mapo-t2i/pick-style-pixel-art

---
license: openrail++
library_name: diffusers
dataset_info:
features:
- name: caption
dtype: int64
splits:
- name: train
num_bytes: 2929653589
num_examples: 1000
download_size: 2929757570
dataset_size: 2929653589
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---

# Margin-aware Preference Optimization for Aligning Diffusion Models without Reference

<div align="center">
<img src="assets/mapo_overview.png" width=750/>
</div><br>

We propose **MaPO**, a reference-free, sample-efficient, memory-friendly alignment technique for text-to-image diffusion models. For more details on the technique, please refer to our paper [here](https://arxiv.org/abs/2406.06424).

## Developed by

* Jiwoo Hong^* (KAIST AI)
* Sayak Paul^* (Hugging Face)
* Noah Lee (KAIST AI)
* Kashif Rasul (Hugging Face)
* James Thorne (KAIST AI)
* Jongheon Jeong (Korea University)

## Dataset

This dataset is *pixel art* split of **Pick-Style**, self-curated with [Stable Diffusion XL](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0). Using the context prompts (i.e., without stylistic specifications), we generate (1) cartoon style generation with stylistic prefix prompt and (2) normal generation with context prompt. Then, (1) is used as the chosen image, and (2) as the rejected image. The *chosen* field comprises pixel art style generations from SDXL, while the *rejected* field comprises the ordinary generations from SDXL.

## Citation

```bibtex
@misc{hong2024marginaware,
title={Margin-aware Preference Optimization for Aligning Diffusion Models without Reference},
author={Jiwoo Hong and Sayak Paul and Noah Lee and Kashif Rasul and James Thorne and Jongheon Jeong},
year={2024},
eprint={2406.06424},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
```

原始信息汇总

数据集概述

数据集信息

• 许可证: openrail++
• 库名称: diffusers

数据集特征

• 特征名称: caption
• 数据类型: int64

数据集拆分

• 拆分名称: train
• 示例数量: 1000
• 数据大小: 2929653589 字节
• 下载大小: 2929757570 字节

配置

• 配置名称: default
• 数据文件路径: data/train-*