yuxi-liu-wired's picture
Update README.md
88429c4 verified
metadata
license: cc0-1.0
language:
  - en
annotations_creators:
  - no-annotation
size_categories:
  - 1M<n<10M
source_datasets:
  - stylebreeder/stylebreeder
pretty_name: stylebreeder
layout: default
tags:
  - stable diffusion
  - prompt engineering
  - prompts
  - research paper
dataset_info:
  - name: style_embedding
    dtype: sequence
    sequence:
      dtype: float16
  - name: content_embedding
    dtype: sequence
    sequence:
      dtype: float16
  - name: image
    dtype: image

This is a version of stylebreeder/stylebreeder, the 2M split.

I took the 2M split, then ran the CSD on every image, resulting in two 768-dimensional embedding vectors for each image. I also saved the images in a much smaller resolution in JPEG format to save space.

I apologize that the dataviewer doesn't work. I uploaded the dataset directly to Huggingface via the website UI and it seems their autoconversion broke something. I can't load the dataset directly because it runs out of memory locally, and I don't know how to stream a dataset from disk.