|
|
|
--- |
|
tags: |
|
- autotrain |
|
- image-classification |
|
widget: |
|
- src: https://huggingface.co/datasets/mishig/sample_images/resolve/main/tiger.jpg |
|
example_title: Tiger |
|
- src: https://huggingface.co/datasets/mishig/sample_images/resolve/main/teapot.jpg |
|
example_title: Teapot |
|
- src: https://huggingface.co/datasets/mishig/sample_images/resolve/main/palace.jpg |
|
example_title: Palace |
|
datasets: |
|
- Colby/autotrain-data-sdxl-detection |
|
--- |
|
|
|
# SDXL Detector |
|
|
|
This model was created by fine-tuning the [umm-maybe AI art detector](https://huggingface.co/umm-maybe/AI-image-detector) on a dataset of Wikimedia-SDXL image pairs, where the SDXL image is generated using a prompt based upon a BLIP-generated caption describing the Wikimedia image. |
|
|
|
This model demonstrates greatly improved performance over the umm-maybe detector on images generated by more recent diffusion models as well as non-artistic imagery (given the broader range of subjects depicted in the random sample drawn from Wikimedia). |
|
|
|
However, its performance may be lower for images generated using models other than SDXL. In particular, this model underperforms the original detector for images generated using older models (such as VQGAN+CLIP). |
|
|
|
# Model Trained Using AutoTrain |
|
|
|
- Problem type: Image Classification |
|
|
|
## Validation Metrics |
|
loss: 0.08717025071382523 |
|
|
|
f1: 0.9732620320855615 |
|
|
|
precision: 0.994535519125683 |
|
|
|
recall: 0.9528795811518325 |
|
|
|
auc: 0.9980461893059392 |
|
|
|
accuracy: 0.9812734082397003 |
|
|
