All case studies
Sports Collectibles · Computer Vision · Production engagement
Automated baseball card grading
Designed and deployed a computer vision workflow that automatically analyzes baseball cards and returns centering measurements in real time — helping collectors decide which cards are worth submitting for professional grading.
Eliminated
Manual centering analysis
Instant pre-grading
Collector insights
Reduced
Unnecessary submissions
Scales to more grading factors
Architecture
Challenge
TradingCard360 wanted to help collectors determine which cards were worth submitting for professional grading by automating one of the most important grading factors: centering. Manual centering analysis was slow, inconsistent, and a barrier to scaling pre-grading insights inside their application. They needed a system that could handle the wide variety of baseball card designs — both framed and frameless — and return grading-relevant measurements through a single API call.
Client
TradingCard360
Our approach
- Built a classification model to identify whether a card contains a frame, used as the entry point of the workflow
- Trained two custom instance segmentation models — one optimized for framed cards, one for frameless designs — to handle each layout with maximum accuracy
- Designed an automated workflow with intelligent model routing: the classifier decides which segmentation model runs for each image
- Derived centering coordinates from the segmentation masks and packaged masks + measurements into a single structured response
- Deployed the full pipeline as a production-ready API endpoint integrated directly into the TradingCard360 application
- Established a scalable architecture that can extend to additional grading factors beyond centering