|
|
Generating training data with neural networks
Ševčík, Pavel ; Kolář, Martin (referee) ; Hradiš, Michal (advisor)
The aim of this thesis was to prepare a training data set for traffic sign detection using generative neural networks. The solution uses a modified U-Net architecture and several experiments with the application of styles using AdaIN layers as in the StyleGAN model have been conducted. By extending the real GTSDB data set with the generated images, mean average precision of 80.36 % has been achieved, which yields an improvement of 19.27 % compared to the mean average precision of the detection model trained on real data only.
|
|
|
I/O Subsystem Optimalization Using SSD
Bělousov, Petr ; Chalupníček, Kamil (referee) ; Kašpárek, Tomáš (advisor)
The bachelor thesis examines IO subsystem optimization using SSD cache to speed up HDDs. I examined possible server loads and identified those that are suitable for caching. In the first part I introduce 2 caching solutions, LVM cache and B-cache with their management capabilities and 2 filesystems Ext4 and XFS. In the second part IO performance of LVM cache and B-cache with Ext4 and XFS filesystem is benchmarked and compared to an uncached HDD array.
|
|
|
Detection of Graffiti Tags in Image
Pavlica, Jan ; Hradiš, Michal (referee) ; Špaňhel, Jakub (advisor)
The thesis is focused on the possible utilization of current methods in the area of computer vision with the purpose of automatic detection of graffiti tags in the image. Graffiti tagsare the most common expression of graffiti, which serves as the author’s signature. In the thesis, state-of-the-art detection systems were tested; the most effective one is the Single Shot MultiBox Detector. The result has reached 75.7% AP.
|
|
|
Detection of Vehicles in Image and Video
Petráš, Adam ; Zemčík, Pavel (referee) ; Špaňhel, Jakub (advisor)
This bachelor thesis is focused on vehicle detection. The thesis deals with the method of vehicle detection using convolutional neural networks, their structures and models. All scripts were implemented using python programming language with Tensorflow Object Detection API interface. The first part of this thesis was devote to the structures of popular neural networks and models of detection neural networks. The next chapter deals with the most famous frameworks that are used for machine learning. Three neural network models were selected and trained on the COD20K dataset. The result of this thesis is statistics that discuss the efficiency and performance of each model on trained dataset and compare performance without displaying video on Nvidia RTX 2060, where the performace archieved by SSD MobileNet V2 network was 300FPS and Nvidia Tegra TX2 8GB, whose performace reached almost 44FPS.
|
|
|
Calculating disparity map from color stereo images
Kulíková, Barbora ; Nováček, Petr (referee) ; Klečka, Jan (advisor)
This bachelor’s thesis deals with a creation of the depth maps. The first chapter concerns with the physiology of a human space perception and the methods of displaying the 3D content which are the topics closely related to the depth map, its creation and practical usage. Subsequently, there is a chapter focused on the description of the used methods of the image processing. The fundamental theoretical chapter deals with the methods of computing the disparity and used principles. In the practical part of the thesis an application has been made with a user interface in the Matlab environment. The application enables the user to create the disparity maps through the local and global methods. The functionalism of the application and the implementation of the methods are experimentaly verified. An experiment comparing the methods and analyzing influence of the local method parameters on the quality of the depth map has been made. The last part of the thesis was to create a simple stereopicture database.
|
|
|
Detector of the Human Head in Image
Svoboda, Jakub ; Orság, Filip (referee) ; Goldmann, Tomáš (advisor)
Detection of human head is an important part of person detection and identification algorithms. This thesis is focused on the detection of human head with methods based on neural networks. The majority the of conventional detectors can identify objects within a limited range of positions, whereas models based on neural networks offer a more robust approach. In this thesis we trained the current state-of-the-art models and compared their accuracy and speed. The most accurate model proved to be RetinaNet which has reached 85.15% AP. This detector can be used to improve current available algorithms for person detection, identification and tracking.
|
|
|
Various File Systems Used on Different Storage Devices
Bortlová, Pavlína ; Krčma, Martin (referee) ; Lojda, Jakub (advisor)
The bachelor thesis deals with storing data on various storage devices, namely on the HDD, SSD, flash drives, SD cards. In the teoretical part are discussed principles of functions storage media and data storage structure through various file system. In the practical part was measured read and write speeds selected combinations of file systems (FAT, NTFS, XFS, ext4, btrfs, exFAT, JFFS, F2FS) and storage devices (HDD, SSD, flash drive, sd card).
|
|
|
Weapon Detection in an Image
Debnár, Pavol ; Drahanský, Martin (referee) ; Dvořák, Michal (advisor)
This thesis is focused on the topic of firearms detection in images. In the theoretic section, the explanation of the term firearm is covered, along with the definition of the most prevalent firearm categories. Then the concept of image noise and the ways it can hinder image detection is covered, along with ways of reducing it. Next, algorithms of image detection are introduced - first those which operate on the basis of neural nets - such as Convolutional Neural Nets and Single Shot Multibox Detection. The next section discusses classic algorithms of object detection such as HOG+SVM and SURF. After that, information on the used libraries and software is provided. The experimental part covers the designed algorithm and database. For detection, the HOG+SVM, SURF and SSD algorithms were used. All the algorithms are tested on the database and, if possible, on video. A final evaluation is provided, along with possible future development options.
|
|
|
Improving Accuracy of Detection and Recognition of Traffic Signs with GANs
Glos, Michal ; Musil, Petr (referee) ; Smrž, Pavel (advisor)
The goal of this thesis was to extend a dataset for traffic sign detection. The solution was based on generative neural networks PatchGAN and Wasserstein GAN of combined DenseNet and U-Net architecture. Those models were designed to synthesize real looking traffic signs from images of their norms. Model for object detection SSD, trained on synthetic data only, achieved mean average precision of 59.6 %, which is an improvement of 9.4 % over the model trained on the original data. SSD model trained on synthetic and original data combined achieved mean average precision of 80.1 %.
|
|
|
Detection of Graffiti Tags in Image
Molisch, Marek ; Herout, Adam (referee) ; Špaňhel, Jakub (advisor)
The goal of this work is to compare today's architecture of object detection models and use them for the purpose of graffiti tag detection. State-of-the-art models, which are compatible with the Tensorflow framework, were used. Faster R-CNN architecture was found to be the most accurate and SSD architecture to be the fastest. Experiments with graffiti tags from Athens in the STORM dasater showed, that it is better to approach graffiti tags as objects rather than writings.
|
guest ::
