|
|
Implementing gesture recognition on ARM as an alternative to traditional device control
Gajdošík, Richard ; Zbořil, František (referee) ; Kočí, Radek (advisor)
Cieľom tejto bakalárskej práce je vývoj a implementácia systému na rozpoznávanie gest s využitím architektúry ARM, konkrétne s použitím dosky i.MX 93 a TensorFlow Lite. Projekt sa zameriava na aplikáciu neurónových sietí pre rozpoznávanie gest rúk, čím poskytuje alternatívu k tradičným metódam ovládania zariadení. Dôležitou súčasťou práce je rozsiahla analýza existujúcich riešení rozpoznávania gest, zameraná na identifikáciu ich silných stránok a možných vylepšení. Práca detailne opisuje proces navrhovania, vývoja a optimalizácie modelu na rozpoznávanie gest v reálnom čase, špeciálne prispôsobeného pre čipy ARM s dôrazom na efektivitu a výkon. Okrem toho práca aj obsahuje vytvorenie demonštračnej aplikácie, ktorá vizuálne reprezentuje rozpoznané gestá. Užívateľské testovanie je uskutočnené na hodnotenie praktickosti a užívateľského zážitku systému rozpoznávania gest, čo poskytuje cennú spätnú väzbu pre budúce vylepšenia.
|
|
|
Neural Networks for Video Quality Enhancement
Sirovatka, Matej ; Juránek, Roman (referee) ; Hradiš, Michal (advisor)
Cieľom tejto práce je vytvoriť novú metódu super rozlíšenia na zlepšenie kvality videa. Táto metóda je založená na myšlienke použitia deformovateľných konvolučných vrstiev a optického toku na zarovnanie príznakov z viacerých po sebe následujúcích snímkov videa. Táto metóda je následne použitá v neuronovej sieti založenej na U-Net architektúre na predikciu snímkov vo vysokom rozlíšení. Vyhodnotenie je prevedené na datasete obsahujúcom snímky z reálneho života a porovnané s inými metódami. Testované sú rôzne konfigurácie navrhnutej metódy a výsledky sú analyzované. Výsledky experimentov ukazujú sľubné výsledky, pričom model prekonáva bilineárnu interpoláciu a metódy založené na jednom snímku. Testované sú rôzne architektúry modulu zarovnávania príznakov spolu s celou architektúrou U-Net, pričom sa ukazuje, že použitie Vgg19 ako enkóderu dáva najlepšie výsledky.
|
|
|
Framework for event modeling a prediction in football.
Geffert, Maroš ; Beneš, Karel (referee) ; Szőke, Igor (advisor)
This thesis investigates current methods of predicting football events such as the number of goals in a match, the outcome of a match, or whether both teams will score. The models analyzed were neural network, RandomForest and XGBoost. Extensive historical data on matches and players were collected as part of the work. The main objectives were to determine whether detailed statistics significantly affect prediction, to evaluate the effectiveness of using betting odds as features, to investigate the impact of historical data on the quality of predictions, and to determine whether success can be achieved in the betting market with such models. The results showed that detailed statistics improve the accuracy of the predictions, but the use of odds as features generally degrades the predictions. The results regarding the use of historical data for predictions were inconclusive. RandomForest and neural network models achieved promising results with ROI of 32.38% and 29.04%, respectively.
|
|
|
Reversibility of Voice Change Methods
Lička, Zbyněk ; Firc, Anton (referee) ; Malinka, Kamil (advisor)
Moderní metody pro změnu hlasu dovolují i nezkušeným uživatelům vytvářet přesvědčívé nahrávky hlasu slavné osoby s pouze pár sekundami nahraného ukázkového hlasu. Existují dvě hlavní kategorie metod pro změnu hlasu: konverze hlasu a text-to-speech. Metody konverze hlasu vyžadují vstupní řeč, která má být konvertována do hlasu jiného řečníka. Moderní metody pro konverzi hlasu se často zabývají odstraněním či redukcí množství informací o původním řečníkovi v konvertovaném hlasu. Tato práce se zabývá možnostmi pro extrakci informací z konvertovaného hlasu s případnou kompletní rekonstrukcí vstupní řeči. Výsledky této práce odhalují poznatky o nestudované vlastnosti těchto metod.
