---
license: mit
datasets:
- deepghs/anime_ai_check
metrics:
- accuracy
pipeline_tag: image-classification
tags:
- art
---

The model used to predict whether an image is generated by AI techniques, mainly diffusion models. Please note that this model is trained on crawled data, and **its predictions should be used for reference purposes only**. It should not be used to determine the source or origin of an image.

|         Model         |  FLOPs | Accuracy |                                                   Confusion Matrix                                                   | Description                                                         |
|:---------------------:|:------:|:--------:|:--------------------------------------------------------------------------------------------------------------------:|---------------------------------------------------------------------|
| caformer_s36_plus_sce | 22.10G |  97.54%  | [Confusion Matrix](https://huggingface.co/deepghs/anime_ai_check/blob/main/caformer_s36_plus_sce/plot_confusion.png) | Model: caformer_s36.sail_in22k_ft_in1k_384 pratrained from timm     |
|    mobilenetv3_sce    |  0.63G |  93.20%  |    [Confusion Matrix](https://huggingface.co/deepghs/anime_ai_check/blob/main/mobilenetv3_sce/plot_confusion.png)    | Model: mobilenetv3_large_100 from timm                              |
|  mobilenetv3_sce_dist |  0.63G |  95.65%  |  [Confusion Matrix](https://huggingface.co/deepghs/anime_ai_check/blob/main/mobilenetv3_sce_dist/plot_confusion.png) | Distillated from caformer_s36_plus_sce, using mobilenetv3_large_100 |