simCataloguer

simCataloguer is a workflow for generating machine-authored text and images from catalogue data. It was created in 2023 as part of the Arts and Humanities Research Council (AHRC) funded project 'Legacies of Catalogue Descriptions and Curatorial Voice: Opportunities for Digital Scholarship.'

Setup

This process assumes assumes familiarity with Windows Powershell.

System Requirements

Please note that you may require a large amount of VRAM to use this library. Data we have gathered so far: 8Gb is not enough but 24Gb is sufficient.

Installation

Clone repo

git clone --recursive git@github.com:dreamingspires/simCataloguer.git

Optional steps

You may need to install zlib, and graphics driver

• get NVIDIA GPU Computing Toolkit, CUDA 11.7.0
• get cuDNN.
• install cuDNN.

Installations

• install python 3.9.
• install pip (may not be required, check with pip --version)
• install Poetry.
• add poetry to path using on screen instructions.
• reboot shell.
• check with poetry --version.
• run poetry config virtualenvs.in-project true.
• poetry install (You may need to run this more than once, it fails to parallel process on occasion)
• poetry run pip install --extra-index-url https://download.pytorch.org/whl/cu117 --no-deps --force-reinstall torch torchvision

Run test

• Go into your examples directory within simCataloguer and run poetry run python test_writer.py'

Notes

• You can adjust num_cuts and quality settings in examples/test_writer depending on the GPU size of your local machine. See the Pixray documentation for details.
• When you first run poetry run python test_writer.py a language model will be built based on the input file in examples/test_writer. If this process fails, you may need to download a pre-trained model. For MDG.txt a pre-trained model is available to download here. This should be unzipped and placed at simCataloguer/checkpoint.
• When you first run poetry run python test_writer.py, four large files will be downloaded to simCataloguer/.venv/Lib/site-packages/pixray_module/models. If the downloads fail for any reason, then rerunning poetry run python test_writer.py will fail. If this happens four files - yfcc_2.pth, ViB-32.pt, ViT-B-16.pt, and RN50,pt can be downloaded seperately and manually placed in this directory. As yfcc_2.pth is the largest and most likely to fail, we include a direct download link here.

Reuse

BM-MDG.zip is derived from a dataset published by the British Museum, data and derived data are licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license.

For more info on this dataset see Baker, James, & Salway, Andrew. (2019, June 13). Creation of the BMSatire Descriptions corpus (Version v1.0). Zenodo. doi: 10.5281/zenodo.3245037.

Credits

simCataloguer was developed by Dreaming Spires Software Development Ltd in 2023.

For more details of the type of projects we develop, please contact contact@dreamingspires.dev or visit dreamingspires.dev.