Manish Garg is the Co-founder and Chief Product Officer at Skan.ai, a computer-vision-based process intelligence platform.
Computer vision conjures up thoughts of autonomous vehicles and facial recognition for many of us. Indeed, billions of dollars have gone into the research and development of such systems. Yet while there is significant promise and progress, the areas are also fraught with technical challenges and moral quandaries.
However, computer vision transcends those high-profile use cases and has practical applications across the enterprise value chain. Corporations can harness the power of computer vision to solve some foundational issues—at a scale and accuracy hitherto unfathomed.
What is computer vision?
Computer vision is a branch of artificial intelligence that deals with how computers “see and understand” digital visual media (images, videos, etc.). While computer vision has been a field of study at universities, computing advances have accelerated the field tremendously in recent years.
In the last couple of decades, visual media and formats have witnessed unprecedented growth—whether it’s YouTube or TikTok videos, Instagram photos or a slew of other social media. With volume, velocity and variety of visual data, no wonder the computer vision applications market is a growing field.
While there are ways computer vision can help companies in their quest for digital transformation, it would also be helpful to distinguish between computer vision and machine vision (even though both are related). Machine vision relies on hardware-enabled components to facilitate industrial engineering and manufacturing use cases. Computer vision is a field of image analytics and understanding, which is agnostic to the methods of inputs and primarily reliant on software to categorize, classify, understand and incorporate various forms and formats of visuals into the business process.
There are hundreds of potential use cases for computer vision, but here are some enterprise use cases to get you started.
• Defect detection and quality conformance. Computer vision-aided models can learn about the appearance of products and flag any deviations and abnormalities for a human operator to take a closer look. In fast-paced manufacturing environments where the acceptable defect rates are extremely small, the ability of computer vision algorithms can help raise the level of quality control.
• Indoor location identification and localization. A repository of existing snapshots can become a point of reference for identifying the location of an individual and potentially directing them to other places. For example, a selfie may help the shopper to the correct site in a large department store, including a snapshot of merchandise. (Or someone lost in a museum can find their way using location identification.)
• Live supply chain tracking. Sensors, satellite imagery and computer vision algorithms can help corporations track the entire life cycle of logistics and transportation, thus enabling more accurate material planning and production scheduling.
• Deepfakes. Deepfakes and fake/misleading news have become a significant challenge. Computer vision can help mitigate this scourge by analyzing images at a pixel level and conducting contextual analysis about the origins and potential distortions.
• Business process monitoring. Knowledge workers predominantly interact with digital systems to complete work tasks where information and workflows from workstation to workstation span various geographies and time zones. Computer vision can observe work as it happens and plot the reference business processes and variations through the law of large numbers. This visual evidence of work can help optimize and transform business processes.
Before embarking on large-scale projects and significant budget allocations, companies should consider the following critical success factors.
• Choosing the problem. Sometimes, companies chase shiny new technologies, and in such cases, it ends as a case of a technology in search of business use cases. It is essential to work on problems with a long-term need and measurable return on investment. Companies should limit science experiments to an incubation center or a center of excellence before implementing the models in production.
• Defining the outcomes. While a computer’s ability to identify a cat or a dog is a monumental achievement, what is the business outcome driving the computer vision use case? Unless your company is a tech giant advancing the field of computer vision, knowing the results and what success means for the business is an essential prerequisite.
• Data availability. Data (in the case of computer vision, visual data) is at the core of both the problem and the solution. In addition to availability, the data quality and the level of categorization and annotations are critical determinants of success.
• Availability of pretrained models. Transfer learning from pretrained commercial or open-source models can cut down research and development time. Unless your company has the resources and need to develop models from the ground up, the path of least resistance is leveraging pretrained models and the resultant transfer learning.
• Capabilities and capacity. There is a shortage of talent in areas of artificial intelligence, and because of this, there is tremendous competition for attracting and retaining high-caliber talent. Before significant computer vision projects, it is paramount to take stock of the talent quotient and then build the capabilities, caliber and capacity to manage computer vision projects successfully.
As the world’s content and knowledge grow more visual in nature, expect computer vision use cases to increase exponentially. This will be true as computer vision technologies, including those at the edge, become more powerful.