ai-detection-cnn

AI Detection CNN Model

This model detects AI-generated images using spatial features extracted from Qwen2.5-VL.

Model Description

• Architecture: 1D CNN with 3 convolutional layers
• Task: Binary classification (AI-generated vs Real images)
• Feature Extractor: Qwen2.5-VL-3B-Instruct
• Framework: PyTorch

Usage

from inference import test_image

# Test an image
result = test_image(
    "path/to/image.jpg",
    hf_repo_id="nahid112376/ai-detection-cnn",
    threshold=0.5
)

print(f"Label: {result['label']}")
print(f"Confidence: {result['confidence']*100:.2f}%")

Model Details

• Input: Spatial features from Qwen2.5-VL (shape: [num_patches, hidden_dim])
• Output: Single regression score (0-1, threshold at 0.5)
• Training: Trained on demectai dataset

Installation

pip install torch transformers huggingface_hub pillow numpy
pip install qwen-vl-utils

Files

• cnn_regression_model.pth: Trained model weights
• inference.py: Inference script
• extract_features.py: Feature extraction script

Citation

If you use this model, please cite:

@misc{ai-detection-cnn,
  author = {bartazable},
  title = {AI Detection CNN Model},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/nahid112376/ai-detection-cnn}}
}