PAVE by Discovery Loft Inc. is making waves in the automotive industry by using computer vision to automate the detailed capture, inspection, and grading of a vehicle, enabling anyone to deliver accurate and comprehensive vehicle inspections within minutes. The process can be accomplished using a smartphone and with no application download required. Since its inception in 2017, the company's mission was to take an industry-focus, collaborative approach to technology by working directly with its intended clients to find ways to accelerate the much-needed digital transformation for the automotive industry.

An Expert Driving the Revolution

Stephen Southin, Co-CEO, PAVE by Discovery Loft Inc. has 25+ years of automotive retail and wholesale experience delivering in-depth domain knowledge, essential in his focus as PAVE's creator and product architect. He also has 15+ years of technical and startup expertise that he gained as an autotech entrepreneur that complemented his role in bringing PAVE to market and building Discovery Loft's dynamic team. Stephen's second recent successful exit was the Bumper App he brought to market in 2012 and successfully exited through Vicimus Inc.'s acquisition in 2017.

Artificial Intelligence Impacting Today's Innovation

PAVE has pioneered an approach that combines human intelligence with artificial intelligence that focuses on automating the vehicle inspection process instead of taking an AI-first approach that was the path of all other startups in the computer vision space. This method allowed Discovery Loft to enter the market with a product that delivers the needed accuracy and consistency with near-real-time results, providing the automotive industry with a transformative solution. The automotive industry has had a sceptical view of computer vision's capacity to inspect a vehicle at the granular level required. By the company's proven results, this has helped open the industry to a better way to do vehicle inspections and opened the door to how Artificial Intelligence can deliver value to the industry in numerous ways.

Continuous Learning from Real-Life Experiences and Peers

Stephen believes in developing one's own trifecta of mastery by putting in the time to gain extensive knowledge and experience to develop three entirely different skills to create a competitive advantage. He mentioned that by combining one's three expertises, one will discover a truly original and compelling idea that can be executed so well that it can generate the groundswell required to attract the right people and traction needed to succeed.  For Stephen, it was his 15 years in the trenches in the automotive industry, 15 years in developing technology, and 15 years as a bootstrap entrepreneur that gave him his three – automotive, technology, and business.  According to him, this method can give one the ability to ask the right questions, to always be adaptive and decisive.

Overcoming the Challenges

Stephen says identifying various classes of damage on vehicles like a cracked windshield, scrapes, scratches, and dents using image recognition is nothing new. Today, a developer can quickly build this skill into their application using many off-the-shelf visual classifiers like IBM's Watson and Amazon SageMaker, along with many open-source repositories available on GitHub.

According to him, 2020 was an exciting time for AI, deep learning and computer vision because virtually anyone could quickly prototype an idea that showed well in their presentation. Stephen was impressed because it had some actual functionality. He says, however, once one digs a little deeper into the use-case (hopefully not after the plan gets sold), one can discover that it is just a prototype. These available models are very limiting for a lot of reasons. For a day, if one followed a professional vehicle inspector at an auction as they do cosmetic inspections on vehicles, it would be shocking at the number of variations of the types of damage and blemishes they have to detect. PAVE needed to extract the elements and objects from digital images for over 10,000 unique vehicle damage types to match their trained skills and years of human experience.

Stephen mentioned that one's most significant limitation would be to identify the exact part of the vehicle that is affected. To automate vehicle inspections that have real value to the industry, one needs to be able to locate the point of interest at the component level, not just "has damage on the front or side of the vehicle." PAVE identifies the complete anatomy of the over 22,000 body types of cars, trucks, and vans sold in North America over the past 15 years, down to knowing the marker light on the side mirrors.

Vital Attributes Every Innovative Leader Should Possess

Stephen says, "Attract the right people for your tribe, always ask the right questions, be adaptive and always be decisive and keep your focus on the things you need to get done and learn to postpone those things you'd like to get done."

Innovative Products to Attract the Target Audience

To innovate products and solutions, Stephen advises: "Get your idea validated as quickly as possible by giving your target market a chance to interact with it so you can benchmark their reaction and, more critically, gain valuable feedback about what you actually should be building. But once you know what you need to create, don't release it too early. Hold off showing anything if possible until it is ready. The world won't be capable of seeing the finished product for what it has become, and they will only remember the half-baked version you pushed out prematurely."

Disruptive Technologies Bringing Digital Transformation

Stephen believes technologies like AI, ML, Big Data, and the Cloud are now an essential part of any tech stack for every organization and every department. They are the real drivers of digital transformation, and the leaders who are fully embracing them are the ones who will gain the needed lead in their industries that will be hard for others to achieve the pace required to catch up.

The Future Ahead

Today is the right time to be going all-in on computer vision. Many recent breakthroughs like Geoffrey Hinton's recent paper on "GLOM" that enables neural networks with fixed architectures to parse an image into a part-whole hierarchy with different structures for each image are fascinating. This type of approach could lead the development to bring solutions to market soon, exceeding a human's own ability of sight. That day would be a big step for the entire AI industry, and it is not that far in the future.