SynthScars
收藏资源简介:
SynthScars是一个高质量且多样化的数据集,包含12,236张完全合成的图像,并带有专家注释。该数据集具有4种不同的图像内容类型、3类伪影,以及涵盖像素级分割、详细文本解释和伪影类别标签的细粒度注释。
SynthScars is a high-quality and diverse dataset consisting of 12,236 fully synthetic images accompanied by expert annotations. It features four distinct image content types, three artifact categories, and fine-grained annotations covering pixel-level segmentation, detailed textual explanations, and artifact category labels.
LEGION: Learning to Ground and Explain for Synthetic Image Detection 数据集概述
📌 数据集基本信息
- 数据集名称: SynthScars
- 数据量: 12,236张全合成图像
- 标注类型: 人工专家标注
- 内容类型: 4种不同图像内容类型
- 标注粒度:
- 像素级分割
- 详细文本解释
- 伪影类别标签
🏆 数据集特点
- 高质量多样性: 包含多种内容类型和伪影类别
- 精细标注:
- 3类伪影标注
- 像素级分割标注
- 详细文本解释
📦 数据集结构
./data └── SynthScars ├── train │ ├── images │ └── annoations │ └── train.json └── test ├── images └── annoations └── test.json
📊 性能表现
- 伪影定位: 在SynthScars、RichHF-18K和LOKI数据集上评估
- 解释生成: 在SynthScars和LOKI数据集上评估
- 深度伪造检测: 在UniversialFakeDetect基准测试上评估
🛠️ 使用方式
- 作为防御者:
- 伪影定位和解释生成训练
- 深度伪造检测训练
- 作为控制器:
- 图像再生
- 区域修复
📜 引用格式
bibtex @misc{kang2025legionlearninggroundexplain, title={LEGION: Learning to Ground and Explain for Synthetic Image Detection}, author={Hengrui Kang and Siwei Wen and Zichen Wen and Junyan Ye and Weijia Li and Peilin Feng and Baichuan Zhou and Bin Wang and Dahua Lin and Linfeng Zhang and Conghui He}, year={2025}, eprint={2503.15264}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2503.15264}, }
🔗 相关资源
- 论文地址: https://arxiv.org/pdf/2503.15264
- 项目页面: https://opendatalab.github.io/LEGION/
- 数据集下载: https://huggingface.co/datasets/khr0516/SynthScars




