Hugging Face's logo

Junwei-Xi
/
Dual-Data-Alignment

Model card Files
xet
 Community
Junwei-Xi commited on
Commit
a6daba8
·
verified ·
1 Parent(s): 302a3cc

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +1 -121
README.md CHANGED
@@ -2,124 +2,4 @@
2
  license: apache-2.0
3
  datasets:
4
  - Junwei-Xi/DDA-Training-Set
5
- ---
6
- <div align="center">
7
- <h1> Dual Data Alignment [NeurIPS'25 Spotlight]</h1>
8
- <h3>Dual Data Alignment Makes AI-Generated Image Detector Easier Generalizable</h3>
9
-
10
- Ruoxin Chen<sup>1</sup>, Junwei Xi<sup>2</sup>, Zhiyuan Yan<sup>3</sup>, Keyue Zhang<sup>1</sup>, Shuang Wu<sup>1</sup>,
11
- Jingyi Xie<sup>4</sup>, Xu Chen<sup>2</sup>, Lei Xu<sup>5</sup>, Isabel Guan<sup>6†</sup>, Taiping Yao<sup>1†</sup>, Shouhong Ding<sup>1</sup>
12
-
13
-
14
- <sup>1</sup>Tencent YouTu Lab
15
- <sup>2</sup>East China University of Science and Technology
16
- <sup>3</sup>Peking University
17
-
18
- <sup>4</sup>Renmin University of China
19
- <sup>5</sup>Shenzhen University
20
- <sup>6</sup>Hong Kong University of Science and Technology
21
-
22
-
23
- <!-- [[GitHub](https://github.com/roy-ch/Dual-Data-Alignment)] [[Paper](https://arxiv.org/abs/2505.14359)] [[Dataset (Coming Soon)]()] -->
24
- [![BFree](https://img.shields.io/badge/Project%20page-222222.svg?style=for-the-badge&logo=github)](https://github.com/roy-ch/Dual-Data-Alignment)
25
- [![arXiv](https://img.shields.io/badge/-arXiv-B31B1B.svg?style=for-the-badge&logo=arXiv)](https://arxiv.org/abs/2505.14359)
26
- </div>
27
-
28
- ## 📣 News
29
-
30
- - `2025/09` : 🎉 Accepted by NeurIPS 2025 as **Spotlight**.
31
- <!-- - `2025/08` : 🏆 DDA (Ke-Yue Zhang's team) wins **1st Prize** at the [The 6th Face Anti-Spoofing Workshop: Unified Physical-Digital Attacks Detection@ICCV2025]((https://sites.google.com/view/face-anti-spoofing-challenge/winners-results/challengeiccv2025)) ! Notably, 🔥 our winner model is exclusively trained on DDA-aligned COCO, without using any competition-provided face data. **A model trained with no face data wins a face anti-spoofing competition**.-->
32
- <!-- - - `2025/10` : 🏆 **[ICCV 2025 FAS Challenge: 1st Prize](https://sites.google.com/view/face-anti-spoofing-challenge/winners-results/challengeiccv2025) (Ke-Yue Zhang’s team)**-->
33
-
34
- ---
35
-
36
- <!-- 两图一行：bias 左边，benchmark 右边 -->
37
- <!-- <div style="display:flex; justify-content:space-between; align-items:center; margin:20px 0;">
38
- <img src="assets/bias.png" style="max-width:48%; height:auto;" />
39
- <img src="assets/BenchmarkComparison.png" style="max-width:48%; height:auto;" />
40
- </div> -->
41
-
42
- ## 🤖 Motivation
43
- <div style="text-align:center; margin:20px 0;">
44
- <img src="assets/motivation.png" style="max-width:60%; height:auto;" />
45
- </div>
46
-
47
- ---
48
- ## 📊 Evaluation on 11 benchmarks
49
- JPEG compression with a quality factor of 96 is applied to the synthetic images in GenImage, ForenSynths, and AIGCDetectionBenchmark to mitigate format bias. The number of generators used in each dataset is reported: G refers to GAN, D to Diffusion, and AR to Auto-Regressive models. Among the 11 benchmarks, Chameleon, Synthwildx, WildRF, and Bfree-Online are the 4 in-the-wild datasets. Notably, DDA is **the first detector** to achieve over 80% cross-data accuracy on Chameleon.
50
-
51
- | Benchmark | NPR (CVPR'24) | UnivFD (CVPR'23) | FatFormer (CVPR'24) | SAFE (KDD'25) | C2P-CLIP (AAAI'25) | AIDE (ICLR'25) | DRCT (ICML'24) | AlignedForensics (ICLR'25) | DDA (ours) |
52
- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
53
- | GenImage (1G + 7D) | 51.5 ± 6.3 | 64.1 ± 10.8 | 62.8 ± 10.4 | 50.3 ± 1.2 | 74.4 ± 8.4 | 61.2 ± 11.9 | 84.7 ± 2.7 | 79.0 ± 22.7 | **91.7 ± 7.8** |
54
- | DRCT-2M (16D) | 37.3 ± 15.0 | 61.8 ± 8.9 | 52.2 ± 5.7 | 59.3 ± 19.2 | 59.2 ± 9.9 | 64.6 ± 11.8 | 90.5 ± 7.4 | 95.5 ± 6.1 | **98.1 ± 1.4** |
55
