Add dataset: brill_iconclass

[davanstrien]

A URL for this dataset

https://labs.brill.com/ictestset/

Dataset description

A test dataset and challenge to apply machine learning to collections described with the Iconclass classification system.

Iconclass is a metadata standard used by some LAM institutions. This dataset is of particular interest for the following reasons:

• the dataset includes images relevant to the LAM domain (art history)
• some images contain multiple iconclass labels whilst others contain a single one
• the iconclass schema itself poses an interesting machine learning challenge since the notation of the schema is broken down into parts. This means that whilst the dataset can be used as a image classification dataset is likely to benefit from more bespoke sequence prediction approaches.

Dataset modality

Image

Dataset licence

Creative Commons Public Domain Dedication and Certification

Other licence

No response

How can you access this data

As a download from a repository/website

Confirm the dataset has an open licence

• To the best of my knowledge, this dataset is accessible via an open licence

Contact details for data custodian

posthumus@brill.com

[cakiki]

Looks great! Removing the candidate tag 😃

[davanstrien]

#self-assign

[davanstrien]

#ready-for-review

[davanstrien]

I have written a script for loading this one: https://huggingface.co/datasets/biglam/cultural_heritage_metadata_accuracy

@albertvillanova I didn't make this one streaming. I did have a version that supported streaming but it was quite a bit slower to load. It's possible I missed something obvious though.

I used the following to generate examples in the streaming version:

    def _generate_examples(self, download_dir):
        with ZipFile(download_dir) as myzip:
            with myzip.open("data.json") as json_file:
                data = json.load(json_file)
                for row, item in enumerate(data.items()):
                    filepath, labels = item
                    image = Image.open(myzip.open(filepath))
                    yield row, {"image": image, "label": labels}

My own feeling is that streaming is less important for this one but if I've missed an obvious way of supporting streaming happy to hear it!

[epoz]

BTW, I am not affiliated with Brill any more, so my contact address should be updated. You can use info@iconclass.org for testset related matters.

[epoz]

Is it an idea to add the core data of the Iconclass system as a dataset?

[davanstrien]

BTW, I am not affiliated with Brill any more, so my contact address should be updated. You can use info@iconclass.org for testset related matters.

I will update that 🙂

[davanstrien]

Is it an idea to add the core data of the Iconclass system as a dataset?

That would be great. I had originally planned to also add a configuration of this dataset that had the 'translation' of the iconclass labels i.e. turing the iconclass code into the associated description. I know there used to be a Python library that allowed for these queries but I think it's no longer maintained? Adding the core data as a dataset would both be nice as its own dataset but could also potentially be used as a way of doing this 'translation'.

[epoz]

Ouch, yes. that Python library was also made by me, but has terrible (read: non-existent) documentation.
Would love to update it, but there have been other things clamouring for attention.

If there is interest in using it, will galvanise me to give it some spit-and-polish.

[davanstrien]

Ouch, yes. that Python library was also made by me, but has terrible (read: non-existent) documentation. Would love to update it, but there have been other things clamouring for attention.

If there is interest in using it, will galvanise me to give it some spit-and-polish.

That would be great, happy to offer some help with that if useful. Its possible some of the functionality could be replicated by having the underlying data available on the hub but some areas might be better served by a specific library.

[epoz]

I have updated the location of this testset, it is now on:
https://iconclass.org/testset/

[davanstrien]

I have updated the location of this testset, it is now on: iconclass.org/testset

Thanks for letting me know — just updated the URLs.