Improving classification performance of endoscopic images with generative data augmentation

Çağlar, Ümit Mert
The performance of a supervised deep learning model is highly dependent on the quality and variety of the images in the training dataset. In some applications, it may be impossible to obtain more images. Data augmentation methods have been proven to be successful in increasing the performance of deep learning models with limited data. Recent improvements on Generative Adversarial Networks (GAN) algorithms and structures resulted in improved image quality and diversity and made GAN training possible with limited data. The process of endoscopic imaging is essential for diseases with symptoms occurring inside the body. Medical experts use gastrointestinal endoscopic imaging to assess their patients and treat them. Ulcerative Colitis (UC) is a gastrointestinal disease where the assessment of a patient's health is done by Mayo scoring, where experts evaluate the severity of the disease symptoms. The classification of endoscopic images according to Mayo classes with deep-learning-based approaches has been studied and proven to be feasible. This thesis proposes adopting a GAN-based synthetic image generation process to increase the number of images in the dataset used by deep-learning-based methods. The results show that the classification performance of deep-learning-based approaches can be improved by 2.7% with the help of synthetic images generated by generative adversarial networks.


Improved Knowledge Distillation with Dynamic Network Pruning
Şener, Eren; Akbaş, Emre (2022-9-30)
Deploying convolutional neural networks to mobile or embedded devices is often prohibited by limited memory and computational resources. This is particularly problematic for the most successful networks, which tend to be very large and require long inference times. Many alternative approaches have been developed for compressing neural networks based on pruning, regularization, quantization or distillation. In this paper, we propose the “Knowledge Distillation with Dynamic Pruning” (KDDP), which trains a dyn...
Improved Image Generation in Normalizing Flows through a Multi-Scale Architecture and Variational Training
Sayın, Deniz; Cinbiş, Ramazan Gökberk; Department of Computer Engineering (2022-8-31)
Generative models have been shown to be able to produce very high fidelity samples in natural image generation tasks in recent years, especially using generative adverserial network and denoising diffusion model based approaches. Normalizing flow models are another class of generative models, which are based on learning invertible mappings between the latent space and the image space. Normalizing flow models possess desirable features such as the ability to perform exact density estimation and simple maximu...
Improvement of Hyperspectral Classification Accuracy with Limited Training Data Using Meanshift Segmentation
Özdemir, Okan Bilge; Çetin, Yasemin (2014-04-25)
In this study, the performance of hyperspectral classification algorithms with limited training data investigated. Support Vector Machines (SVM) with Gaussian kernel is used. Principle Component Analysis (PCA) is employed for preprocessing and meanshift segmentation is used to incorporate spatial information with spectral information to observe the effect spatial information. Pattern search algorithm is used to optimize meanshift segmentation parameters. The performance of the algorithm is demonstrated on h...
A method for quadruplet sample selection in deep feature learning Derin Öznitelik Öǧrenme için Dördüz Örnek Seçme Yöntemi
Karaman, Kaan; Gundogdu, Erhan; Koc, Aykut; Alatan, Abdullah Aydın (2018-07-05)
Recently, the deep learning based feature learning methodologies have been developed to recognize the objects in fine-grained detail. In order to increase the discriminativeness and robustness of the utilized features, this paper proposes a sample selection methodology for the quadruplet based feature learning. The feature space is manipulated by using the hierarchical structure of the training set. In the training process, the quadruplets are selected by considering the distances between the samples in the...
Key protected classification for collaborative learning
Sariyildiz, Mert Bulent; Cinbiş, Ramazan Gökberk; Ayday, Erman (Elsevier BV, 2020-08-01)
© 2020Large-scale datasets play a fundamental role in training deep learning models. However, dataset collection is difficult in domains that involve sensitive information. Collaborative learning techniques provide a privacy-preserving solution, by enabling training over a number of private datasets that are not shared by their owners. However, recently, it has been shown that the existing collaborative learning frameworks are vulnerable to an active adversary that runs a generative adversarial network (GAN...
Citation Formats
Ü. M. Çağlar, “Improving classification performance of endoscopic images with generative data augmentation,” M.S. - Master of Science, Middle East Technical University, 2022.