1 code implementation • 25 Mar 2023 • Caner Ozer, Arda Guler, Aysel Turkvatan Cansever, Ilkay Oksuz
We apply a variety of techniques to measure the faithfulness of the saliency detectors, and our explainable pipeline relies on NormGrad, an algorithm which can efficiently localise image quality issues with saliency maps of the classifier.
no code implementations • 11 Aug 2022 • Caner Ozer, Arda Guler, Aysel Turkvatan Cansever, Deniz Alis, Ercan Karaarslan, Ilkay Oksuz
While we obtain a classification accuracy of 87. 1% and 95. 48% on the Object-CXR and LVOT datasets, our experimental results suggest that the use of Swin Transformer improves the Object-CXR classification performance while obtaining a comparable performance for the LVOT dataset.