|
|
Doplnění chybějící části obrazu pomocí hlubokého učení
Zobaník, Radek ; Kubík, Tibor (oponent) ; Šilling, Petr (vedoucí práce)
V této práci vznikla aplikace pro testování a porovnávání metod pro doplnění chybějící části obrazu za využití hlubokého učení a byly natrénovány dvě metody, pconv s konvoluční architekturou, respektive AOT-GAN s GAN architekturou. Práce popisuje návrh výsledné aplikace, její funkcionalitu a důležité body implementace. Byla zvolena datová sada, na které byly vybrané modely optimálně natrénovány. Proběhly experimenty na AOT-GAN modelu, kdy se zkoumal vliv počtu AOT bloků v generátoru na výsledný doplněný obraz. Všechny experimenty byly kvalitativně a kvantitativně porovnány. Výsledky ukázaly úctyhodné výsledky při práci s přírodní scenérií.
|
|
|
Rozpoznávání nemocí rostlin pomocí umělé inteligence
Juliš, Adam ; Kubík, Tibor (oponent) ; Bažout, David (vedoucí práce)
Cílem práce bylo prozkoumat možnosti detekce nemocí rostlin při nedostatku trénovacích dat. Zkoumala se možnost extrakce vzoru jednotlivých onemocnění a aplikace těchto vzorů na neznámé rostliny. Ještě v teoretické části práce byl tento přístup označen za chybný. Dále se analyzovaly a srovnávaly datové sady se snímky rostlinných patogenů. Vytvořil se generátor augmentovaných snímků a několik modelů nad menší datovou sadou ověřující existující přístupy.
|
|
|
Deep Neural Networks for Landmark Detection in Images
Stolárik, Samuel ; Španěl, Michal (oponent) ; Kubík, Tibor (vedoucí práce)
Landmark detection in medical images is essential for the diagnosis of various medical conditions. However, it remains laborious, time-consuming, and therefore costly. Deep neural networks have proved themselves to be an applicable solution for the creation of fully automatic landmark detection systems. Automatic landmark detection in the medical domain faces the obstacle of insufficient training data; because of this, using the most efficient representation of the data is important. In this work, three different approaches to the representation of data during the training of deep neural networks were tested. The first two use the U-Net model for heatmap regression and differ in the way how the heatmaps are created. One uses 2D Gaussian heatmaps, which, however, ignore the underlying anatomical information in the input image, and the second one addresses this problem by using novel contour-hugging heatmaps. The third approach uses a graph neural network to process the hierarchical graph representation of the input image. Experiments demonstrated that the combination of the U-Net and the novel contour-hugging heatmaps could detect landmarks in cephalometric images with a mean error of 1.8 mm. This represents a 3.34% improvement over the 2D Gaussian heatmaps. Additionally, the method predicted 70.21% of landmarks with an error of less than 2 mm.
|
|
|
Generative Models for 3D Shape Completion
Zdravecký, Peter ; Španěl, Michal (oponent) ; Kubík, Tibor (vedoucí práce)
In many real-world scenarios, scanned 3D models contain missing parts due to occlusion, scanning errors, or the incomplete nature of the data itself. The goal of this work is to create an automated process for 3D shape completion using a supervised deep learning-based method. The proposed solution is based on the prior work of DiffComplete, which uses a diffusion-based model operating over distance field representation and handles the task as a generative problem. The results showed a high capability of this model with an 81.6 IoU metric on the custom-prepared test set of furniture objects. The model also demonstrates strong generalization capabilities on shapes that are out of the training distribution (average 70.9 IoU metric). Apart from more detailed data-centric experiments, this work further extends current state-of-the-art in two ways. Firstly, it addresses the most crucial shortcoming, expensive computation, by processing the input in a low-resolution domain. Secondly, it utilizes user input (Region of Interest), which gives the user more control over generation in ambiguous scenarios.
|
|
|
Deep Neural Networks for Landmark Detection on 3D Models
Kubík, Tibor ; Kodym, Oldřich (oponent) ; Španěl, Michal (vedoucí práce)
Landmark detection is a frequent step during medical data analysis. More and more often, these data are represented in the form of 3D models. An example is a 3D intraoral scan of dentition. Deep neural networks are an appropriate way of detecting landmarks in images. In terms of 3D data, the processing comes with high memory requirements and computational time, which does not meet the needs of medical applications. In this work, I propose a method that eliminates this problem and detects landmarks on the surface of polygonal models of jaws. Different architectures of neural networks, all of which are based on the U-Net architecture, are used in this work. The multi-view approach transforms the task into a 2D domain, where the suggested networks detect landmarks by heatmap regression from several viewpoints. Using a consensus method, final estimates from multiple views are predicted in 3D space. This work introduces experiments with two consensus methods: a centroid of predictions and a geometric approach based on the RANSAC algorithm and least-squares fit. Experiments have shown that a combination of Attention U-Net, 100 viewpoints, and RANSAC consensus method, is able to detect landmarks with an error of 1.20 +- 1.81 mm, while 94.01% of landmarks is predicted with an error of less than 2 mm.
|
|
|
Deep Neural Networks for Landmark Detection on 3D Models
Kubík, Tibor ; Kodym, Oldřich (oponent) ; Španěl, Michal (vedoucí práce)
Landmark detection is a frequent step during medical data analysis. More and more often, these data are represented in the form of 3D models. An example is a 3D intraoral scan of dentition. Deep neural networks are an appropriate way of detecting landmarks in images. In terms of 3D data, the processing comes with high memory requirements and computational time, which does not meet the needs of medical applications. In this work, I propose a method that eliminates this problem and detects landmarks on the surface of polygonal models of jaws. Different architectures of neural networks, all of which are based on the U-Net architecture, are used in this work. The multi-view approach transforms the task into a 2D domain, where the suggested networks detect landmarks by heatmap regression from several viewpoints. Using a consensus method, final estimates from multiple views are predicted in 3D space. This work introduces experiments with two consensus methods: a centroid of predictions and a geometric approach based on the RANSAC algorithm and least-squares fit. Experiments have shown that a combination of Attention U-Net, 100 viewpoints, and RANSAC consensus method, is able to detect landmarks with an error of 1.20 +- 1.81 mm, while 94.01% of landmarks is predicted with an error of less than 2 mm.
|
host ::
RSS.