| # NOTICE |
|
|
| StatMapCorpus distributes **annotations, identifiers and image embeddings only**. It does |
| not contain, mirror or redistribute any images. Each record points to an image hosted by a |
| third party (`url`). It also carries verification surrogates (`sha256`, `phash`, |
| `image_width`, `image_height`) that let a user confirm that a file retrieved later is the |
| file that was annotated. |
|
|
| ## Provenance |
|
|
| ``` |
| Common Crawl → CommonPool / DataComp → MapPool → StatMapCorpus |
| ``` |
|
|
| - **MapPool**: 75,890,407 map-like images identified in CommonPool, released under |
| CC BY 4.0. |
| - Reference: Schnürer, R. (2024). MapPool – bubbling up an extremely large corpus of |
| maps for AI. *Abstracts of the ICA* 8, 22. https://doi.org/10.5194/ica-abs-8-22-2024 |
| - Available at https://huggingface.co/datasets/sraimund/MapPool |
| - StatMapCorpus takes `uid`, `url` and the CLIP ViT-L/14 image embeddings from MapPool, |
| which carries them from CommonPool. |
| - **Modified:** a subset was selected and annotated anew. |
| - **CommonPool / DataComp**: an index of image–text pairs from Common Crawl. |
| - Reference: Gadre, S. Y. et al. (2023). DataComp: in search of the next generation of |
| multimodal datasets. *Advances in Neural Information Processing Systems* 36, |
| 27092–27112. |
| - Its authors distribute the url–text samples and metadata under CC BY 4.0. They state |
| that the individual images remain under their own copyrights, that the index is |
| provided as is, and that any content downloaded from it is used at the user's own risk |
| and subject to the terms or copyright limitations accompanying that content. |
| |
| Like the collections it derives from, StatMapCorpus stores links rather than images, for |
| copyright reasons. |
|
|
| Neither the author of MapPool nor the DataComp authors are affiliated with StatMapCorpus |
| or endorse it. |
|
|
| ## Licensing |
|
|
| | Component | Licence | |
| |---|---| |
| | Annotations, metadata, documentation (including `candidates.parquet` and `validation.csv`) | **CC BY 4.0**: https://creativecommons.org/licenses/by/4.0/legalcode | |
| | `download_images.py` | **MIT**: see the licence header in the file | |
| | CLIP embeddings | derived from MapPool, **CC BY 4.0** (attribution to MapPool required) | |
| | **Source images** | **Rights remain with their original owners. Not redistributed and not licensed here.** | |
|
|
| Downloading images retrieves files directly from their original hosts. Whether a given |
| image may be used for a given purpose depends on that image's own copyright status, not on |
| this dataset's licence. Users are responsible for their own compliance. |
|
|
| ## Removal requests |
|
|
| **Rights holders.** Removal requests are honoured. |
| - Write to **Marta Solarz, m.solarz2@uw.edu.pl**, Faculty of Geography and Regional |
| Studies, University of Warsaw, giving the `uid` or `url` of the affected record. |
| - Requests are acknowledged within 30 days. |
| - The affected records are removed and a corrected version of the dataset is published, |
| without waiting for a scheduled release. |
|
|
| **Personal data.** The images were collected by CommonPool from the public web, and an |
| audit of CommonPool found personal information in the wider collection (Hong et al. 2025, |
| https://arxiv.org/abs/2506.17185). StatMapCorpus does not redistribute images and consists |
| of maps rather than photographs of people. Even so, a linked image may contain personal |
| data. Data subjects may request erasure of a record under Art. 17 GDPR, either at the |
| address above or through the University of Warsaw Data Protection Officer at |
| iod@adm.uw.edu.pl. |
|
|
| Because the dataset contains no images, removal means deleting the identifier, its |
| annotations and its embedding from the dataset. The image itself remains under its owner's |
| sole control at all times. |
|
|
| ## Citation |
|
|
| Solarz, Marta, 2026, "StatMapCorpus: 23,549 English-language statistical maps from |
| MapPool, annotated by thematic map type using vision-language models", |
| https://doi.org/10.58132/FSJGSP, Dane Badawcze UW, V2. |
|
|
