cholpona/pizza-burger-sushi-recipe
pizza-burger-sushi (recipe-only) This repo contains no image data. It contains a single script (recreate.py) and instructions for rebuilding the merged 3-class dataset used to fine-tune cholpona/vit-base-patch16-224-pizza-burger-sushi. The underlying images are not redistributed here because the burger source dataset does not specify a license. Reproducing the data locally from the original sources respects each upstream's terms. What it produces… See the full description on the dataset page: https://huggingface.co/datasets/cholpona/pizza-burger-sushi-recipe.
pizza-burger-sushi (recipe-only)
This repo contains no image data. It contains a single script (recreate.py) and instructions for rebuilding the merged 3-class dataset used to fine-tune `cholpona/vit-base-patch16-224-pizza-burger-sushi`.
The underlying images are not redistributed here because the burger source dataset does not specify a license. Reproducing the data locally from the original sources respects each upstream's terms.
What it produces
<out>/pizza_burger_sushi/data/
train/{pizza,burger,sushi}/ ~150 images per class (454 total)
test/{pizza,burger,sushi}/ 46 images per class (138 total)Source datasets
How to reproduce
pip install huggingface_hub Pillow
python recreate.py --out ./dataThen point your training code at ./data/pizza_burger_sushi/data/{train,test}/<class>/.
The recipe
- Download
Shad0wKillar/pizza_steak_sushi(already hasdata/{train,test}/{pizza,steak,sushi}/). - Download
NexaAIalex/Burger. - Resize burger images: max side 384 px, RGB, JPEG quality 90.
- Shuffle the resized burger images with
random.Random(42), take the first 150 fortrain/, the next 46 fortest/. - Assemble the final layout
pizza_burger_sushi/data/{train,test}/{pizza,burger,sushi}/, dropping thesteakclass from the source.
recreate.py does all of the above in one shot and is fully deterministic given SEED=42. It is also idempotent — re-running skips files that already exist.
License
- This recipe (script + README): MIT.
- The source datasets retain their own licenses (see table above). This repo does not redistribute their data.