- | DDA-COCO (5D) | 42.2 ± 5.4 | 52.4 ± 1.5 | 51.7 ± 1.5 | 49.9 ± 0.3 | 51.3 ± 0.6 | 50.0 ± 0.4 | 60.2 ± 4.3 | 86.5 ± 19.1 | **92.2 ± 10.6** |
56
- | EvalGEN (3D + 2AR) | 2.9 ± 2.7 | 15.4 ± 14.2 | 45.6 ± 33.1 | 1.1 ± 0.6 | 38.9 ± 31.2 | 19.1 ± 11.1 | 77.8 ± 5.4 | 68.0 ± 20.7 | **97.2 ± 4.2** |
57
- | Synthbuster (9D) | 50.0 ± 2.6 | 67.8 ± 14.4 | 56.1 ± 10.7 | 46.5 ± 20.8 | 68.5 ± 11.4 | 53.9 ± 18.6 | 84.8 ± 3.6 | 77.4 ± 25.0 | **90.1 ± 5.6** |
58
- | ForenSynths (11G) | 47.9 ± 22.6 | 77.7 ± 16.1 | 90.0 ± 11.8 | 49.7 ± 2.7 | **92.0 ± 10.1** | 59.4 ± 24.6 | 73.9 ± 13.4 | 53.9 ± 7.1 | 81.4 ± 13.9 |
59
- | AIGCDetectionBenchmark (7G + 10D) | 53.1 ± 12.2 | 72.5 ± 17.3 | 85.0 ± 14.9 | 50.3 ± 1.1 | 81.4 ± 15.6 | 63.6 ± 13.9 | 81.4 ± 12.2 | 66.6 ± 21.6 | **87.8 ± 12.6** |
60
- | Chameleon (Unknown) | 59.9 | 50.7 | 51.2 | 59.2 | 51.1 | 63.1 | 56.6 | 71.0 | **82.4** |
61
- | Synthwildx (3D) | 49.8 ± 10.0 | 52.3 ± 11.3 | 52.1 ± 8.2 | 49.1 ± 0.7 | 57.1 ± 4.2 | 48.8 ± 0.8 | 55.1 ± 1.8 | 78.8 ± 17.8 | **90.9 ± 3.1** |
62
- | WildRF (Unknown) | 63.5 ± 13.6 | 55.3 ± 5.7 | 58.9 ± 8.0 | 57.2 ± 18.5 | 59.6 ± 7.7 | 58.4 ± 12.9 | 50.6 ± 3.5 | 80.1 ± 10.3 | **90.3 ± 3.5** |
63
- | Bfree-Online (Unknown) | 49.5 | 49.0 | 50.0 | 50.5 | 50.0 | 53.1 | 55.7 | 68.5 | **95.1** |
64
- | **Avg ACC** | 46.1 ± 16.1 | 56.3 ± 16.5 | 59.6 ± 14.6 | 47.6 ± 16.0 | 62.1 ± 15.6 | 54.1 ± 12.8 | 70.1 ± 14.6 | 75.0 ± 11.1 | **90.7 ± 5.3** |
65
- | **Min ACC** | 2.9 | 15.4 | 45.6 | 1.1 | 38.9 | 19.1 | 50.6 | 53.9 | **81.4** |
66
-
67
- ---
68
-
69
- ## 📦 Training data
70
-
71
- The training dataset has been released on [ModelScope](https://www.modelscope.cn/datasets/JunweiXi/DDA-Training-Set) and [HuggingFace](https://huggingface.co/datasets/Junwei-Xi/DDA-Training-Set).
72
-
73
- ---
74
-
75
- ## 📑 Checkpoints
76
-
77
- The checkpoint has been released on [ModelScope](https://modelscope.cn/datasets/roych1997/Dual_Data_Alignment/files).
78
-
79
- ---
80
-
81
- ## ⭐ New Challenging Benchmarks
82
-
83
- DDA-COCO Benchmark has been released on [ModelScope](https://www.modelscope.cn/datasets/JunweiXi/DDA-COCO) and [HuggingFace](https://huggingface.co/datasets/Junwei-Xi/DDA-COCO).
84
-
85
- EvalGEN Benchmark has been released on [ModelScope](https://www.modelscope.cn/datasets/JunweiXi/EvalGEN) and [HuggingFace](https://huggingface.co/datasets/Junwei-Xi/EvalGEN).
86
-
87
- ---
88
-
89
- ## 🎯 ToDo List <a name="todo"></a>
90
-
91
- - [x] Release arxiv paper with complete BibTeX citation
92
- - [x] Release checkpoint and inference code
93
- - [x] Release training set and training script
94
- - [ ] Release code for DDA data construction
95
- - [ ] Release model and code for ICCV 2025 FAS Challenge
96
-
97
-
98
- ## 📨 Contact
99
-
100
- If you have any questions or suggestions, please feel free to contact us
101
- at [cusmochen@tencent.com](cusmochen@tencent.com).
102
-
103
- Welcome to discuss with us if you have any questions
104
- <p align="center">
105
- <img src="assets/QRcode.jpg" width="200">
106
- </p>
107
-
108
-
109
-
110
- ## 😄 Acknowledgement
111
-
112
- Part of this codebase is adapted from [UniversalFakeDetect](https://github.com/WisconsinAIVision/UniversalFakeDetect). Huge thanks to the original authors for sharing their excellent work!
113
-
114
-
115
- ## ✍️ Citing
116
- If you find this repository useful for your work, please consider citing it as follows:
117
- ```shell
118
- @inproceedings{chen2025dual,
119
- title={Dual Data Alignment Makes {AI}-Generated Image Detector Easier Generalizable},
120
- author={Ruoxin Chen and Junwei Xi and Zhiyuan Yan and Ke-Yue Zhang and Shuang Wu and Jingyi Xie and Xu Chen and Lei Xu and Isabel Guan and Taiping Yao and Shouhong Ding},
121
- booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems},
122
- year={2025},
123
- url={https://openreview.net/forum?id=C39ShJwtD5}
124
- }
125
- ```
 
2
  license: apache-2.0
3
  datasets:
4
  - Junwei-Xi/DDA-Training-Set
5
+ ---