i1-datasets

i1: A Simple and Fully Open Recipe for Strong Text-to-Image Models
Boya Zeng, Tianze Luo, Shu Pu, Jucheng Shen, Taiming Lu, Gabriel Sarch, Zhuang Liu
Princeton University
[arXiv][code][model][project page]

This organization (along with zlab-princeton) stores the processed TFRecord datasets used for the controlled experiments and final i1-1B and i1-3B model training in the i1 paper.

We note that the processed TFRecord files for YFCC, RedCaps, and Megalith are currently not provided due to image license constraints. Please follow our data processing pipeline to download the images from URLs and combine them with our captions into TFRecord files.

256-Resolution Datasets

512-Resolution Datasets

1024-Resolution Datasets