Catalogue management refers to the process of managing the content of a brand's catalogue of products. As the name suggests, catalogue management includes creating catalogues, setting up product categories, entering product information, setting pricing, and controlling the catalogue's content.
The catalogue is the foundation for developing IR (Image Recognition) technology because, ultimately, the catalogue dictates what the IR technology should be capable of recognising. As a result, catalogue management in the context
of IR is vital because without an accurate catalogue, building IR technology is a fruitless exercise, as IR tools cannot extract insights related to the product catalogue of interest.
For IR to work effectively, associated imagery must be available for every SKU product to train the IR tech. Traditionally, training IR tools to identify individual SKUs has been a complex and laborious process, given the amount of real imagery needed. Therefore, maintenance and management of catalogues is a real headache. Although there is no foolproof solution to guarantee 100% product catalogue management, there are some considerable advantages when taking a synthetic data approach.
In this article, we’ll explore how Neurolabs ZIA and its use of synthetic data can help make CPG catalogue management a breeze.
Choosing the right data approach to Image recognition in catalogue management
Currently, there are two main approaches to catalogue management, these include:
Approach one: The use of real data in product catalogue management
Traditional IR technologies use real imagery to upload to the catalogue. Imagery can be photogrammetry images (i.e. images from various sides of the product) or just one to two product catalogue images (i.e. front-facing)
Using traditional imagery puts the onus on the CPG brand as they need to supply the imagery, which is sometimes challenging or unfeasible as this approach is slow and expensive.
However, in some cases, image recognition solution providers will send an agent to gather this data on the CPG's behalf. This service, of course, will be priced into your solution package.
Approach two: Scraping catalogue information from online sources
Alternatively, if you or your IR solution provider cannot gather real data to train image recognition learning algorithms, it can gather catalogue information from online sources such as e-commerce websites.
The problem with this approach is that it is often incomplete, as there is no guarantee that solution providers can find the information they need to identify SKUs for full catalogue coverage accurately. As a result, many CPGs could struggle to find the time and resources to 'fill in the gaps' by updating their catalogues with real data. This, in turn, impacts the cost-efficiency of seeking a traditional IR solution for catalogue management.
Issues with maintaining catalogue accuracy with online data collection
Trusting unverified online sources to generate vital catalogue data may not be the wisest move for CPGs looking to push ahead of their competitors. The information available online can be inaccurate for many reasons, such as:
Online data sources can be insufficient in training IR for SKU recognition in diverse retail environments
E-commerce websites may only include front and back images of products on a plain background, which is not enough to train the traditional IR algorithms for production-level performance. For IR technology to work correctly, the tech needs several different angles of an image, and ultimately, the IR tech needs to be capable of recognising SKUs in any environment.
Typically, traditional IR vendors deploying web scraping techniques for catalogue management often use online image recognition technologies like Google Lens to visibly search for SKUs. However, this method is less robust than solutions offered by dedicated image recognition providers.
Google (the world’s largest search engine) naturally provides the underlying technology for Google Lens, which functions well under “ideal” conditions, such as perfect lighting, product placement and when the test image closely resembles the product catalogue imagery. However, retail environments are highly diverse. For example, product arrangements and overall conditions on the shop floor, such as bad lighting, partial occlusions etc., can vary. Therefore, the online images used by Google to train its one-shot algorithms can often fail in real-world SKU detection scenarios.
E-commerce SKU imagery and annotations may contain poor quality information
Additionally, traditional IR solution providers using web research techniques for catalogue management depend on their sources displaying correct online product information, which, of course, is virtually impossible to guarantee if a third party has provided the images.
For example, SKU images may be of poor quality; they may have distracting product image backgrounds, graphics overlaid on the product packaging, poor image resolution etc.
Moreover, traditional IR vendors may also need to use tools like Google Translate to gather data. This may yield inaccurate results when product copy (used for image annotations) is translated into their desired language.
Overall, insufficient information hinders IR vendors' ability to construct IR learning algorithms.
Synthetic image recognition and data optimises CPG catalogue management
Next-generation image recognition solutions like Neurolabs ZIA use synthetic data and synthetic computer vision (SCV) to train image recognition algorithms.
In practice, we make onboarding and catalogue management as simple and streamlined as possible.
All CPGs need to do is provide us with the following:
SKU artwork - At Neurolabs, we can leverage your existing SKU artwork assets (print label, typically a pdf file) to produce the synthetic data required to build your master catalogue. This is our preferred method for data gathering as it saves CPGs the most time and effort.
However, if these artwork resources are unavailable for whatever reason, don’t worry; we’ve got your back.
ZIA Capture App - Our dedicated app empowers anyone with an iPhone to scan and onboard an SKU in less than 30 seconds. From here, in just a few minutes, your products can be represented as 3D models to train our synthetic IR algorithms.
So for a typical use case of 50 - 100 SKUs, you can onboard your entire catalogue in less than two hours and have the synthetic IR training model ready for deployment in less than a day, achieving 96%+ accuracy from day one. This is a game-changer for the industry!
These robust data capture methods enable you to start training the synthetic IR technology to detect products before items hit the shelf, giving you a competitive edge.
In addition, our image recognition datasets include an extensive database of 100,000+ SKUs, making catalogue management a breeze.
Furthermore, our technology integrates into your existing end-to-end Sales Force Automation (SFA) solution via our API. ZIA’s effortless integration enables you to significantly reduce the time it takes to onboard our synthetic data sets and image recognition technology, streamlining your catalogue management processes.
Neurolabs ZIA streamlines product catalogue management
With our next-generation synthetic image recognition technology, you can streamline catalogue onboarding and management, ensure a high degree of accuracy in your product catalogue management practices and move ahead of your competitors.
If you want to learn more about how Neurolabs ZIA works and how it can boost your company's bottom line, download our free eBook now.
Alternatively, if you want to see for yourself how effective our solution is, get in touch with us today to book a free demo.
At Neurolabs, we are revolutionising in-store retail performance with our advanced image recognition technology, ZIA. Our cutting-edge technology enables retailers, field marketing agencies and CPG brands to optimise store execution, enhance the customer experience, and boost revenue as we build the most comprehensive 3D asset library for product recognition in the CPG industry.
Comments