|
|
Counting Crates in Images
Mičulek, Petr ; Špaňhel, Jakub (referee) ; Herout, Adam (advisor)
V této práci se zabývám tématem počítání beden v obrazových datech pomocí technik hlubokého učení. V práci jsem navrhl řešení pro počítání beden, které představuji na fotkách krabiček sirek. Ačkoli původní řešení počítalo s využitím datové sady beden ze skladu pivovaru, sada nakonec nebyla dodána a na doporučení vedoucího práce byly pro řešení vybrány bloky krabiček sirek. Implementované řešení využívá plně konvoluční neuronovou síť založenou na klasifikaci, umožňující výstup ve vysokém rozlišení. Tato síť je trénována na výřezech fotek z datové sady, díky čemuž je řešení rychlé a síť je vhodná i pro použití na menších datových sadách. Síť detekuje ve fotkách klíčové body krabiček sirek, které jsou následně zpracovány algoritmem pro odhad klíčových bodů z predikce sítě a výpočet finálního počtu beden. Na validačním datasetu dosahuje řešení následujících výsledků: ve 12,5 % případů predikce selže a ve zbylých případech má průměrnou absolutní odchylku (MAE) 11,14. Pomocí rozsáhlých experimentů bylo řešení vyhodnoceno a výsledky potvrzují, že tento přístup může být použit pro počítání objektů.
|
|
|
Automatic Industrial Quality Control from Image
Kruták, Martin ; Hradiš, Michal (referee) ; Zemčík, Pavel (advisor)
The goal of this thesis is to create overall, automatic and non-contact quality control of a pellet. The issue is divided into two separate parts. The first part deals with precise dimensional measuring of pellet - its length and head diameter so that it is precise and reasonably fast. Precise measuring is achieved with help of algorithms which achieve the sub-pixel precision by polynomial approximation of the edges extracted from the image gradients. The second part deals with the defects of a pellet. Detecting defects like longitudinal furrows or skirt cuts is achieved with convolutional neural networks. The measurement modules work with the resulting precision up to 0.025 mm in case of length measuring and up to 0.01 mm in case of head diameter measuring. In case of defect detections, neural network shows very high classification success rate. The contribution of this thesis is a presentation of innovative approaches in automatic quality control of pellets with use of neural networks and a demonstration of its usage in real manufacturing process.
|
|
|
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.
|
|
|
Non-contact measurement of the dimensions of determination scales
Šemora, Petr ; Matoušek, Radomil (referee) ; Škrabánek, Pavel (advisor)
This thesis deals with non-contact measuring the dimensions of the sand lizard anal plate. First the thesis briefly summarizes the techniques used to measure object dimensions and the techniques used for image segmentation. Subsequently, the thesis provides a basic overview of neural networks and convolutional neural networks. The practical part describes a system for measuring the dimensions of the sand lizard anal plate. The proposed algorithms are implemented in a graphical user interface enabling automatic and manual measurements.
|
|
|
Comic Images Super-Resolution Using Deep Learning
Zdravecký, Peter ; Juránek, Roman (referee) ; Španěl, Michal (advisor)
Táto práca demonštruje metódu super rozlíšenia na zlepšenie kvality komiksových obrázkov pomocou hlbokého učenia. Náročnou časťou tejto úlohy bolo súčasne zachovať kvalitu textových a kreslených častí, bez výraznej deformácie ktorejkoľvek časti z nich. Na dosiahnutie uspokojivých výsledkov boli skúmané dve hlboké neurónové siete. Sieť U-Net a modifikácia s názvom Robustný U-Net (RUNet). Zvolené stratové funkcie na trénovanie týchto sietí boli stredná kvadratická chyba a perceptuálna strata. Práca obsahuje experimenty na týchto sieťach v kombinácii s každou stratovou funkciou. Ďalšie experimenty sa zamerali na vplyv počtu použitých blokov zo stratovej siete VGG16 na funkciu perceptuálnej straty. Experimenty ukázali, že sieť RUNet využívajúca perceptuálnu stratu s tromi extrahovanými blokmi dosiahla najlepšie výsledky.
