FAQ: Using Neurolabs’ Synthetic Computer Vision Solution in Retail
All your questions about our retail-first Computer Vision software answered.
Q. What does Neurolabs do? Neurolabs provides Synthetic Computer Vision (SCV) software to retailers and Consumer Packaged Goods (CPG) brands to enable them to automate visual-based, in-store processes such as On-Shelf Availability.
Q. What problem does Neurolabs solve? Currently, Neurolabs is focused on retail. Our software helps retailers and CPG brands address any on-shelf issues that they may be having. It enables them to automate manual processes like shelf monitoring and shelf auditing, saving them time and money.
Our Synthetic Computer Vision Approach
Q. What is Synthetic Computer Vision? Synthetic Computer Vision (SCV) is a form of image recognition software. SCV uses Synthetic Data to train Computer Vision (CV) models to detect objects in images and videos. Learn more.
Q. What is Synthetic Data? Synthetic Data is a virtual replica of real data that is used to train Computer Vision models. For Neurolabs, our Synthetic Data takes the form of virtual, 3D models of supermarket products. Learn more.
Q. What imagery is needed to create these 3D models? Neurolabs can create a virtual, 3D model of any Fast Moving Consumer Good (FMCG) product using either the digital packaging of the product or using six images that have been taken of the product (top, bottom, front, back, left side, and right side).
Q. What makes you different to other Computer Vision solutions? Most CV solutions on the market today rely on a process that requires massive amounts of costly and time-consuming, real data as input. This makes it impossible to adapt and scale to the demands of the masses, and unrealistic for most companies to consider its use for domain-specific applications. Using Synthetic Data to train Computer Vision i.e. Synthetic Computer Vision makes the technology considerably more flexible, accessible, and scalable. Check out this post for a comparison of traditional Computer Vision and Synthetic Computer Vision.
Real-Time Shelf Monitoring
Q. How do you monitor the real retail environment once your solution is deployed? We can access images from the real retail environment using fixed cameras that are FTP-enabled. A photograph is taken at pre-defined intervals using these cameras. These images are then sent to the Neurolabs API for detections to be carried out. The detection data is made instantly available to our customers for the purposes of reporting and automation.
We can also access these images via mobile devices or mobile applications. Simply capture an image of the retail environment and send the images to the to the Neurolabs API for detections to be carried out similar to the above. Synthetic Computer Vision allows us to deploy on mobile devices without making a tradeoff in terms of accuracy.
Q. What kind of camera setup is required for fixed shelf monitoring? We recommend a resolution between 2k — 4k but can work with lower resolution models. Once the camera has a clear view of the shelf or fridge that you wish to be monitored, we can collect the imagery that we need.
Q. After I install a camera, who maintains it? It is either the end customer or the System Integrator (SI) working with the customer responsible for the maintenance of the solution.
Q. How do your models detect products that are at the back of the shelf?As long as there’s a glimpse of the SKU in the camera view, the technology will be able to detect it. A good test is the following: if you can tell with your eyes that there’s a product in the image, our technology should pick it up too.
Q. What happens when the camera is obstructed by a person or an object? We have pre-built automated processes in place to continually check for such situations. In the case of a person being detected in the image, we instantly discard the photo and never store it. This is in line with privacy laws. That’s why we set the camera to take photos at very frequent time intervals (e.g. every 30 seconds). For the situation when an object is obstructing the camera, we alert the designated staff representatives to resolve the issue.
Q. What happens when a product changes packaging? The fact that 23% of all Stock-Keeping Unit (SKU) packaging changes quarterly, makes this a very important question. Our solution is designed to adapt to SKU changes with no downtime. This is possible due to our Synthetic Computer Vision approach. We source the SKU’s packaging straight from the brand, allowing us to prepare the computer vision detection algorithms ahead of time in the event of any packaging changes.
Q. How fast can Neurolabs deploy a working solution? We can implement a real-world deployment for you in less than one week.
Q. What level of accuracy can Neurolabs achieve with Synthetic Computer Vision? We can achieve 96% accuracy for SKU-level product recognition from day 1. This increases further as the model carries out detections.
Q. Once the system is installed, how can I use the data? The solution is very flexible in terms of deliverables. For example, the solution can be designed to send product availability insights straight into your Business Information (BI) tool for visualisation. The same insights can be fed into your internal systems (e.g. ERP) to enhance other processes (e.g. demand forecasting). End of day SKU-level reports can be set up to be sent to specific store representatives’ inboxes. Most importantly, live notifications and alerts can be sent to the designated store staff. All data (images and shelf insights) are saved and readily available for visual checks, when needed, through our platform.
