CoolFace
Datasetpublic

Amr292/mamluk-cairo-style-dataset

Mamluk Cairo architecture — style LoRA dataset 42 curated photos of 17 Mamluk-era buildings in Cairo, each paired with a training caption, used to train Amr292/mmlkcairo-sdxl-lora. Contents NNNN.jpg + NNNN.txt — image and caption (kohya / diffusers compatible), numbered 0002–0043 (0001, a skyline shot, was removed). metadata.csv — per image: building, view type, Wikimedia Commons title, source URL, author, license, license URL. scripts/collect_images.py —… See the full description on the dataset page: https://huggingface.co/datasets/Amr292/mamluk-cairo-style-dataset.

sourceHugging Faceotherupdated 3d agoView on Hugging Face
0likes34downloads
Dataset Card

Mamluk Cairo architecture — style LoRA dataset

42 curated photos of 17 Mamluk-era buildings in Cairo, each paired with a training caption, used to train Amr292/mmlkcairo-sdxl-lora.

Contents

  • —NNNN.jpg + NNNN.txt — image and caption (kohya / diffusers compatible), numbered 0002–0043 (0001, a skyline shot, was removed).
  • —metadata.csv — per image: building, view type, Wikimedia Commons title, source URL, author, license, license URL.
  • —scripts/collect_images.py — collection from Commons with license filtering; scripts/check_captions.py — caption rules checker.

How it was built

  1. 1.Collection: a hand-picked list of 17 Mamluk buildings, pulled from their Commons categories. Kept only freely licensed photos (CC BY, CC BY-SA, CC0, public domain) with a short side ≥ 1024 px, max 10 per building (a second pass allowed 25 for five buildings, to find more portals).
  2. 2.Exclusions for style purity: buildings with major Ottoman-era parts (Cairo Citadel, Bayt al-Suhaymi, the Blue Mosque), and historical prints, postcards and paintings.
  3. 3.Curation: ~237 unique candidates across two passes → 42, removing heavy clutter, filters, HDR, fisheye and near-duplicates; at most 5 per building.
  4. 4.Captioning rule — describe everything except the style: every caption starts with the trigger mmlkcairo, then names subject, viewpoint, light and all non-architectural content. Style words (Mamluk, Islamic, muqarnas, ablaq, ornate, carved…) are banned and enforced by check_captions.py.
ViewCount
Facades11
Domes8
Minarets8
Portals6
Courtyards5
Interiors4

Licensing

Each photo keeps its own Wikimedia Commons license, listed in metadata.csv with its author; CC BY-SA images remain under CC BY-SA. Please credit the authors when reusing. Captions and scripts: CC BY 4.0, Amr Mohammed.