Neurolabs:Explained

Portfolio Explained:Neurolabs

Welcome to ‘Portfolio: Explained,’ a series dedicated to exploring the dynamic world of startups within our portfolio. Our goal is to unravel the complexities of these innovative companies, offering easy-to-understand, clear, and concise explanations of their advanced technologies and solutions.

In this edition of “Portfolio: Explained,” we delve into Neurolabs, a trailblazer in synthetic computer vision for retail.

Origin of Neurolabs

The origin of Neurolabs can be traced back to 2009 at the University of Edinburgh, where the founding team of experienced computer scientists and mathematicians met. From there, ten years of collaborating on-and-off across academia and industry, they founded Neurolabs in 2018. It was their shared frustration with the existing limitations in technology application in the real world that sparked the idea of starting a company focused on bridging this gap.

Their aim was to develop groundbreaking services that could revolutionize the retail industry.

Since its inception, Neurolabs has gained a reputation for excellence, thanks to the exceptional feedback received from their clients. This positive feedback has further propelled the company’s growth and success. With a passionate team spread across various locations in Europe, Neurolabs continues to push the boundaries of machine learning and AI to provide innovative solutions that transform the way businesses operate in the retail sector.

How did you come up with the idea?

“Drawing bounding boxes to train Artificial Intelligence? That’s a dead end! You’re limiting AI and undermining human intelligence. Our bet: let AI self-train in virtual worlds. DeepMind has done it with chess; Neurolabs is doing it in computer vision” — Paul Pop

The 3 founders were living together in London back in 2018. All of them were training Neural Networks for a living and they all saw the same pattern — data is gathered, curated and then manually annotated with intense human labor. To them, this manual annotation process seemed mind boggling and it was clear that in the future this was not going to be the industry standard. They tried training Neural Networks in Virtual Environments, first to follow football players in Premier League games. As the method showed promise, they decided to try their luck on the entrepreneurship path and approach Computer Vision algorithm training fundamentally different — with synthetic data.

The Problem

Neurolabs is addressing significant challenges in the retail sector, particularly in inventory management, product recognition, retail execution and efficient shelf auditing. These areas are crucial for retailers as they directly impact sales, customer satisfaction, and operational efficiency.

Traditional methods for handling these tasks are often labor-intensive, prone to errors, and costly. Additionally, the collection of real-world data for product recognition can restrict scalability, raise privacy concerns, degrade overtime and be expensive. Neurolabs’ synthetic computer vision technology offers a solution to these problems by providing a more efficient, accurate, and privacy-compliant way to manage and recognize retail products, in-store promotions and competitor activity.

Neurolabs’ Solution

Employing Synthetic Computer Vision and Zero Image Annotations (ZIA), Neurolabs is transforming retail processes. By training its system with highly accurate 3D product models (digital twins), it ensures precise product recognition. The ZIA technology allows the system to learn and identify new products autonomously, streamlining shelf auditing and inventory management.

Synthetic Computer Vision

Imagine you want to teach a computer to recognize everyday objects, just like you teach a child. Normally, you would show the computer thousands of real photos of these objects from different angles and under various lighting conditions. But collecting and preparing all these real photos can be very time-consuming and expensive. Synthetic Computer Vision changes this approach. Instead of using real photos, it uses computer-generated images. Think of it like using highly detailed and realistic video game graphics. These computer-generated images are of 3D models of objects, which look and behave just like real objects. Because these images are created by a computer, you can easily make thousands of them with different backgrounds, angles, and lighting conditions. This helps the computer learn to recognize these objects in various real-world situations without the need to collect actual photos.

Zero Image Annotations (ZIA)

Now, let’s talk about teaching the computer what it’s seeing in these images, whether they’re real or synthetic. Usually, this involves a process called “annotation” — like pointing out and labeling objects in a photo. For example, in a picture of a street, you might label cars, trees, and people. This helps the computer understand what each object is. But, just like collecting photos, this labeling process can be very time-consuming and labor-intensive. Zero Image Annotations (ZIA) is a technology that eliminates the need for this manual labeling. It allows the computer to learn and recognize new objects without someone having to go through and label every single image. This is like having a smart assistant that can learn to identify and understand new objects on its own, just by looking at examples. It’s a more efficient way for computers to learn, saving a lot of time and effort.

This method not only enhances product recognition capabilities but also offers a cost-effective, privacy-compliant alternative to traditional data collection methods, thereby addressing key challenges faced by retailers and CPG brands (Consumer Packaged Goods brands). The result is a streamlined, accurate, and efficient system for inventory management, where every product detail is meticulously captured and analyzed, paving the way for a more intelligent and responsive retail ecosystem.

How It Works

Neurolabs’ technology is akin to training a sharp-eyed, ever-vigilant assistant who never forgets a face — or in this case, a product. By using digital twins, Neurolabs trains computer vision models to identify real-world products with remarkable accuracy. This process sidesteps the limitations of traditional data collection methods, which are often costly, time-consuming, error-prone and fraught with privacy concerns. It’s a smart solution, using synthetic data to train algorithms in a way that is both efficient and effective.

Imagine a supermarket where keeping track of products on the shelves and marketing promotions is crucial for both inventory management and customer satisfaction. Neurolabs’ synthetic computer vision technology simplifies this process. When implemented in a store, the system can accurately identify and track products on the shelves and promotional products, ensuring that inventory is correctly managed and any out-of-stock situations are quickly identified and addressed.

Who Benefits?

Neurolabs’ solutions are a boon to retailers, field marketing agencies and consumer packaged goods (CPG) brands. By automating visual-based processes like shelf monitoring, they offer a significant boost in operational efficiency. This technology is particularly beneficial for physical stores, where keeping track of stock, promotions and presentation can be a daunting task. As CEO Paul Pop notes, creating synthetic data is a learning process perfected over years. Their method has proven its efficiency and accuracy, even outperforming real datasets in training computer vision models. This success is due to the precise, controlled nature of synthetic data, allowing for more accurate and bias-free algorithm training.

Who Uses It?

While giants like Tesla and Google pour resources into AI for high-tech products, Neurolabs offers a practical, easily implementable solution for various industries, especially non-tech ones, struggling to adopt automation technologies. Their end-to-end solution simplifies the implementation of computer vision, democratizing this advanced technology for broader use.

The Journey

Neurolabs’ journey is marked by significant milestones. From completing a €1M pre-seed fundraising round to raise €3.5 million in a seed funding round led by LAUNCHub Ventures, with participation from Techstart, 7% Ventures, and Lunar Ventures, the company has demonstrated a clear vision and capability. Their focus has been on scaling operations, expanding into various consumer packaged goods use cases, and amassing the largest 3D product repository in retail. Neurolabs’ success in pitting synthetic datasets against real datasets, and emerging victorious, underscores the efficacy and potential of their technology.

Conclusion

Neurolabs stands as a testament to the power of innovation in addressing real-world problems. Their journey from a university collaboration to a leading startup in synthetic computer vision reflects a deep commitment to revolutionizing the retail sector. With their advanced technology, Neurolabs is not just solving current challenges but also paving the way for future advancements in retail automation and beyond. As they continue to grow and expand their offerings, Neurolabs is set to redefine the landscape of retail technology and open new horizons for businesses worldwide.