artbench-pd-256x256

# Dataset Card for ArtBench Public Domain 256x256


[日本語はこちら](README_JA.md)

This repository is the subset of ArtBench.
[ArtBench](https://github.com/liaopeiyuan/artbench) is the dataset for historical arts such as Art Nouveau and Ukiyo-e.
I picked up public domain images from ArtBench. Then, I create new dataset.

# Usage
You can use huggingface datasets to download the dataset.
You can also download [the tar file](artbench-pd-256x256.tar).

```python
from datasets import load_dataset

dataset = load_dataset("alfredplpl/artbench-pd-256x256")
```

# Intended Use
- Research Vision & Language
- Develop text-to-image model or image-to-text model.

# Procedure
- caption

We captioned the images by Florence-2-large.
We gave the model DETAILED_CAPTION task.
- caption_jp

We captioned the images by Qwen2-VL-2B-Instruct.

# Captioning Cost
- 30 GPUhours (RTX 4090)

# Citation

```bibtex
@article{liao2022artbench,
title={The ArtBench Dataset: Benchmarking Generative Models with Artworks},
author={Liao, Peiyuan and Li, Xiuyu and Liu, Xihui and Keutzer, Kurt},
journal={arXiv preprint arXiv:2206.11404},
year={2022}
}
```

# Acknowledgement
We appreciate [Peiyuan Liao](https://liaopeiyuan.com/) to create the great dataset.
This dataset leads futher researchs and development.

# ArtBench 公有领域 256x256 数据集卡片

[查看日文版请点击此处](README_JA.md)

本仓库为ArtBench数据集的子集。ArtBench（链接：https://github.com/liaopeiyuan/artbench）是面向新艺术运动（Art Nouveau）、浮世绘（Ukiyo-e）等历史艺术作品的数据集。本数据集从ArtBench中遴选公有领域图像，重新构建而成。

## 使用方法
您可通过Hugging Face Datasets库下载本数据集，亦可直接下载[打包文件（tar格式）](artbench-pd-256x256.tar)。

python
from datasets import load_dataset

dataset = load_dataset("alfredplpl/artbench-pd-256x256")

## 适用场景
- 视觉与语言方向研究
- 开发文本生成图像（text-to-image）或图像生成文本（image-to-text）模型

## 标注流程
- 英文标注（caption）
我们使用Florence-2-large模型对图像进行标注，为模型配置**DETAILED_CAPTION**（详细描述）任务。
- 日文标注（caption_jp）
我们使用Qwen2-VL-2B-Instruct模型对图像进行标注。

## 标注算力成本
30 GPU小时（基于RTX 4090显卡）

## 引用格式
bibtex
@article{liao2022artbench,
title={The ArtBench Dataset: Benchmarking Generative Models with Artworks},
author={Liao, Peiyuan and Li, Xiuyu and Liu, Xihui and Keutzer, Kurt},
journal={arXiv preprint arXiv:2206.11404},
year={2022}
}

## 致谢
感谢Peiyuan Liao（链接：https://liaopeiyuan.com/）构建了这一优秀数据集，本数据集为后续研究与开发工作提供了重要支撑。