When you’re a superstar like Taylor Swift, security can be a major concern. Now there are new tools that can help celebrities stay safe: AI combined with image detection. A recent article in The Verge reported that at a May 2018 Taylor Swift concert, her security team employed facial recognition to identify potential stalkers. Images from a facial recognition camera were cross-referenced with a database of hundreds of the pop star’s known stalkers to see if there was a match. When you think of deep learning for computer vision, this type of use case might not be the first thing that comes to mind, but it just goes to show that its possibilities are endless.
Computer vision captures, processes and analyzes real world images and videos to provide meaningful information, and we’re just beginning to harness its real capabilities. According to research firm, Tractica it’s expected to soar from an estimated $1.1 billion in 2016 to $26.2 billion by 2025.
Deep learning is giving a major boost to the capabilities of computer vision and sparking renewed interest in its capabilities. As a form of AI that enables algorithms to learn by example, deep learning uses learning data representations, as opposed to task-specific algorithms to derive deeper and more independent insights than other forms of machine learning.
Consider the following examples: of deep learning for computer vision in action:
- Ensuring safety in autonomous cars. It’s being applied in autonomous driving to navigate roads and make quick decisions in real time, such as identifying an oncoming vehicle or slowing down on icy pavement.
- Designing better ads. Companies, such as Gannett, are turning to deep learning and computer vision to design better online ads, determining which colors, images and fonts work best. The company says that this has boosted click-through rates across different news sites.
- Improving patient outcomes. It can help physicians diagnose diseases, among other applications. For example, a physician or radiologist can use it to review brain scans to determine healthy or not so healthy areas of the brain.
- Improving urban planning. Computer vision and deep learning solutions can detect the number of buses and cars on busy highways and side roads to more effectively manage traffic.
- Gauging emotions. In areas like education or retail, it can be used to determine the emotions of consumers or students and their reactions to the classroom or in-store experience. In education in particular, surveillance cameras can determine if classroom instruction is interesting by how engaged students appear to be.
Sending Computer Vision and Deep Learning to the Moon
When deep learning for computer vision applications are deployed in satellites, the possibilities are extended even further. Satellite imagery gives us an elevated look at massive amounts of images for applications such as:
- Fighting deforestation. Computer vision and deep learning can help identify and count the trees throughout forests and parks, detect whether deforestation is occurring and suggest possible causes.
- Monitoring economic growth. Monitoring urban activity, such as counting cars, electric lights, or construction work can produce indicators of development and economic growth in countries around the world.
- Disaster relief. Following natural disasters like hurricanes and earthquakes, relief organizations can use automatically labeled maps of structures to visit, use computer vision powered drones to conduct inspections and estimate impact of destruction.
How Do These Computer Vision Applications Deployed? It’s Not that Simple.
While there are clear benefits to the use of deep learning-based computer vision solutions the question of the hour is, how do companies get there? The democratization of this type of AI, through pre-packaged apps such as SalesForce.com’s Einstein, or Google AI tout the availability of “AI for everyone,” yet, when everyone has access to the same benefits, no one can really stand apart.
The democratization of AI has made it easy to create plug-and-play models, but it is still hard to create good models. The plethora of code and tutorials make it possible for a basic programmer to pretty easily build a basic model, but there’s a huge gap between a basic AI solution to answer rudimentary questions and one with the deep understanding of a specific business. When it comes to deep learning, skill is less important than experience in knowing which parameters to choose for each dataset and business problem.
While starting with ready-made deep learning solutions might be a good way to get your feet wet, companies serious about leveraging deep learning-enabled computer vision for competitive advantage eventually need to step away from the quick fix and develop custom applications from scratch. Custom solutions provide not only the deep dive analysis that comes from understanding your unique business needs and specific customer behaviors, but it uses these data-and image-driven insights to solve your business challenges and drive your specific growth path. Ready-made solutions simply do not have the robustness and experience required to do the job.
The problem is that true deep learning-enabled computer vision requires a very specific and highly honed expertise that can be hard to find. Data scientists spend years understanding not only the tech involved, but specific industry customer behaviors and challenges. They learn to understand what resonates in certain markets and they know how to transfer their understanding to machines. In addition, even with out-of-the-box solutions, really beneficial AI requires constant care and feeding. Deep learning solutions need to be constantly fed new data, images, video and other content to be accurate and up-to-date. Very few organizations have the resources to accomplish this in-house. So, while it may be okay to dabble with some of the out-of-the-box AI platforms, custom solutions built from scratch are unmatched in providing true business advantage.
Powerful deep learning insights derived from computer vision technologies are enabling a whole new level of awareness, understanding and insights, improving lives, making people safer, cities more efficient and health diagnoses more accurate. These benefits require a deep commitment among the organizations that deploy them and a trust in what they can really accomplish. As Taylor Swift asks in the title of one of her top songs, “Are you Ready for It?”