|
|
|
Pedestrian Attribute Analysis
Studená, Zuzana ; Špaňhel, Jakub (referee) ; Hradiš, Michal (advisor)
This work deals with obtaining pedestrian information, which are captured by static, external cameras located in public, outdoor or indoor spaces. The aim is to obtain as much information as possible. Information such as gender, age and type of clothing, accessories, fashion style, or overall personality are obtained using using convolutional neural networks. One part of the work consists of creating a new dataset that captures pedestrians and includes information about the person's sex, age, and fashion style. Another part of the thesis is the design and implementation of convolutional neural networks, which classify the mentioned pedestrian characteristics. Neural networks evaluate pedestrian input images in PETA, FashionStyle14 and BUT Pedestrian Attributes datasets. Experiments performed over the PETA and FashionStyle datasets compare my results to various convolutional neural networks described in publications. Further experiments are shown on created BUT data set of pedestrian attributes.
|
|
|
Volumetric Segmentation of Dental CT Data
Berezný, Matej ; Kodym, Oldřich (referee) ; Čadík, Martin (advisor)
The main goal of this work was to use neural networks for volumetric segmentation of dental CBCT data. As a byproducts, both new dataset including sparse and dense annotations and automatic preprocessing pipeline were produced. Additionally, the possibility of applying transfer learning and multi-phase training in order to improve segmentation results was tested. From the various tests that were carried out, conclusion can be drawn that both multi-phase training and transfer learning showed substantial improvement in dice score for both sparse and dense annotations compared to the baseline method.
|
|
|
Anti-Drone Perimeter Protection
Janík, Roman ; Dvořák, Michal (referee) ; Drahanský, Martin (advisor)
Developement of drone technology brings opportunities for many fields of human activity, but simultaneously brings security threats. A need to effectively face these threats arises. In this work is described the problematics and state-of-the-art methods for object detection in a video captured by moving camera. A system for detecting and locating a drone or a flock of drones has been proposed. Algorithm for detection is based on convolutional neural network, specifically on SSD algorithm. The convolutional neural network was trained on self-made dataset. The system was implemented using OpenCV library with possible algorithm acceleration on GPU using OpenCL. Created solution was tested on video.
|
|
|
Weather Estimation Based on Images of Clouds
Kukaň, Tomáš ; Goldmann, Tomáš (referee) ; Orság, Filip (advisor)
The main purpose of this thesis is a creation of a simple mobile application that would be able to give weather predictions based on a cloud photo through the usage of convolu- tional neural networks. I have analyzed all types of clouds and joined them with weather prediction. Then there are the results of experiments with different neural networks archi- tectures and different datasets. In the end of this thesis I have described the creation of the Android application as well as the problems I had to solve.
|
|
|
Interpretability of Neural Networks in Speech Processing
Sarvaš, Marek ; Mošner, Ladislav (referee) ; Žmolíková, Kateřina (advisor)
S rastúcou popularitou hlbokých neurónových sietí, nedostatok transparentnosti spôsobenejich funkciou čiernej skrinky, zvyšuje dopyt po ich interpretácii. Cieľom tejto práce je získať nový pohľad na hlboké neurónové siete v úlohách spracovania reči. Konkrétne klasifikácia pohlavia z AudioMNIST datasetu a klasifikácia rečníka z filter bánk VoxCeleb datasetu s použitím konvolučnej a reziduálnej neurónovej siete. Na interpretáciu týchto neurónových sietí bola použitá metóda propagácie relevancií cez vrstvy. Táto metóda vytvorí tepelnú mapu, ktorá vyznačí príznaky, ktoré prispeli ku správnej klasifikácii pozitívne a ktoré negatívne. Ako výsledky interpretácie ukazujú, klasifikácie boli založené najmä na nižších frekvenciách v reči a čase. V prípade klasifikácie pohlavia sa mi podarilo nájsť vysokú závislosť modelu na veľmi malom počte príznakov. Pomocou získaných informácií som vytvoril rozšírený trénovací set, ktorý zvýšil robustnosť modelu.
|
guest ::
