A convenience mirror of the two ground-truth image sets used as benchmarks for diffusion-based inverse problems (super-resolution, Gaussian/motion deblurring, inpainting, …), bundled with the text prompts that accompany each set. Everything here is the clean ground truth — no degraded / measurement images are included. Degradations are meant to be generated on the fly by the operator A of whichever task you are running.

Contents… See the full description on the dataset page: https://huggingface.co/datasets/dwchoi/div2k800-ffhq1000.