Mini-YoChameleon-Data

# [Mini Yo'Chameleon Data](https://thaoshibe.github.io/YoChameleon) (CVPR 2025)

This is a mini-training-data for Yo'Chameleon, with example of personalized subject called `<bo>` (From [Yo'LLaVA](https://thaoshibe.github.io/YoLLaVA)).

What you will find:
- 10/10 positive images for training/ testing
- 1000 HARD negative images for training (retrieved from LAION-5B based on similarity with subject)
- 1000 random images for training

The folder structure:

```
mini-yochameleon-data
|_ random_negative_example
| |_ [1000 random images example for training recognition abilities]
|_ test
| |_ bo
| |_ [10 test images for <bo> subject]
|_ train
|_ bo
|_ [10 positive images for <bo> subject]
|_ negative_example
| |_ [1000 HARD negative images for <bo> subject, retrieved from LAION-5B]
|_ json
|_ 1000.json # image reconstruction
|_ recognition.json # image reconstruction -- with HARD NEGATIVE images (or so-called soft-positive)
|_ text_conversation.json # image reconstruction
```


[Paper](https://huggingface.co/papers/2504.20998)

```
@article{yochameleon,
title={Yo'Chameleon: Personalized Vision and Language Generation},
author={Thao Nguyen and Krishna Kumar Singh and Jing Shi and Trung Bui and Yong Jae Lee and Yuheng Li},
journal={2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2025},
}
```

# [Mini Yo'Chameleon 数据集](https://thaoshibe.github.io/YoChameleon) (CVPR 2025)

本数据集为Yo'Chameleon的小型训练子集，包含名为`<bo>`的个性化主体示例（源自[Yo'LLaVA](https://thaoshibe.github.io/YoLLaVA)）。

数据集内容如下：
- 用于训练与测试的10张正样本图像
- 1000张训练用难负样本图像（基于与目标主体的相似度从LAION-5B中检索得到）
- 1000张训练用随机采样图像

数据集目录结构如下：

mini-yochameleon-data
|_ random_negative_example
| |_ 用于训练识别能力的1000张随机样本图像
|_ test
| |_ bo
| |_ 用于<bo>主体的10张测试图像
|_ train
|_ bo
|_ 用于<bo>主体的10张正样本图像
|_ negative_example
| |_ 从LAION-5B检索得到的1000张<bo>主体难负样本图像
|_ json
|_ 1000.json # 图像重构任务
|_ recognition.json # 图像重构任务——搭配难负样本（或称软正样本）
|_ text_conversation.json # 图像重构任务


[论文](https://huggingface.co/papers/2504.20998)

@article{yochameleon,
title={Yo'Chameleon: 个性化视觉与语言生成},
author={Thao Nguyen and Krishna Kumar Singh and Jing Shi and Trung Bui and Yong Jae Lee and Yuheng Li},
journal={2025 IEEE/CVF计算机视觉与模式识别会议（CVPR 2025）},
year={2025},
}