PixelGPT is a pixel art image dataset designed to provide training and evaluation data for small text-based image models, native low-resolution image generation, or autoregressive image modeling. This dataset contains 20,000 24×24 pixel art images, with 18,000 images in the training set and 1,000 images each in the validation and test sets. The dataset is divided into 19 primary categories and 133 secondary categories. Each pixel art image uses no more than 5 colors and is accompanied by an original English caption and two-level semantic classification labels. This dataset is a publicly available subset of the PixelGPT 24×24 dataset; the full version contains over 218,000 images and can be used for larger-scale model training.

Data fields:

• Image: Native 24x24 pixel art image
• caption: English text description of the image
• family: primary semantic category
• category: secondary semantic classification
• tags: descriptive tags

  

Citation

If you use this dataset in your research, please cite the data source: Pixelgpt-24×24-20k — https://huggingface.co/datasets/unstonio/pixelgpt-24×24-20k