Employment of cycle-spinning in deep learning

Download

Ulku_UZUN_Ph_D.pdf

Date

2022-11-29

Author

Uzun, Ülkü

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

175
views

179
downloads

Cycle-spinning (CS) method has been used in wavelet domain processes such as signal denoising, and image enhancement with great success. In this thesis, CS is adapted to be used in deep-learning algorithms, particularly it is integrated into GANbased raindrop removal and CNN based image classification, and object detection models. Experiments on commonly-used architectures, such as AlexNet, DenseNet, ResNet, YOLOv5, EffcientDet, CenterNet, and TensorFlow Object Detection (TFOD) show that the application of the CS method produces favorable results with higher perceptual quality in terms of full-reference metrics for raindrop removal, increased accuracy in classification and better object detection performance. It is shown that the proposed method reduces the signal degradation caused by aliased components when it is employed before down-sampling and it can increase the performance, without introducing any extra learnable parameters. Another advantage of the CS method is that it inherently provides a smoothing-out effect and, as such, it has potential uses in denoising.

Subject Keywords

Cycle-spinning, Generative Adversarial Network (GAN), Classification, Object detection, Shift-invariant, Aliasing

URI

https://hdl.handle.net/11511/101141

Collections

Graduate School of Informatics, Thesis

Suggestions

OpenMETU
Core

Utilization of navigational queries for result presentation and caching in search engines Ozcan, Rifat; Altıngövde, İsmail Sengör; Ulusoy, Özgür (2008-12-01) We propose result page models with varying granularities for navigational queries and show that this approach provides a better utilization of cache space and reduces bandwidth requirements.
Improving classification performance of endoscopic images with generative data augmentation Çağlar, Ümit Mert; Temizel, Alptekin; Department of Modeling and Simulation (2022-2-8) 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 lim...
Evaluation of potential novel variations and their interactions related to bipolar disorders: Analysis of genome-wide association study data AÇIKEL, Cengizhan; Aydın Son, Yeşim; ÇELİK, Cemil; GÜL, Husamettin (2016-11-24) Background: Multifactor dimensionality reduction (MDR) is a nonparametric approach that can be used to detect relevant interactions between single-nucleotide polymorphisms (SNPs). The aim of this study was to build the best genomic model based on SNP associations and to identify candidate polymorphisms that are the underlying molecular basis of the bipolar disorders. Methods: This study was performed on Whole-Genome Association Study of Bipolar Disorder (dbGaP [database of Genotypes and Phenotypes] study ac...
Paired 3D Model Generation with Conditional Generative Adversarial Networks Öngün, Cihan; Temizel, Alptekin (2018-09-14) Generative Adversarial Networks (GANs) are shown to be successful at generating new and realistic samples including 3D object models. Conditional GAN, a variant of GANs, allows generating samples in given conditions. However, objects generated for each condition are different and it does not allow generation of the same object in different conditions. In this paper, we first adapt conditional GAN, which is originally designed for 2D image generation, to the problem of generating 3D models in different rotat...
Density-Aware Probabilistic Clustering in Ad Hoc Networks Ergenç, Doǧanalp; Eksert, Levent; Onur, Ertan (2018-06-07) Clustering makes an ad hoc network scalable forming easy-to-manage local groups. However, it brings an extra control overhead to create and maintain clustered network topology. In this paper, we propose Probabilistic Clustering Algorithm that is a simple and efficient clustering algorithm with minimal overhead. In this algorithm, cluster heads are determined probabilistically in a distributed fashion. An analytic model is introduced for nodes to compute the probability of declaring themselves as cluster hea...

Citation Formats

Ü. Uzun, “Employment of cycle-spinning in deep learning,” Ph.D. - Doctoral Program, Middle East Technical University, 2022.