|
|
Convolutional Networks for Historic Text Recognition
Vešelíny, Peter ; Kolář, Martin (referee) ; Kišš, Martin (advisor)
This thesis deals with text line recognition of historical documents. Historical texts dating back to the 17th - 19th centuries are written in fraktur typeface. The character recognition problem is solved using neural network architecture called sequence-to-sequence . This architecture is based on encoder-decoder model and contains attention mechanism. In this thesis a dataset, from texts originated from German archiv called Deutsches Textarchiv , was created. This archive contains 3 897 different German books that have available transcripts and corresponding images of pages. The created dataset was used to train and experiment with the proposed neural network. During the experiments, several convolutional models, hyperparameters and the effects of positional embedding were investigated. The final tool can recognize characters with accuracy 99,63 %. The contribution of this work is the~mentioned dataset and neural network, which can be used to recognize historical documents.
|
|
|
Deep Neural Network Pruning for Text Recognition
Petráš, Simon ; Hradiš, Michal (referee) ; Kišš, Martin (advisor)
This document is a work on pruning neural network for handwriting recognition. The aim of the work is to create a program for pruning the network. We prune two types of neural networks, namely convolutional and recurrent neural networks. During the pruning of the convolution part, various criteria of parameter selection were experimented with. The result of the work is a model that achieves 20% acceleration while increasing the network inaccuracy by only 0.4%, but also a number of other models that are faster but also acquire higher inaccuracies.
|
|
| |
|
|
Detection of objects and tracking the route of movement of traffic participants for the needs of intelligent transport nodes
Vymazal, Tomáš ; Kiac, Martin (referee) ; Burget, Radim (advisor)
The master‘s thesis is focused on the object detection. The aim of this thesis is to desine an experiment to assess the detection models YOLOv5, YOLOR, Scaled-YOLOv4 and EfficientDet and to compare their properties (detection speed, memory requirements, accuracy and certainty of detection). For this purpose a custom data set is created to investigate these parameters. The study shows that the YOLOv5 network is performd as the best solution. Deep SORT is used for object tracking which is important for the subsequent extraction of training data from video footage for object movement prediction. The added value is the design of the prediction algorithm which is based on a polynomial regression model.
|
|
|
Improving Consistency in Text Recognition Datasets
Tvarožný, Matúš ; Hradiš, Michal (referee) ; Kišš, Martin (advisor)
This work is concerned with increasing the consistency of datasets for text recognition. This paper describes the problems that cause the inconsistency and then presents solutions to eliminate it. The effect of the properties of the polygons defining the text line boundaries and hence how the modified version of the dataset, which is composed of ideal text line variants, affected the accuracy of the model is investigated. Further, the work focuses on detecting and then removing or modifying text lines whose ground truth transcription does not match the actual text they contain. Experimentation showed that removing the visual inconsistency on the training set did not have a significant effect on the trained model, but modifying the test set improved the OCR accuracy of the model by 1.1\% CER. By modifying the dataset so that it did not contain mutually inconsistent pairs of recognized text and the corresponding ground truth, the model improved by a maximum of only 0.2\% CER after re-training. The main finding of this work is, above all, the proven beneficial effect of removing inconsistencies on test suites, thanks to which it is possible to determine a more realistic error rate of the OCR model.
|
|
|
Deep Neural Network Pruning for Text Recognition
Petráš, Simon ; Hradiš, Michal (referee) ; Kišš, Martin (advisor)
This document is a work on pruning neural network for handwriting recognition. The aim of the work is to create a program for pruning the network. We prune two types of neural networks, namely convolutional and recurrent neural networks. During the pruning of the convolution part, various criteria of parameter selection were experimented with. The result of the work is a model that achieves 20% acceleration while increasing the network inaccuracy by only 0.4%, but also a number of other models that are faster but also acquire higher inaccuracies.
