---
title: FakeCOCO
canonical_url: "https://www.modelscope.cn/datasets/acmiracle/FakeCOCO"
md_url: "https://www.modelscope.cn/datasets/acmiracle/FakeCOCO.md"
repository: acmiracle/FakeCOCO
last_updated: 2026-08-21
license: "MIT License"
storage_size: "207 GB"
downloads: 233
stars: 1
---

# FakeCOCO

> FakeCOCO - acmiracle 在 ModelScope 开源的数据集。Introduce FakeCOCO: AI-generated Image Detection Dataset for Incremental Learning from "Incremental Learning for AI-generated Image Detection".

acmiracle/FakeCOCO 是 ModelScope 魔搭社区上的数据集，存储大小 207 GB，采用 MIT License 许可。

- **Repository**: acmiracle/FakeCOCO
- **License**: MIT License
- **Storage size**: 207 GB
- **Downloads**: 233
- **Stars**: 1
- **Last updated**: 2026-08-21

Source: https://www.modelscope.cn/datasets/acmiracle/FakeCOCO

---

# Introduce
FakeCOCO: AI-generated Image Detection Dataset for Incremental Learning from "[Incremental Learning for AI-generated Image Detection](https://github.com/ac-miracle/FakeCOCO)".

# Download 
:modelscope-code[]{type="sdk"}
:modelscope-code[]{type="git"}

# Directory Tree
Each folder contains compressed files. After unzip the file, files under the data root directory can be organized as follows. The Fake folder contains the AI-generated images, and the Real folder contains the collected images from COCO.
- 📁 FakeCOCO/
  - 📁 Base/
    - 📁 train/
      - 📁 Real/
        - 🖼️ 000000004426.jpg
        - 🖼️ 000000004428.jpg
        - ...
      - 📁 Fake/
        - 🖼️ SD3_00001_.png
        - 🖼️ SDXL_00001_.png
        - ...
    - 📁 test/
    - 📁 val/
  - 📁 Midjourneyv7/
    - 📁 train/
      - 📁 Real/
      - 📁 Fake/
    - 📁 test/
    - 📁 val/
  - 📁 ChatGPT4o/
  - ...
- 📄 README.md

# More Details
<div align="center">
The FakeCOCO Details on the Dataset Split Among All Domains



| Domain | Total | Train (Real) | Train (Fake) | Validation (Real) | Validation (Fake) | Test (Real) | Test (Fake) |
| :--- | :---: | :---: | :---: | :---: | :---: | :---: | :---: |
| **Base** | 234,778 | 81,814 | 82,654 | 11,540 | 11,810 | 23,345 | 23,615 |
| **MidjourneyV6.1**| 1,380 | 420 | 420 | 135 | 135 | 135 | 135 |
| **Flux** | 1,380 | 420 | 420 | 135 | 135 | 135 | 135 |
| **SD3.5Large** | 1,380 | 420 | 420 | 135 | 135 | 135 | 135 |
| **Imagen3** | 1,344 | 408 | 408 | 132 | 132 | 132 | 132 |
| **NOVA** | 1,380 | 420 | 420 | 135 | 135 | 135 | 135 |
| **Ideogram3.0** | 1,212 | 364 | 364 | 121 | 121 | 121 | 121 |
| **MidjourneyV7** | 1,212 | 364 | 364 | 121 | 121 | 121 | 121 |
| **Seedream3** | 1,332 | 404 | 404 | 131 | 131 | 131 | 131 |
| **ChatGPT-4o** | 1,176 | 352 | 352 | 118 | 118 | 118 | 118 |
| **Harmon** | 1,380 | 420 | 420 | 135 | 135 | 135 | 135 |

</div>

# Reference
If you find our repository useful for your research, please consider citing our paper:

```bibtex
@ARTICLE{Cai2026Incremental,
  author={Cai, Zihao and Song, Xue and Li, Xinghan and Li, Bo and Shan, Haijun and Chen, Jingjing},
  journal={IEEE Transactions on Multimedia},
  title={Incremental Learning for AI-Generated Image Detection},
  year={2026},
  pages={1--10},
  doi={10.1109/TMM.2026.3716041}
}
```