Q. What are the types of products you have difficulties recognising? Human sight provides the best proxy to assess the difficulty of product recognition. Compact objects (e.g. canned food, boxes) are easy to detect. On the other hand, transparent, reflective packaging might prove more challenging to deal with depending on environmental conditions such as lighting. However, the fact that Synthetic Computer Vision uses Synthetic Data makes these edge cases a lot less likely to cause an issue. Given the virtual nature of our Synthetic Data, we can simulate conditions that would typically make detection more difficult and train our models to detect the products even under those conditions.
Q. Can you detect differences between the same products but in different sizes (e.g. small vs. large products)? Size is one of the characteristics that the our SKU detection models “learn” about the physical products. As long as your eyes can tell the difference between the products, the model will be able to do so too.
Q. Do you provide a dashboard to visualise the status of the shelf? We integrate your product availability insights into straight into your Business Information (BI) tool of choice for visualisation.
Q. How much time does it take to run the detection? Is it usable in real-time? Inference time is highly dependent on the computational power available. We see the inference time as a variable to be optimised based on the business use-case and the business needs. Real-time inference can be easily achieved with the online solution.
Q. How much assistance do you need from store managers/store employees? Store managers and store employees are the ones that make use of the insights provided by the solution. We designed the solution so that it requires no assistance from their side and so that they can prioritise acting on the insights instead e.g. restocking a product that is Out-Of-Stock.
Q. How can I get support? Is there a specific person? When can I reach you? We usually scale deployments through our technical implementation partners, therefore, they would be the primary point of contact for support. However, while working with us, there’s always a dedicated person available from our side to ensure the solution meets your needs and expectations.
Q. Can I integrate this with my own internal systems? Absolutely. Our solution is designed from the bottom up to integrate with your existing systems. Integration flexibility and speed is one of the core distinguishing features of our solution.
Q. Can I create alerts in your platform? Currently the platform and operational in-store actions are kept separately from each other. That’s in line with our goal to offer high user customisation when it comes to insights-driven actions. However, we integrate seamlessly with services that can trigger alerts and much more such as Robotic Process Automation (RPA) tools like UiPath.
Q. Why do you have a platform? Can’t you just built the product detection model for me? The model is built for you automatically through the platform on the backend. We currently do not expose this step of the process as we want to make the experience as simple and user friendly for you as possible
Q. Why do I have to deal with the 3D models? I already outsource product images. 3D models are critical to the scalability of the solution. Whereas product images are static data, 3D models allows us to reuse the same input data multiple times across different stores. It’s the only way to build a solution that scales across your hundreds or thousands of locations.
Q. Are my 3D models shared with everyone else on the platform? It depends on our agreement. We are very flexible in terms of 3D asset ownership. We fully respect your privacy and, therefore, give you the choice. Of course, the more we can leverage existing 3D assets, the better the quality of the services we can provide and the lower the costs of the solution in the long term.
Q. What kind of support / investments will be needed if we would like to go for PoC? During the PoC, we take care of the majority of the workload. However, we would need your help with the following:
Business Case Definition
Hardware & Network Configuration
Q. What is the price to use Neurolabs? For a Proof-of-Concept (PoC), we usually agree on a one-off price. Once in production, we have a pricing model based on the number of camera deployments (if it’s a fixed-camera solution) or the number of users (if it’s a mobile solution).
Q. Why have you got a monthly fee? In retail, products and packaging change frequently. These changes require regular updating. We think a monthly fee best serves this.
Q. Why can’t I pay for a company licence? To our knowledge, all enterprise software licences have some form of term, so after 3 years or so, a new licence is required. With the Neurolabs model, you get the benefit of being able to scale up or down as you need.
Getting Started Costs
Q. Why do you charge a Getting Started Fee? The simple answer is flexibility. Every client is different. However, all of our clients so far have wanted to implement with a combination of in-house, System Integration, and Neurolabs, with the project costs and hardware often coming from a different budget.
Q. What happens if peoples faces are in the stills Every picture is checked for Personally Identifiable Information (PII) before any insights are extracted from it. In case we detect any humans in the photo, the image is automatically deleted.
Q. Where does the service run? We use all the major cloud providers and place the instance in a jurisdiction that you, the client, find acceptable.
Q. How do you enforce GDPR regulations? Each image captured by a store camera is automatically verified for the presence of human content (customers or employees of the store) once it is stored in the cloud. If people are positively detected, the image is automatically deleted. The solution only processes images with no human content, therefore preventing any PII data being exposed during the process. In addition, our use of synthetically generated data to train the algorithms reduces the need for large image datasets collected in the store, thus decreasing the risk of PII exposure further.
Q. How do you ensure the solution is secure?
We use all the major cloud providers, so we rely on their processes and procedures to maintain and update operating systems.
All data at rest is encrypted.
All user interfaces are secured via username/password. Thus, we can interface with many of the single sign-on models available.