|
|
|
Automatic Creation of Animated Video based on Textual Story
Kuchař, Josef ; Švec, Tomáš (referee) ; Smrž, Pavel (advisor)
The aim of this work is to link a diffusion model for generating human motion with a diffusion model for generating video. The solution uses current methods for generating video and motion. Video generation is carried out using an image generator equipped with an adapter for temporal consistency. The work introduces a method of connecting both diffusion models using the ControlNet network. The created solution allows for generating video from a simple text description, or a detailed scenario. The program was tested in a user study.
|
|
|
Optimalizace řízení podmínek v inteligentním skleníku
Vilimovský, Dan ; Beran, Vítězslav (referee) ; Bažout, David (advisor)
The work focuses on finding ways to optimize the operation of a smart greenhouse by reducing operating costs and facilitating the sustainability of an ideal environment for plant development in the greenhouse. The aim is to model the conditions in the greenhouse using a developed neural network model that is able to predict the values of environmental parameters such as temperature and humidity. This model is trained on real data obtained from long-term measurements in a test facility. Furthermore, the thesis deals with the possibility of using this neural network model to optimize the control of the greenhouse operation so that ideal development conditions for the plants grown or reduced electricity consumption used for the operation of appliances can be ensured in the future.
|
|
|
Cellular Automata Design Methods
Hranický, Jan ; Strnadel, Josef (referee) ; Bidlo, Michal (advisor)
This thesis addresses the problem of cellular automata design. A new method of Discrete neural cellular automaton, abbreviated DNCA, is proposed and implemented. The thesis thus deals with the problem of optimizing the transition function of the automaton using the Adam algorithm as well as a less traditional approach of differential evolution. The proposed model is successfully trained on more than ten behavior scenarios varying in difficulty.
|
|
|
Methods for initializing neural network weights and their effect on network learning
Prukner, Jakub ; Nemčeková, Petra (referee) ; Chmelík, Jiří (advisor)
This thesis examines the use of various methods for initialising the weights of artificial neural networks and monitoring their impact on network learning. Image classification from two databases, MNIST and CIFAR-10, is selected as the task for the network. The theoretical section provides an overview of the field of artificial neural networks, along with an analysis of different methods for initialising weights. The practical section includes a description of the experiments conducted, an explanation of the architectures and their associated hyperparameters. The individual experiments observe the effect of the selected methods and their respective configurations on the learning of different artificial neural network architectures. The results are compared for each dataset and architecture type, and the methods with which a the network achieved the best learning are selected. Furthermore, the methods with which the optimal learning of the network was achieved the fastest are selected. The results obtained are discussed.
|
|
|
Identifikace člověka podle fotografie dlaně / hřbetu ruky
Štanga, Miroslav ; Vaško, Marek (referee) ; Herout, Adam (advisor)
This work focuses on using contrastive self-supervised learning method for creating model of deep learning intended for person recognition based on hand photographs. The paper outlines fundamentals of machine learning, utilized tools and dataset. The method was developed using PyTorch library. The proposed model draws inspiration from the SimCLR architecture and its use of contrastive representation learning. The proposed approach utilizes the triplet loss function for optimization. Then the optimization process is described and impact of individual hyperparameters on the model´s accuracy is compared. The resulting model was trained on 1696 hand photos and achieves 98% accuracy on validation set. The accuracy achieved using self-supervised methods is higher than the accuracy achieved using supervised methods.
|
|
|
Radial Basis Function Neural Network
Nevoral, Leoš ; Rozman, Jaroslav (referee) ; Zbořil, František (advisor)
The thesis focuses on explaining the field of RBF neural networks, specifically RCE networks, through a demo application. The demo application primarily visualizes the network learning process, as well as the state of the neural network and different shapes of basis functions. Additionally, the utilization of EBF neural networks is explored. Conventional approaches to EBF networks are compared and tested against new design of OEBF network. Which is based on deriving the elliptical areas from euclidean distance from both focal points of ellipse. The new design shows no signs of improving the properties of these networks and rather produces results almost identical to those of the classic RCE network, which are, however, several percentage points less accurate. Finally, methods for improving this solution in future are proposed.
|
guest ::
