Datasets:
Tasks:
Image Classification
Modalities:
Image
Formats:
imagefolder
Size:
100K - 1M
Tags:
art
License:
license: openrail | |
task_categories: | |
- image-classification | |
tags: | |
- art | |
size_categories: | |
- 100K<n<1M | |
Used to detect severely stylistically corrupted anime images generated by Stable Diffusion. | |
There are two classes: `corrupted` and `normal`, comprising 61,100 and 63,004 images, respectively. | |
For the `corrupted` type, it includes several cases such as: | |
* Entirely black images | |
* Mosaics or stylistic anomalies due to low sampling steps | |
* Stylistic anomalies due to excessively weighted tags | |
* Stylistic anomalies due to embedding with excessive weights | |
It's important to note that the following content is not included: | |
* Distortions in facial and hand details caused by resolution issues | |
* Distortions in human and object structures | |
For the `normal` type, it contains regular AI-generated images as well as approximately 15,000 images hand-drawn by humans. | |