xiaoxiaxu/highresolution-laioncoco-aesthetic-MEG

This dataset is filtered from laioncoco-aesthetic, which is used for academic research on mobile edge generation (MEG). It includes high-resolution 1024-by-1024 text-to-image samples generated by a distilled SDXL with 4-12 denoising steps. The dataset mainly involves the following fields: caption: The text prompt of the image. image: The target image corresponding to the prompt. diffusion: The generative results of the distilled SDXL. latents: The latent features of the distilled SDXL.