|
|
|
Virtual Robot Control Using EEG
Drla, Michal ; Goldmann, Tomáš (referee) ; Tinka, Jan (advisor)
This bachelor thesis aimed to create an application where is user able to control the virtual robot with an EEG signal. The thesis contains a brief introduction that explains how BCI systems which are using EEG work. This introduction not only explains the basics of EEG analysis but also explains brain biology and shows different signals which are extractable from the brain. This thesis also explains the theory of neural networks which are used to implement the analysis. In implementation are shown scripts that were used to collect data and there is also shown the design of the neural network. Results of testing are good, the neural network was making correct decisions and the user was able to control the virtual robot.
|
|
|
Visual Question Answering
Kocurek, Pavel ; Ondřej, Karel (referee) ; Fajčík, Martin (advisor)
Visual Question Answering (VQA) je systém, kde je vstupem obrázek s otázkou a výstupem je odpověď. Navzdory mnoha pokrokům ve výzkumu se VQA, na rozdíl od počítačově generovaných popisů obrázků, v praxi používá jen zřídka. Cílem této práce je zúžit mezeru mezi výzkumem a praxí. Z tohoto důvodu byla kontaktována komunita zrakově postižených a byla jim nabídnuta demonstrativní aplikace VQA a následně byla vytvořena mobilní aplikace. Byla provedena studie s 20 účastníky z komunity. Nejprve účastníci zkoušeli demonstrativní aplikaci po dobu dvou týdnů a následně byli požádáni o vyplnění dotazníku. 80 % respondentů hodnotilo přesnost aplikace VQA jako dostatečnou nebo lepší a většina z nich by ocenila, kdyby jejich aplikace pro generování popisů podporovala také VQA. Po tomto zjištění práce porovná získané znalosti z VQA se znalostmi z popisů v různých scénářích. Byla vytvořena datová sada 111 obrázků různorodých scén s ručně anotovanými popisky. Experiment porovnávající získané znalosti ukázal úspěšnost 69,9 % pro VQA a 46,2 % pro popisy obrázků. V dalším experimentu v 70,9 % případů účastníci vybrali správný popis za pomocí VQA. Výsledky naznačují, že pomocí VQA je možné zjistit více znalostí o detailech obrázků než je to v případě generovaných popisů.
|
|
|
Radio Modulation Recognition Networks
Pijáčková, Kristýna ; Maršálek, Roman (referee) ; Götthans, Tomáš (advisor)
Bakalářská práce se zabývá klasifikací rádiových modulací pomocí metod hloubkového učení. V práci jsou navrženy čtyři architektury, kde tři z nich jsou tvořeny pomocí konvolučních a rekurentních neuronových sítí a čtvrtá využívá architekturu transformátorů. Při návrhu architektur byl brán v potaz výsledný počet parametrů jednotlivých sítí, který může výrazně ovlivňovat výslednou velikost sítě. Pro účely návrhu byl využit programovací jazyk Python a knihovna Keras, která umožňuje práci s neuronovými sítěmi. Výsledky práce jsou následně zhodnoceny a porovnány s výsledky sítí navržených v článcích zabývajících se tímto tématem.
|
|
|
Convolutional Networks for Handwriting Recognition
Sladký, Jan ; Kišš, Martin (referee) ; Hradiš, Michal (advisor)
This thesis deals with handwriting recognition using convolutional neural networks. From the current methods, a network model was chosen to consist of convolutional and recurrent neural networks with the Connectist Temporal Classification. The Vertical Attention Module, which selects the relevant information in each column corresponding to the text in the figure was subsequently implemented in such a model. Then, this module was compared with other possibilities of vertical aggregation between convolutional and recurrent networks. The experiments took place on a data set containing over 80,000 lines of text from Czech letters from the 20th century. The results show that the Vertical Attention Module almost always achieves the best results on all used types of convolution networks. The resulting network achieved the best result with 8,9% of the character error rate. The contribution of this work is a neural network with a newly introduced element that can recognize lines of text.
|
guest ::
