File size: 2,405 Bytes
357e504
432b7a0
dbc4771
 
432b7a0
 
 
 
 
 
 
357e504
432b7a0
 
 
 
 
 
 
dbc4771
 
 
 
432b7a0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
dbc4771
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
---
pretty_name: DMImage Test Set
license: other
license_name: research-use-only
task_categories:
- image-classification
tags:
- ai-generated-image-detection
- synthetic-image-detection
- deepfake-detection
- benchmark
---

# DMImage Test Set

A constructed test set for **AI-generated image detection**, containing **21,000 images**:
**5,000 real** photographs and **16,000 synthetic** images produced by 30 generator
configurations spanning GANs, diffusion models, and autoencoder-based generators.

The images are distributed as a single archive, `DMImage-FullTestSet.zip` (5.5 GB).
Extract it to obtain the `TestSet/<subset>/` layout. Subsets whose name starts with
`real_` contain real photographs; all other subsets contain images generated by the
model indicated by the folder name.

## Composition

| Family | Subsets | Images |
|---|---|---:|
| Real | `real_imagenet_val`, `real_coco_valid`, `real_ucid` | 5,000 |
| GAN | `progan_lsun`, `biggan_256/512`, `stylegan2_*`, `stylegan3_*`, `eg3d` | 5,000 |
| Diffusion | `stable_diffusion_256`, `latent-diffusion_*`, `guided-diffusion_*`, `glide_text2img_valid`, `dalle-mini_valid`, `dalle_2` | 9,000 |
| Autoencoder | `taming-transformers_*` | 2,000 |

- Real: ImageNet val (2,000, `.JPEG`), COCO val (2,000, `.jpg`), UCID (1,000, `.tif`)
- Synthetic: `.png`, resolution indicated by the folder name (256 / 512 / 1024)

## operations.csv

`TestSet/operations.csv` records, for every image, the randomized crop and JPEG
quality parameters used in the robustness evaluation protocol:

| Column | Meaning |
|---|---|
| `src` | image path relative to `TestSet/` |
| `cropsize`, `x1`, `y1` | crop size and top-left offset |
| `qf` | JPEG quality factor |
| `typ` | subset name |
| `label` | `REAL` for real images, `FALSE` for generated images |

## Usage

```python
from pathlib import Path
from PIL import Image

root = Path("TestSet")
for path in root.glob("*/*"):
    is_real = path.parent.name.startswith("real_")
    image = Image.open(path)
```

This is an evaluation-only test set; it is not intended for training.

## License

Released for **research use only**. The real images originate from ImageNet, COCO,
and UCID and remain subject to their original licenses; the synthetic images were
generated with the corresponding models. Commercial use is not permitted, and users
are responsible for complying with the terms of the original sources.