our machine learning models are extremely power intensive, and we need the biggest, most powerful GPUs we can get," says Alex. "With Google Cloud, we've been able to access alpha programs such as early GPUs and get the technical support we need. As a partner, it's been there from the start."
"A GPU can handle three or four images, depending on their size. Sometimes, we need hundreds of machines, then a minute later, we might need just three. With GPUs on Compute Engine, we have a system that scales as flexibly as our customers need us to be."
—Alex Ciorapciu, CEO and co-founder, autoRetouchPowering up custom machine learning models with scalable GPUs
When building its platform, autoRetouch faced two key challenges. The first was developing custom models to segment images accurately, assigning semantic labels to every pixel so photos can then be modified according to the customer requirements. The second was accessing enough computational power to run those models in a way that was cost-effective, responding flexibly to its clients' needs.
The first challenge was solved by the hard work of the autoRetouch team. "As a starting point for our models, we used an open source Google TensorFlow project on semantic image segmentation called DeepLab," says Alex. "After a year of hard work, and multiple iterations, we've managed to achieve a Mean Intersection over Union (MIoU) score of 99.7% in benchmarking categories. MIoU is used to evaluate semantic segmentation, and our score indicates a very high level of accuracy."
To solve the second challenge, autoRetouch turned to Google Cloud. "We did look at competitors, but we chose Google Cloud because of the exceptional service from the Google team," says Alex. "We can access specialized technical support whenever we need it and have had early access to Cloud GPUs in order to evaluate them. It was a key contributing factor to our choice."
autoRetouch uses AI Platform to host its models and run online predictions using Google Kubernetes Engine, scaling CPU-based models automatically. "We also use Cloud Storage buckets for both our training data and our output data," says Alex. To train its power-intensive machine learning models, the company uses GPUs on Compute Engine.
"Our customers can be literally anywhere in the world, from Southeast Asia to the Americas, or Europe," says Alex. "A GPU can handle three or four images, depending on their size. Sometimes, we need hundreds of machines, then a minute later, we might need just three. With GPUs on Compute Engine, we have a system that scales as flexibly as our customers need us to be."
"Without GPUs on Google Cloud, a platform such as autoRetouch wouldn't be possible, as the hardware would require a significant investment. This way, we can access exactly what we need, our costs are covered by what we earn, and we're able to scale as we grow."
—Alex Ciorapciu, CEO and co-founder, autoRetouchStreamlining operations and keeping an eye on costs
While some of its developers hadn't previously worked with Google Cloud, that hasn't been an issue for autoRetouch. "It's really straightforward to get started," says Alex. "Even developers who are using it for the first time haven't had any problems getting up to speed."
It's also been able to streamline its development and operations (DevOps), running a continuous deployment CI/CD pipeline so it can roll out fixes and new features quickly. "That helps us be super quick in our development," says Till Lorentzen, Chief Software Architect at autoRetouch. "And because every developer is responsible for their own features, we don't need any dedicated operations personnel."
The autoRetouch team also appreciates transparent billing on Google Cloud. "As the CEO of the business, it's really useful to have a clear insight into where our money is going," says Alex. "Is it computing, storage, or the AI hub? From a business perspective, it absolutely makes sense to be able to understand exactly what we're spending on."
Supporting business continuity for clients during COVID-19
Because autoRetouch can easily scale to meet the requests of retailers, they can upload as many images as they need to. Having set the workflow and parameters that suit them, their images are ready in minutes, and at just €0.10 per processed image, the potential cost savings are significant.
"By reducing costs, we enable retailers to upload more images per product, conveying a more accurate impression to the customer and thereby improving conversion rates," explains Julian Eckerle, Chief Revenue Officer and co-founder at autoRetouch. "Retailers can also be more flexible and creative with their images, as it becomes easier to make simple but effective changes across their site. Changing the background color of the images, for example, can make a big difference for certain products.”
The platform’s cost-effectiveness for end users has become more important in the era of COVID-19, as companies all over the world shift their business online amid the challenges brought by social distancing. “It’s a difficult time for everyone,” says Alex, “and we want to do our part to remove some of the burden, whether it’s by helping retailers to do business as usual, or by reducing repetitive tasks across the image editing supply chain to free up time for value-add work that helps the business grow.”
Scaling even further to bring autoRetouch to more customers
With exponential growth in its client base, the plan is to make sure it's ready to serve its new customers. "Without GPUs on Google Cloud, a platform such as autoRetouch wouldn't be possible, as the hardware would require a significant investment," says Alex. "This way, we can access exactly what we need, our costs are covered by what we earn, and we're able to scale as we grow."
While autoRetouch is currently focused on bulk editing for fashion product imagery, it has other markets in its sights, as its product is applicable to any company needing quality images. It also plans to start offering its services through Google Marketplace, so customers can easily access autoRetouch through their Google accounts. "We want to extend our hand to Google customers and let them know they are also valued clients of autoRetouch by making it easy to access our services," says Alex.
Meanwhile, the company continues to work with Google Cloud to deliver more value to its customers by enhancing technical capabilities of its platform. "We're always pushing the limits,” says Alex, ”and our customers expect that. It’s good to have Google Cloud working alongside our team to help us meet, and exceed, customer expectations."
Tell us your challenge. We're here to help.
Contact us
About autoRetouch
Based in Germany and serving fashion image producers around the world, from retailers to editing services, autoRetouch is a revolutionary, end-to-end automated image processing platform. With its powerful machine learning models, autoRetouch enables its clients to easily define and apply edits such as background removal, reducing costs and cutting delivery times from days to minutes.