APIs — We use OAuth to grant access to our APIs.
Q. Can we use our cameras? The simple answer is, it depends. We have several models, but ideally, we need a resolution of 2K to take “stills” rather than a continuous feed. The key requirement, of course, is that the camera is pointed at the shelf. Our ideal camera is Power Over Ethernet (PoE) enabled with the ability to take one 4K still per minute as a minimum requirement.
Q. We use a specific cloud services provider. Can we continue to use it with Neurolabs? We use all the major cloud providers, Amazon, Azure, and Google.
Q. Can you deploy the solution offline? We currently support online deployments only. With the increasing transition to cloud services in the grocery space, we are following the trend and prioritising online deployments. However, we do have plans to make our solution available offline in the future.
Q. How do you connect the store to your service The solution needs reliable connectivity to the internet. This is usually achieved by connecting directly to the store router/switch. The bandwidth needed is minimal since we are working with still images as opposed to video stream.
Q. Who owns the 3D Models? If you, the client, provide the complete model and skins, you own it. We would ask for an unlimited licence to use. If we create the model, then Neurolabs owns the model. If you stop using Neurolabs, we would provide access for a fee.
Q. Who owns the Algorithms? Trained Computer Vision algorithms are not transferable.
Q. Who owns the Digital Imagery? If you, the client, own the digital imagery, you continue to own it. You provide Neurolabs with a no-cost licence to use.
Working With Neurolabs
Q. How Can You Install Hundreds of Cameras A typical project has 3 parties: you, the client, ourselves as the solution owner and an integrator. The integrator can be your own in-house team, an integrator you already work with or one we can recommend. Our objective is to make it as straightforward and as low risk as possible. We have successfully done all three options.
Q. Who have you worked with successfully in the past? In terms of end customers, we have worked with Tier 1 supermarkets in Europe such as Auchan and Uvesco Group. As implementation partners, we are working with medium-sized regional Solution Providers (Xabet) as well as worldwide Solution Providers such as ITAB and StrongPoint.
Q. What happens to our model and data if you go out of business You can export your data. The model is not portable. You would need to use your data to train another Computer Vision algorithm.
Q. Why don’t you do an end-to-end solution? Our solution has everything one needs to optimise On-Shelf Availability. All of our clients have looked for the flexibility that our API and Synthetic Data approach provides. Our platform seamlessly integrates with your existing technology, easily connecting to your supply chain and in-store systems. Our output formats are easily consumed by reporting systems such as Tableau and other downstream systems. Our event systems are easily integrated into your help-desk systems.
Q. Who maintains the system once we get up and running? It can be either your internal IT/automation team, your System Integrator, or your Solution Provider partner. Neurolabs is fully responsible for the technological side of the solution, while working alongside other parties to ensure smooth implementation and maintenance of the solution.
Q. What makes Neurolabs better than existing OSA solutions that are on the market? Current shelf monitoring solutions just tell you about the problem, which is not enough. Our solution goes further in getting to the root cause of the problem, allowing you to address the cause not the symptoms (i.e. the Out-Of-Shelf or Out-Of-Stock).
We use Synthetic Data and create digital twins of each of your SKUs to implement (and scale) faster and at a drastically reduced cost compared to using real data. We have access to the world’s largest digital SKU repository and have a process in place to deal with any new additions to the market. This is truly unique and currently the only scalable way of operating an effective Computer Vision solution in a space as dynamic and complex as retail.
Q. How do you compare to just using existing store staff for improving OSA? Our solution is the equivalent of your store staff standing still in front of each shelf in your store 24/7 while reacting instantly to any anomalies spotted on the shelf, first by noticing the problem, then finding the root cause of it, before coming up with the right action to be taken and communicating that action to the right person. That’s without any decrease in monitoring and reporting accuracy due to inevitable fatigue or human error.
Our solution allows you to move away from proxy sources and find a better way to track on-shelf reality. Providing 24/7 shelf truth data, the solution helps you get smarter and more productive at delivering human based interventions to reduce OOS, reduces the complexity of retail, and improves the accuracy of forecasting for increased OSA.
Return On Investment
Q. How can you guarantee a return on our investment? We cannot. However, we have built a robust business case with very conservative assumptions and worst-case costs, and in all our scenarios, the calculations show a positive return. We would very much like to show you our model and get your feedback.
Written by Luke Hallinan, Product Marketing Manager at Neurolabs.
Retailers worldwide lose a mind-blowing $634 Billion annually due to the cost of poor inventory management with 5% of all sales lost due to Out-Of-Stocks alone. 🤯 Neurolabs helps optimise in-store retail execution for supermarkets and CPG brands using a powerful combination of Computer Vision and Synthetic Data, called Synthetic Computer Vision, improving customer experience and increasing revenue. 🤖 🛒