A few months ago, when we first started developing computer vision algorithms to assess real estate condition and quality, we were approached by a single-family value-add developer who wanted to use our real estate image extraction API to find the "worst house on the best block" at scale. He ran our QualityScore API over most of the city of Philadelphia, processing millions of interior and exterior listing photos to identify mismanaged properties or those in disrepair, and actually found a few good deals from the effort.
This process also helped us realize that we needed to add appraisal and site inspection photos to our dataset - listing photos are just too perfect. With most of our training set comprised of photos with ideal staging, lighting and angles, this customer showed that we were unfairly penalizing photos with poor lighting or staging. Fortunately, we were able to add hundreds of thousands of labeled appraisal images to the training set, balancing out our algorithms to detect underlying real estate quality vs image quality.
The use case of identifying single and multifamily value-add deals at scale is one that resonates with us, and we have been working on a way to analyze market rents and photos at the same time to identify deals with below market rents and poor condition. In this post, we talk about what computer vision is and how it can be used in real estate.
Computer vision is a field of artificial intelligence (AI) and computer science that focuses on enabling computers to interpret and make decisions based on visual data, such as images and videos, in a way that is similar to human visual perception. The ultimate goal of computer vision is to teach machines to "see" and interpret the visual world as humans do, and even beyond human capabilities in certain aspects.
Key concepts and applications of computer vision include:
To achieve these tasks, computer vision employs various algorithms and techniques that often involve deep learning, a subset of machine learning, especially convolutional neural networks (CNNs) and Transformer (ViT). These neural networks are trained using vast datasets of labeled images, enabling them to recognize and interpret new, unseen images. Essentially, computer vision seeks to replicate the intricacies of human vision and perception but also aims to surpass human capabilities by extracting insights and details that might be missed by the human eye.
Using computer vision to identify value-add deals in the multifamily real estate sector can be a game-changer. Value-add deals refer to properties that offer the potential for increased returns through various enhancements, such as physical improvements, operational changes, or market position improvements. Computer vision, when applied correctly, can automate and enhance many tasks associated with spotting such opportunities. Here's how:
Clearly there are many applications of computer vision in real estate, but we still think the holy grail is using it to identify value-add deals in any market. Next we'll talk a bit about how HelloData is using computer vision today and how we plan to leverage it in the future.
Of the above approaches, the most important for our algorithms are automating the analysis of property condition, extracting amenities and attributes, and detecting comparable properties with computer vision. Right now, our rent comp detection model is used by dozens of real estate investors, property managers and PropTech companies to deliver highly accurate rent comps in any U.S. market. We have several clients using our QualityScore API directly too, processing listing photos to assess condition and quality for market studies and appraisals.
In the near future, we're going to release a map-based feature that allows you to search for deals that are potentially undervalued based on their rents and the assessment or algorithms perform on their listing photos and street views. We've also discussed a "daily deals email" that looks at every new multifamily for sale listing that hits the market, analyzes its quality, condition and attributes, and determines the degree to which it is undervalued based on the market. Should be a gold mine for identifying value-add deals at scale. Stay tuned!
Whether you are interested in our Market Survey tool, APIs or Bulk Data, feel free to book a time slot with us to learn more about how we can help you.Book a demo