README.md

---
license: mit
task_categories:
- image-to-image
tags:
- computer-vision
- image-editing
- exemplar-based
size_categories:
- 1K<n<10K
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
dataset_info:
  features:
  - name: x_original
    dtype: image
  - name: x_edited
    dtype: image
  - name: y_original
    dtype: image
  - name: y_edited
    dtype: image
  splits:
  - name: train
    num_bytes: 1207235952.0
    num_examples: 1472
  download_size: 1142011780
  dataset_size: 1207235952.0
---

# ReEdit-Bench: Benchmark Dataset for Exemplar-Based Image Editing

A curated dataset of ~1,500 samples for evaluating exemplar-based image editing methods, as presented in our WACV '25' paper - **ReEdit: Multimodal Exemplar-Based Image Editing with Diffusion Models**

[![Project Page](https://img.shields.io/badge/Project-Page-blue)](https://reedit-diffusion.github.io/)  [![arXiv](https://img.shields.io/badge/arXiv-Paper-b31b1b)](https://arxiv.org/abs/2411.03982)  [![GitHub](https://img.shields.io/badge/GitHub-Code-181717?logo=github)](https://github.com/reedit-diffusion/reedit-diffusion)


<img src="https://cdn-uploads.huggingface.co/production/uploads/675d4f54dfdaa78daabe8e71/G1G1laTDtfd598f_z7t6v.jpeg" alt="reedit_overview" width="800"/>

## Dataset Structure

Each sample contains 4 images representing an exemplar edit pair:
- `x_original`: Source image before editing
- `x_edited`: Source image after editing (defines the edit operation)  
- `y_original`: Target image before editing
- `y_edited`: Target image after the same edit is applied

## Dataset Description

This dataset was carefully curated from InstructP2P samples, with manual visual inspection to ensure high quality. The edit operation demonstrated on (x_original → x_edited) should be apllied to y_original. y_edit denotes the gold standard ground truth for the edited target

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("{dataset_name}")
sample = dataset[0]

# Access images
x_orig = sample['x_original']
x_edit = sample['x_edited'] 
y_orig = sample['y_original']
y_edit = sample['y_edited']
```

## Citation

If you use this dataset, please cite the associated paper
```
@InProceedings{Srivastava_2025_WACV,
    author    = {Srivastava, Ashutosh and Menta, Tarun Ram and Java, Abhinav and Jadhav, Avadhoot Gorakh and Singh, Silky and Jandial, Surgan and Krishnamurthy, Balaji},
    title     = {ReEdit: Multimodal Exemplar-Based Image Editing},
    booktitle = {Proceedings of the Winter Conference on Applications of Computer Vision (WACV)},
    month     = {February},
    year      = {2025},
    pages     = {929-939}
}
```