Image Classifier Application

A lightweight Vue.js web application for running image classifiers on IIIF manifests or local images directly in the browser.

The application loads ONNX image-classification models, applies the corresponding preprocessing configuration, classifies IIIF canvases, and lets users export a filtered IIIF manifest containing the selected images and machine-generated annotations.

Try a demo

Test this app with a demo app on

Features

• Load ONNX models from local artifacts or from Hugging Face.
• Classify all canvases in a IIIF manifest.
• Classify a single local image.
• Adjust the classification threshold.
• Reset the threshold to the model’s recommended value.
• Filter results by positive, negative, pending, error, or manually corrected items.
• Apply optional negative-label heuristics based on canvas labels.
• Correct model's suggestions
• Export the current filtered selection as a new IIIF Presentation API v3 manifest.
• Run entirely in the browser with no backend server.

Limitations

For now this application is configure only for binary classification task.

Configuration

The application is configured through app.config.json.

This file defines:

• the application mode: local or online;
• available model runs;
• labels and aliases;
• default IIIF manifest URL;
• project metadata;
• heuristic keywords;
• UI texts.

Example:

{
  "mode": "online",
  "ui": {
    "title": "Medieval Illumination Detector",
    "defaultManifestUrl": "https://example.org/iiif/manifest.json"
  },
  "labels": {
    "positive": "illuminated",
    "negative": "non_illuminated"
  },
  "online": {
    "runs": [
      {
        "id": "mobilenetv3_large",
        "label": "MobileNet v3 large",
        "baseUrl": "https://huggingface.co/ORG/REPO/resolve/main/mobilenetv3_large"
      }
    ]
  }
}

Each model run should contain:

run/
├── onnx/
│   └── model.onnx
├── preprocess.json
└── inference_config.json

Run locally

From the repository root:

python -m http.server 8000

Then open: http://localhost:8000

Do not open index.html directly with file://, because browsers restrict local file access and CORS behavior.

Citation

If you use this application in your work, please cite it using the metadata provided in CITATION.cff.