famousdetectiveadrianmonk's picture
Upload dataset
b3a32e0 verified
metadata
pretty_name: ㅇㄹ
dataset_info:
  features:
    - name: input_image
      dtype: image
    - name: edited_image
      dtype: image
    - name: edit_prompt
      dtype: string
    - name: input_caption
      dtype: string
    - name: edited_prompt
      dtype: string
    - name: new_edited_image
      dtype: image
  splits:
    - name: train
      num_bytes: 1239264513
      num_examples: 1000
  download_size: 1239217373
  dataset_size: 1239264513
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Description

This dataset is an enhanced version of the original dataset provided by https://huggingface.co/fusing. The original dataset, https://huggingface.co/datasets/fusing/instructpix2pix-1000-samples, consisted of input images and edit prompts designed to generate edited images.

In this custom dataset, we have added additional fields and data transformations to facilitate advanced image editing tasks. Specifically, the dataset now includes:

  • input_caption: Captions generated using a vision-language model (VLM) from the input images.
  • edited_prompt: Prompts generated using a large language model (LLM) by combining the input captions with the original edit prompts.
  • new_edited_image: Images generated using a ControlNet img2img model by applying the edited prompts to the input images.

The transformations provide a robust framework for tasks such as image captioning, prompt-based image editing, and comparing different image editing models.

Changes Made

  • Added input_caption field generated with a vision-language model (VLM).
  • Added edited_prompt generated from a large language model (LLM) using the input_caption and edit_prompt.
  • Added new_edited_image generated from a ControlNet img2img model using the input_image and edited_prompt.

Attribution

This dataset is based on the original dataset provided by https://huggingface.co/fusing. The original dataset can be found at https://huggingface.co/datasets/fusing/instructpix2pix-1000-samples.