AI Image Classification & Object Detection
AI Image Classification & Object Detection
This client's website is the leading consumer resource for detailed information on various new home builders, communities, and floor plans. It also provides tools and guides to assist buyers in navigating the home buying process, from selecting the right home to securing financing.
Industry
Market Intelligence
Project
Computer Vision
Challenge
The client integrates millions of home listing records, including product images such as community visuals, home interiors and exteriors, and floor plans.
About 95% of these images lacked labels, and the remaining had inconsistent labels added by disparate content contributors.
Approach
Given the large volume of data (around 2.5 million images), manual processing was impractical.
The client required an automated system with batch processing and on-demand RESTful APIs, capable of generating and updating metadata in near-real-time, with a 93% accuracy target.
Solutions
EX Squared’s AI-based solution generates standardized image labels, unique image metadata, and enables personalized image search for increased UX, engagement, and SEO organic traffic.
To ensure standardized metadata, we analyzed the data input points and defined criteria to differentiate labels and objects. This was crucial for consistency across multiple annotators.
In addition to classifying images with confidence scoring and identifying objects within each image, our solution also added metadata for OCR content, SEO, and a proprietary aesthetic quality score to help promote “better looking” images above others.
Models were trained using the Caffe and TensorFlow deep-learning frameworks; PyTorch is also an option. Our models achieved 97% accuracy within 8 major categories. TensorFlow Serving was used for API deployment.
Approximately 100,000 images (20,000 with objects) were classified, with around 40+ labels. Our quality control process prevented models from publishing unexpected outputs.
Results
EX Squared’s AI image classifier accurately predicts and annotates real estate images with over 96% accuracy. It can detect 300+ objects within an image, and classify each image into 30+ categories and subcategories.
Centralized data improves targeting and follow up, while automated distribution keeps listings consistent on more than 50 partner sites, expanding reach and improving lead quality. The result is faster updates, fewer errors, higher engagement, and a smoother path to conversion.
Talk With Us
Tell us what you're working on. We'll ask questions, then come back with a straight view on what's worth doing, including if the answer is not yet.
Prefer to start with a call?
Book A Call