03859ntm a22006017i 4500 000728052 CZ-PrVSE 20260201122909.0 m d cr n|||||||||| 260201s2025 xr fsbm 000 0 eng d NEZPRACOVANÝ IMPORT ABA006 cze ABA006 ABA006 rda Rychnovská, Hedvika ISIS:162859 dis Grafové neuronové sítě pro klasifikaci obrazu : empirické porovnání s konvolučními sítěmi eng Graph Neural Networks for Image Classification : An Empirical Comparison with Convolutional Networks Hedvika Rychnovská 2025 ?? stran : digital, PDF soubor Vedoucí práce: Ondřej Zamazal Bakalářská práce (Bc.)—Vysoká škola ekonomická v Praze. Fakulta informatiky a statistiky, 2026 Obsahuje bibliografii Textový (vysokoškolská kvalifikační práce) Rok obhajoby 2026 Image classification is one of the core tasks in computer vision and is commonly addressed using Convolutional Neural Networks (CNNs), which are designed for grid-like data. Graph Neural Networks (GNNs), designed for graph data, offer an alternative approach by representing images as graphs and can be considered as a generalization of CNNs. The goal of this thesis is to review GNNs with a focus on their use in image classification, to select suitable datasets and architectures, and to compare CNNs and GNNs on this task. Images are converted into superpixel graphs using the SLIC algorithm and the Region Adjacency Graph (RAG) method, and the selected models include GCN and GraphSAGE for the GNNs and two custom CNN architectures. The experiments address four aspects: accuracy, trade-off between accuracy and computational cost, the effect of the number of superpixel segments, and per-class classification. The results show that CNNs achieve higher accuracy overall, while GNNs perform comparably only on simpler images but offer advantages in certain efficiency-related cases. The number of superpixel segments used in graph construction was shown to influence their performance. Overall, the thesis provides a theoretical background and an empirical comparison of CNNs and GNNs on image data across several evaluation criteria. Způsob přístupu: Internet data analytics [obor bakal. práce] bakalářské práce fd132403 czenas bachelor's theses eczenas convolutional neural networks image classification Graph Convolutional Network GraphSAGE deep learning artificial neural networks graph neural networks superpixels Zamazal, Ondřej ISIS:7282 ths Vacura, Miroslav mzk2006343245 opn Vysoká škola ekonomická v Praze. Fakulta informatiky a statistiky kn20010709399 dgg https://insis.vse.cz/zp/92828/podrobnosti VŠKP v InSIS https://insis.vse.cz/zp/92828 Hlavní práce https://insis.vse.cz/zp/92828/posudek/vedouci Hodnocení vedoucího https://insis.vse.cz/zp/92828/posudek/oponent/88824 Oponentura https://insis.vse.cz/zp/92828/priloha/33008 Přiloha k práci https://insis.vse.cz/zp/92828/priloha/33009 Přiloha k práci https://insis.vse.cz/zp/92828/priloha/33010 Přiloha k práci https://insis.vse.cz/zp/92828/priloha/33011 Přiloha k práci https://insis.vse.cz/zp/92828/podrobnosti dc:identifier NEPOSILAT VSKP vse92828 260130 92828