Your Name and Title:
David Palmquist, Systems Administration
Library, School, or Organization Name:
California State University, Fullerton Pollak Library
Co-Presenter Name(s):
Area of the World from Which You Will Present:
Fullerton, CA
Language in Which You Will Present:
English
Target Audience(s):
Library Administration Staff
Short Session Description (one line):
Building a practical book spine recognition system with YOLOv8 detection and Qwen2.5-VL OCR reconciled to Alma data.
Full Session Description (as long as you would like):
This session presents the design and implementation of an automated book spine recognition system for library inventory management.The project builds upon the open-source foundation started by Min-Han Li (https://github.com/MinHanLiWesley/book-spine-recognition), which uses YOLOv8 for spine detection, RealESRGAN for image enhancement, Google Cloud Vision for OCR, and Gemini for metadata refinement.
In our version, we retain the YOLOv8 detection backbone; however, we made several key architectural decisions differently to better suit our environment and goals:
- We replaced the cloud-based OCR and refinement pipeline with a single local Qwen2.5-VL vision-language model that directly performs text recognition on spine crops.
- Instead of public book databases such as Open Library, we reconcile the extracted metadata against our institutional Alma library catalog data without enrichment steps.
Websites / URLs Associated with Your Session:
https://www.dropbox.com/scl/fi/9h4vp7kvshm0mqoz0aswj/stacks_inventory.pdf?rlkey=sh9hnjejkxw5s21w32rijyegl&st=uhdfrrsf&dl=0
Dataset and Public repo will be available by presentation date.
Replies