Was this the DeepSeek Moment for Computer Vision?

After 33 NRFs, few moments stand out to me as transformative milestones. Most were a mishmash of buzzword bingo with every vendor saying pretty much the same thing. NRF 2025, however, may have marked one those transformative moments for computer vision, with the introduction of a groundbreaking mini-edge solution that redefines what is possible.

To fully appreciate the impact of this innovation, it is useful to draw a parallel with Deepseek’s contributions to AI. Deepseek made waves by compressing existing open-source models, drastically reducing their footprint and processing requirements. This enabled enterprises to deploy powerful AI solutions at a fraction of the cost.

The potential game-changer in computer vision that we saw at NRF 2025 combines the NVIDIA Jetson Nano device, priced under $250, with a compact model from UltronAI. This interesting approach echoes the seminal impact of Deepseek in the realm of AI, promising to democratize computer vision with unprecedented efficiency and cost-effectiveness as it can be used in small edge devices rather than heavy compute, potentially lowering the barrier to entry for many retailers. Also, although not specifically open source, the UltronAI solution is democratized and available to vendors, systems integrators and retailers alike.

Computer vision has carved out a critical role in the retail sector, addressing a multitude of needs from theft prevention to enhancing customer experience. Growth in retail theft took this technology from a nice to have to a must-have for retailers. However, its wider adoption has been hampered by often heavy installation costs and significant computational demands that are difficult to do in real-time. Retailers have an urgent need for faster, more efficient computer vision solutions, particularly in areas like the aforementioned theft prevention but also inventory management, traffic counting and beyond.

Current Adoption Rates

According to a recent IHL study, 51% of the retailers planned to use computer vision coming into 2025. This means that more than half of the surveyed retailers are either using or planning to use computer vision, indicating a high level of interest and demand for this technology.

The study also found that the highest sales growth retailers, were 33x more likely to be using computer vision than sales laggards. They were also doubling down for computer vision in 2025, showing 207% greater plans to deploy new computer vision solutions. Similarly, profit leaders were 13 times more likely to be using computer vision and 585% more likely to make a decision in 2025.

As Jerry Sheldon, VP of Technology at IHL Group, observed during NRF 2025, “One of the really cool things that I saw at NRF 2025 was a demo at the Fujitsu booth where they’re using the NVIDIA Jetson running the UltronAI Foundation Model for Product Identification in a self-checkout setup.” HP was also showing a similar product in their booth with Nvidia’s competition Hailo SoC also running UltronAI Foundation Model in real-time for check-out”

In our view, the NVIDIA Jetson platform, when integrated with UltronAI’s compact model, represents a significant leap forward in the use of computer vision that is integrated with other specific devices. The Jetson Nano, with its compact size and impressive processing power, is designed to target the IoT space, enabling rapid image processing with minimal overhead with some models as low as $400. In other words, what the Raspberry Pi did to minimize CPU, Jetson Nano does for GPU intensive tasks.

What makes this solution revolutionary beyond just more computer in smaller scale is its ability to train for specific use cases without the burden of a full-scale computer vision model. This tailored approach with a core set of computer vision focused, with further training specifically on your use case, reduces the computational footprint and lowers costs, making advanced computer vision accessible to a broader range of retailers. Also, although not specifically open source, the UltronAI solution is democratized and available to vendors, systems integrators and retailers alike.

There are some remarkable advantages of this compact computer vision AI engine approach. These benefits include:

• Accuracy: High precision in recognizing and processing images as database is much smaller.
• Speed: Fast inference capabilities, ensuring rapid decision-making.
• Scalability: Architecture that can handle 250,000 SKUs, making it suitable for large retail inventories for a single specific use case.
• Single Image Product Enrollment: Simplifies the process of adding new products to the system.
• Low-Cost Edge Devices: Economically feasible for a wide range of retailers.

This approach not only democratizes access to computer vision to more enterprise retailers but also provides a viable solution for smaller retailers who might have previously found the costs prohibitive.

The solution demonstrated was embedded in self-checkout, but the use cases are multiple within a retail operation for this edge AI approach.  These include inventory accuracy/shelf monitoring, shipping and receiving compliance, traffic monitoring, asset tracking or even theft prevention. One use case is facial recognition of known organized retail crime offenders. IHL data shows that retailers using facial recognition saw 74% less increase in consumer theft in 2023/2024. An edge device for this use case linked to law enforcement could alarm situations before they happen through recognition.

Looking at the future, the integration of computer vision with edge computing and other technologies will further enhance its capabilities. This is not the only game in town or approach to lowering computer vision cost but seems to be one that is driving to that democratization of the technology.  But this approach should make each retailer and vendor re-evaluate how they do things today. 

And our research suggests the combination of compute on the edge and functions like computer vision and RFID are exactly what retailers are looking for.  The IHL study points out that those already using computer vision are 10x more likely to also be using edge computing. This synergy promises to deliver faster, more accurate results while reducing latency and bandwidth issues.

Moreover, advancements in AI, machine learning, and even quantum computing are poised to further improve computer vision. And without question, Jevon’s paradox (increased efficiency can lead to increased consumption, rather than decreased consumption) is alive and well with computer vision. These technologies will enable new algorithms and methods, enhancing the performance and efficiency of computer vision systems.

This moment represents a significant leap forward, one that could reshape the future of retail and beyond.

For more on the entire study mentioned with raw data, see here.

For the complete NRF Recap go here.