Pick-High Dataset
收藏资源简介：
Pick-High是一个高质量的数据集，包含360,000张由SD3.5-Large生成的图像，这些图像使用了Claude-3.5-Sonnet链式思维推理生成的精细提示，并结合Pick-a-Pic形成带有全面偏好注释的图像三元组，用于训练和评估奖励模型。
Pick-High is a high-quality dataset containing 360,000 images generated by SD3.5-Large. These images utilize fine-grained prompts created via Claude-3.5-Sonnet's Chain-of-Thought reasoning, and are combined with Pick-a-Pic to form image triplets with comprehensive preference annotations, which are designed for training and evaluating reward models.
ICTHP数据集概述
数据集基本信息
- 名称: Pick-High Dataset
- 用途: 用于训练和评估高质量图像生成的奖励模型
- 规模: 包含360,000张图像
- 生成方式: 使用SD3.5-Large模型生成，提示词通过Claude-3.5-Sonnet链式思考推理优化
- 标注类型: 图像三元组偏好标注
数据集结构
Pick-High-Dataset/ ├── Pick-High/ │ ├── train.pkl # 训练集标注 │ ├── val.pkl # 验证集标注 │ └── test.pkl # 测试集标注 ├── pick_easy_img/ # 基础质量图像 │ └── train/val/test/ └── pick_refine_img/ # 高质量优化图像 └── train/val/test/
相关模型
- ICT Reward Model: 评估文本-图像对齐质量
- HP Reward Model: 评估美学质量和人类偏好
获取方式
- Hugging Face: https://huggingface.co/datasets/8y/Pick-High-Dataset
- Git LFS:
git clone https://huggingface.co/datasets/8y/Pick-High-Dataset - datasets库:
load_dataset(8y/Pick-High-Dataset)
引用信息
bibtex @misc{ba2025enhancingrewardmodelshighquality, title={Enhancing Reward Models for High-quality Image Generation: Beyond Text-Image Alignment}, author={Ying Ba and Tianyu Zhang and Yalong Bai and Wenyi Mo and Tao Liang and Bing Su and Ji-Rong Wen}, year={2025}, eprint={2507.19002}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2507.19002}, }
许可证
- 类型: MIT License
