|
|
--- |
|
|
license: apache-2.0 |
|
|
metrics: |
|
|
- accuracy |
|
|
- f1 |
|
|
base_model: |
|
|
- google/vit-base-patch16-224-in21k |
|
|
pipeline_tag: image-classification |
|
|
library_name: transformers |
|
|
--- |
|
|
**Note to users who want to use this model in production** |
|
|
|
|
|
Beware that this model is trained on a dataset collected **about 1 year ago**. |
|
|
Since then, there is a remarkable progress in generating deepfake images with common AI tools, resulting in a significant concept drift. |
|
|
To mitigate that, I urge you to retrain the model using the latest available labeled data. |
|
|
As a quick-fix approach, simple reducing the threshold (say from default 0.5 to 0.1 or even 0.01) of labelling image as a fake may suffice. |
|
|
However, you will do that at your own risk, and retraining the model is the better way of handling the concept drift. |
|
|
|
|
|
======= |
|
|
|
|
|
Predicts with about 98% accuracy whether an attached image is AI-generated. |
|
|
|
|
|
See https://www.kaggle.com/code/dima806/ai-vs-human-generated-images-prediction-vit for details. |
|
|
|
|
|
 |
|
|
|
|
|
``` |
|
|
Classification report: |
|
|
|
|
|
precision recall f1-score support |
|
|
|
|
|
human 0.9655 0.9930 0.9790 3998 |
|
|
AI-generated 0.9928 0.9645 0.9784 3997 |
|
|
|
|
|
accuracy 0.9787 7995 |
|
|
macro avg 0.9791 0.9787 0.9787 7995 |
|
|
weighted avg 0.9791 0.9787 0.9787 7995 |
|
|
``` |
