Scripts, figures, and working notes for our team's participation in ImageCLEFmedical GAN task 2023, part of the ImageCLEF labs at the 14th CLEF Conference, 2023.

Implementation Stack: Python, PyTorch, Scikit-learn.

• Research Paper [PDF] describing the methods, rationale, and results.
• Contest Description and Resources.
• Model Pipelines.

Link to the Research Paper.

If you find our work useful in your research, don't forget to cite us!

@article{hb2023correlating,
  url = {https://ceur-ws.org/Vol-3497/paper-116.pdf},
  title={Correlating Biomedical Image Fingerprints between GAN-generated and Real Images using a ResNet Backbone with ML-based Downstream Comparators and Clustering: ImageCLEFmed GANs, 2023},
  author={Bharathi, Haricharan and Bhaskar, Anirudh and Venkataramani, Vishal and Desingu, Karthik and Kalinathan, Lekshmi},
  year={2023},
  keywords={Generative Adversarial Network, Support Vector Machines, Heirarchical Clustering, Machine Learning, Deep Learning, ResNet, Convolutional Neural Networks, Few shot learning, Relational model
},
  journal={Conference and Labs of the Evaluation Forum},
  publisher={Conference and Labs of the Evaluation Forum},
  ISSN={1613-0073},  
  copyright = {Creative Commons Attribution 4.0 International}
}

• A relation neural network based on few-shot learning to capture the underlying similarities between real and artificial images.
• The network learns a custom difference function to effectively compare images for artificialness arising from GAN-based generation.
• For comparison, it uses hierarchical clustering to evaluate the quality of image feature separation between real and artificial images.
• The proposed relation network achieves a 61.4% F1-score in distinguishing real and artificial medical images on a blinded contest test set.

See the paper and contest page for more information.