davanstrien's picture
davanstrien HF Staff
Add IIIF Illustration Detector demo
efa9a57
metadata
title: IIIF Illustration Detector
emoji: 🖼️
colorFrom: green
colorTo: gray
sdk: static
pinned: false
short_description: Find illustrated pages in digitized historical books

IIIF Illustration Detector

Automatically identify illustrated pages in digitized historical books using a lightweight AI model that runs entirely in your browser.

Features

  • Client-side inference: Uses transformers.js - no server required
  • Privacy-preserving: Images never leave your browser
  • IIIF v2 & v3 support: Works with manifests from libraries worldwide
  • Fast: ~20ms per page classification
  • Export: Download results as IIIF Annotations

How to Use

  1. Select a sample manifest or paste your own IIIF manifest URL
  2. Click "Load Manifest" to fetch the book
  3. Click "Classify" to detect illustrated pages
  4. Browse results, adjust confidence threshold, export annotations

Model

Uses small-models-for-glam/historical-illustration-detector - a 2.5MB quantized MobileNetV2 fine-tuned for historical document classification.

  • Accuracy: 95% on test set
  • Classes: illustrated / not-illustrated

What counts as "illustrated"?

  • Engravings, woodcuts, lithographs
  • Photographs and artwork
  • Maps, diagrams, charts
  • Scientific illustrations

Not counted: Decorative drop caps, ornamental borders, printer's devices.

Part of small-models-for-glam

This tool is part of small-models-for-glam - efficient, task-specific models for cultural heritage applications.

Feedback

Found an incorrect prediction? Use the "Report Issue" button in the app, or open a discussion.