Deepfake-Quality-Classifier-SigLIP2 is an image classification vision-language encoder model fine-tuned from google/siglip2-base-patch16-224 for a single-label classification task. It is designed to assess the quality of deepfake images using the SiglipForImageClassification architecture.

python

Classification Report:
                       precision    recall  f1-score   support

    Issue In Deepfake     0.8350    0.7870    0.8103      3750
High Quality Deepfake     0.8025    0.8500    0.8257      3750

             accuracy                         0.8185      7500
            macro avg     0.8188    0.8185    0.8180      7500
         weighted avg     0.8188    0.8185    0.8180      7500

The model categorizes images into two classes:

Class 0: "Issue In Deepfake" – indicating that the deepfake image has noticeable flaws or inconsistencies.
Class 1: "High Quality Deepfake" – indicating that the deepfake image is of high quality and appears more realistic.

Run with Transformers🤗

!pip install -q transformers torch pillow gradio

python

import gradio as gr
from transformers import AutoImageProcessor
from transformers import SiglipForImageClassification
from transformers.image_utils import load_image
from PIL import Image
import torch

# Load model and processor
model_name = "prithivMLmods/Deepfake-Quality-Classifier-SigLIP2"
model = SiglipForImageClassification.from_pretrained(model_name)
processor = AutoImageProcessor.from_pretrained(model_name)

def deepfake_detection(image):
    """Predicts deepfake probability scores for an image."""
    image = Image.fromarray(image).convert("RGB")
    inputs = processor(images=image, return_tensors="pt")
    
    with torch.no_grad():
        outputs = model(**inputs)
        logits = outputs.logits
        probs = torch.nn.functional.softmax(logits, dim=1).squeeze().tolist()
    
    labels = {0: "Issue In Deepfake", 1: "High Quality Deepfake"}
    predictions = {labels[i]: round(probs[i], 3) for i in range(len(probs))}
    
    return predictions

# Create Gradio interface
iface = gr.Interface(
    fn=deepfake_detection,
    inputs=gr.Image(type="numpy"),
    outputs=gr.Label(label="Prediction Scores"),
    title="Deepfake Quality Detection",
    description="Upload an image to check its deepfake probability scores."
)

# Launch the app
if __name__ == "__main__":
    iface.launch()

Intended Use:

The Deepfake-Quality-Classifier-SigLIP2 model is designed to evaluate the quality of deepfake images. It helps distinguish between high-quality deepfakes and those with noticeable issues. Potential use cases include:

Deepfake Quality Assessment: Identifying whether a generated deepfake meets high-quality standards or contains artifacts and inconsistencies.
Content Moderation: Assisting in filtering low-quality deepfake images in digital media platforms.
Forensic Analysis: Supporting researchers and analysts in assessing the credibility of synthetic images.
Deepfake Model Benchmarking: Helping developers compare and improve deepfake generation models.