README.md

Filtered and Recaptioned Dataset

This dataset was created from nbeerbower/NikuX-DPO-filtered with manually selected samples and edited prompts.

Dataset Information

• Original dataset: nbeerbower/NikuX-DPO-filtered
• Number of samples: 620
• Selected indices: 620 samples were selected and potentially recaptioned

Columns

• prompt: The edited/recaptioned text prompt
• chosen: The chosen/preferred image
• rejected: The rejected/non-preferred image

Creation Process

This dataset was created using a manual review tool that allowed:

1. Browsing through the original dataset
2. Selecting samples to keep
3. Editing/recaptioning the prompts
4. Exporting the filtered dataset

Usage

from datasets import load_dataset

dataset = load_dataset("nbeerbower/NikuFix")