emu_edit_test_set

# Dataset Card for the Emu Edit Test Set

## Table of Contents
- [Table of Contents](#table-of-contents)
- [Dataset Description](#dataset-description)
- [Dataset Summary](#dataset-summary)
- [Additional Information](#additional-information)
- [Licensing Information](#licensing-information)
- [Citation Information](#citation-information)

## Dataset Description

- **Homepage: https://emu-edit.metademolab.com/**
- **Paper: https://emu-edit.metademolab.com/assets/emu_edit.pdf**

### Dataset Summary

To create a benchmark for image editing we first define seven different categories of potential image editing operations: background alteration (background), comprehensive image changes (global), style alteration (style), object removal (remove), object addition (add), localized modifications (local), and color/texture alterations (texture).
Then, we utilize the diverse set of input images from the [MagicBrush benchmark](https://huggingface.co/datasets/osunlp/MagicBrush), and for each editing operation, we task crowd workers to devise relevant, creative, and challenging instructions.
Moreover, to increase the quality of the collected examples, we apply a post-verification stage, in which crowd workers filter examples with irrelevant instructions.
Finally, to support evaluation for methods that require input and output captions (e.g. prompt2prompt and pnp), we additionally collect an input caption and output caption for each example.
When doing so, we ask annotators to ensure that the captions capture both important elements in the image, and elements that should change based on the instruction.
Additionally, to support proper comparison with Emu Edit with publicly release the model generations on the test set [here](https://huggingface.co/datasets/facebook/emu_edit_test_set_generations).
For more details please see our [paper](https://emu-edit.metademolab.com/assets/emu_edit.pdf) and [project page](https://emu-edit.metademolab.com/).

### Licensing Information

Licensed with CC-BY-NC 4.0 License available [here](https://creativecommons.org/licenses/by-nc/4.0/legalcode?fbclid=IwAR2SYZjLRywwUMblkWg0LyAxHVVTloIFlvC-ju3BthIYtOM2jpQHgbeXOsM).

### Citation Information
```
@inproceedings{Sheynin2023EmuEP,
title={Emu Edit: Precise Image Editing via Recognition and Generation Tasks},
author={Shelly Sheynin and Adam Polyak and Uriel Singer and Yuval Kirstain and Amit Zohar and Oron Ashual and Devi Parikh and Yaniv Taigman},
year={2023},
url={https://api.semanticscholar.org/CorpusID:265221391}
}
```

# Emu Edit 测试集数据集卡片

## 目录
- [目录](#table-of-contents)
- [数据集描述](#dataset-description)
- [数据集摘要](#dataset-summary)
- [附加信息](#additional-information)
- [授权信息](#licensing-information)
- [引用信息](#citation-information)

## 数据集描述

- **官方主页: https://emu-edit.metademolab.com/**
- **相关论文: https://emu-edit.metademolab.com/assets/emu_edit.pdf**

### 数据集摘要

为构建图像编辑基准测试集，我们首先定义了7类不同的图像编辑操作范畴：背景修改（background alteration，缩写为background）、全局图像修改（comprehensive image changes，缩写为global）、风格修改（style alteration，缩写为style）、对象移除（object removal，缩写为remove）、对象添加（object addition，缩写为add）、局部修改（localized modifications，缩写为local）以及色彩/纹理修改（color/texture alterations，缩写为texture）。
随后，我们选用来自[MagicBrush基准测试集](https://huggingface.co/datasets/osunlp/MagicBrush)的多样化输入图像，并针对每一类编辑操作，招募众包工作者设计贴合需求、富有创意且具备挑战性的编辑指令。
此外，为提升采集样本的质量，我们增设了后验证环节，由众包工作者筛选掉指令无关的样本。
最后，为支持需要输入/输出图像字幕（caption）的方法（如prompt2prompt与pnp）的评估工作，我们为每个样本额外采集了输入图像字幕与输出图像字幕。
在此环节中，我们要求标注人员确保图像字幕同时涵盖图像中的重要元素以及需根据编辑指令修改的元素。
此外，为确保与Emu Edit的公平对比，我们在此[公开发布了该测试集上的模型生成结果](https://huggingface.co/datasets/facebook/emu_edit_test_set_generations)。
如需了解更多细节，请参阅我们的[相关论文](https://emu-edit.metademolab.com/assets/emu_edit.pdf)与[项目主页](https://emu-edit.metademolab.com/)。

### 授权信息

本数据集采用CC-BY-NC 4.0授权协议，协议详情可参见[此处](https://creativecommons.org/licenses/by-nc/4.0/legalcode?fbclid=IwAR2SYZjLRywwUMblkWg0LyAxHVVTloIFlvC-ju3BthIYtOM2jpQHgbeXOsM)。

### 引用信息

@inproceedings{Sheynin2023EmuEP,
title={Emu Edit: Precise Image Editing via Recognition and Generation Tasks},
author={Shelly Sheynin and Adam Polyak and Uriel Singer and Yuval Kirstain and Amit Zohar and Oron Ashual and Devi Parikh and Yaniv Taigman},
year={2023},
url={https://api.semantics.org/CorpusID:265221391}
}