Skip to content

Latest commit

 

History

History
92 lines (67 loc) · 3.59 KB

README.md

File metadata and controls

92 lines (67 loc) · 3.59 KB
Raw

VIKUS IIIF Generator

This GitHub Repo contains code to build a VIKUS IIIF generator instance, which can be used to view IIIF Collections (Presentation 3.x API) with VIKUS Viewer. The code crawls all manifests in a IIIF Collection (v3) recursively, downloads images of each canvas, creates spritesheets, extracts metadata, and generates similarity layouts using UMAP and Rasterfairy. The code also includes a REST API and websocket for real-time progress updates, and allows for the download of a ZIP archive for the VIKUS Viewer. The code uses redis cache for manifests and feature vectors and also extract embeddings with CLIP using huggingface transformers.

vikusdocker

Features

  • Crawling of all manifests in a IIIF Collection (v3) recursively
  • Asynchronous queueing for crawling manifests and images
  • Downloads an image of each canvas and save it to disk
  • Create spritesheets
  • Extract metadata from manifests and generate keywords using spacy
  • Extract embeddings with CLIP using huggingface transformers
  • Redis cache for manifests and feature vectors
  • Generate similarity layouts with UMAP and Rasterfairy
  • REST Api
  • Websocket real time progress
  • Download data ZIP archive for VIKUS Viewer

RUN

To start up the Docker instance, run:

docker compose up

If you have a GPU available, you can increase the speed of CLIP features extraction by running:

docker compose -f docker-compose.gpu.yml up

If you need to rebuild the image, run:

docker-compose down
docker-compose up --force-recreate --build -d

CLI

To run the CLI, run:

docker exec -it vikus-docker-vikusdocker-1 python cli.py https://resource.swissartresearch.net/manifest/zbz-collection-100

You can also use the CLI with the following options:

python cli.py https://resource.swissartresearch.net/manifest/zbz-collection-100 collection.worker=1 collection.depth=1

Usage

To use the VIKUS IIIF generator, open the frontend in a browser at http://localhost:8000/frontend/. Paste an IIIF Collection (v3) URL and click "Create". Click "Generate" to run all the steps.

If you want, you can fiddle with the settings of each step and/or run them individually. Once the process is finished, you can open a VIKUS Viewer preview, or download the data ZIP and extract it into the /data folder of the VIKUS Viewer.

Tech

  • Python3
  • Redis
  • Huggingface Transformers
  • CLIP
  • Spacy
  • UMAP
  • Rasterfairy
  • FastAPI
  • Svelte
  • Docker

API

Open the VIKUS IIIF generator API in a browser: http://localhost:3000/docs

Sub-projects

Requirements

  • Docker
  • Docker Compose

Notebooks

To start the notebook server on port 5000 instead of the server, connect to the vikusdocker and run

  • pip install jupyter-lab
  • jupyter-lab --port 5000

Funding

The work on this project was made possible through the Swiss Art Research Infrastructure (SARI) in the framework of the Bilder der Schweiz Online project funded by the Stiftung Familie Fehlmann.