|
|
Assessment of Uncertainty of Neural Net Predictions in the Tasks of Classification, Detection and Segmentation
Vlasák, Jiří ; Kohút, Jan (referee) ; Herout, Adam (advisor)
This work focuses on comparing three widely used methods for improving uncertainty estimations: Deep Ensembles, Monte Carlo Dropout, and Temperature Scaling. These methods are applied to six computer vision models that are pretrained as well as trained from scratch. The models are then evaluated on computer vision datasets for classification, semantic segmentation, and object detection using a wide range of metrics. The models are also evaluated on distorted versions of these datasets to measure their performance on out-of-distribution data. These modified models achieve promising results. Ensembles outperform the other models by as high as 70 % in accuracy and 0.2 in IOU on the distorted MedSeg COVID-19 segmentation dataset while also outperforming the other models on the CIFAR-100 and FMNIST datasets.
|
|
|
Semantic segmentation of images using convolutional neural networks
Špila, Filip ; Věchet, Stanislav (referee) ; Krejsa, Jiří (advisor)
Tato práce se zabývá rešerší a implementací vybraných architektur konvolučních neuronových sítí pro segmentaci obrazu. V první části jsou shrnuty základní pojmy z teorie neuronových sítí. Tato část také představuje silné stránky konvolučních sítí v oblasti rozpoznávání obrazových dat. Teoretická část je uzavřena rešerší zaměřenou na konkrétní architekturu používanou na segmentaci scén. Implementace této architektury a jejích variant v Caffe je převzata a upravena pro konkrétní použití v praktické části práce. Nedílnou součástí tohoto procesu jsou kroky potřebné ke správnému nastavení softwarového a hardwarového prostředí. Příslušná kapitola proto poskytuje přesný návod, který ocení zejména noví uživatelé Linuxu. Pro trénování všech variant vybrané sítě je vytvořen vlastní dataset obsahující 2600 obrázků. Je také provedeno několik nastavení původní implementace, zvláště pro účely použití předtrénovaných parametrů. Trénování zahrnuje výběr hyperparametrů, jakými jsou například typ optimalizačního algoritmu a rychlost učení. Na závěr je provedeno vyhodnocení výkonu a výpočtové náročnosti všech natrénovaných sítí na testovacím datasetu.
|
|
|
Urban Element Detection Using Satellite Imagery
Oravec, Dávid ; Herout, Adam (referee) ; Zlámal, Adam (advisor)
Táto práca sa zameriava na správnu detekciu objektov v satelitných snímkach pomocou konvolučných neuronových sietí. Cieľom práce je pomocou natrénovaného modelu detekovať bazény a tenisové ihriská v satelitných snímkach z rôznych miest. Model pracuje s dátami z 10 rôznych miest. Pri vypracovaní bol využitý model neurónovej siete RetinaNet a knižnica Detectron2. Model, ktorý sa podarilo vytrénovať, dokáže detekovať objekty s priemernou presnosťou (AP50) na úrovni 63,402 %. Práca môže byť prínosom v oblasti automatizovania získavania štatistík o povrchu zeme.
|
|
|
Semantic Segmentation in Mountainous Environment
Pelikán, Jakub ; Čadík, Martin (referee) ; Brejcha, Jan (advisor)
Semantic segmentation is one of classic computer vision problems and strong tool for machine processing and understanding of the scene. In this thesis we use semantic segmentation in mountainous environment. The main motivation of this work is to use semantic segmentation for automatic location of geographic position, where the picture was taken. In this thesis we evaluated actual methods of semantic segmentation and we chose three of them that are appropriate for adapting to mountainous environment. We split the dataset with mountainous environment into validation, train and test sets to use for training of chosen semantic segmentation methods. We trained models from chosen methods on mountainous data. We let segments from the best trained models get evaluated in electronic survey by respondents and we evaluated these segments in process of camera orientation estimation. We showed that chosen methods of semantic segmentation are possible to use in mountainous environment. Our models are trained on 11, 5 or 4 mountainous classes and the best of them achieve on 4 class mean IU 57.4%. Models are usable in practise. We show it by their deployment as a part of camera orientation estimation process.
|
|
|
Computer Aided Recognization and Classification of Coat of Arms
Vídeňský, František ; Kočí, Radek (referee) ; Zbořil, František (advisor)
This master thesis describes the design and development of the system for detection and recognition of whole coat of arms as well as each heraldic parts. In the thesis are presented methods of computer vision for segmentation and detection of an object and selected methods that are the most suitable. Most of the heraldic parts are segmented using a convolution neural networks and the rest using active contours. The Histogram of the gradient method was selected for coats of arms detection in an image. For training and functionality verification is used my own data set. The resulting system can serve as an auxiliary tool used in auxiliary sciences of history.
|
|
|
Predicting Trajectories of Vehicles and Pedestrians for Driving Assistent Systems
Mudroň, Marek ; Musil, Petr (referee) ; Smrž, Pavel (advisor)
This bachelor thesis deals with representation of a traffic scene by processing monocular video sequence. I try to predict a trajectory of detected vehicles in a short time horizon, based on created representation. Current approaches use multiple expensive sensors to gather instant information of environment. In the thesis I introduce technique, which is able to extract data from an environment by image processing techniques without the need of expensive sensors. The result of this work is a system creating opportunity to reduce the sensor costs of a system for scene representation and trajectory prediction of vehicles in the scene. In addition, comparison of models trained on differently processed data is provided, as well as data about how my system approximates the most reliable prediction models.
|
|
| |
|
|
Semantic Segmentation of Pathologies in Retinal Images
Čabala, Roman ; Orság, Filip (referee) ; Kavetskyi, Andrii (advisor)
The thesis aimed to segment pathology visible in the retina images, such as exudates, hemorrhages, and microaneurysms. For that, two well known deep neural networks, named U-Net and SegFormer, were trained. To test the performance of the models, one publicly available dataset was used, named IDRiD. Obtained results were reported after analyzing different factors which affected the performance of the models U-Net and Segformer.
|
|
| |
|
|
Cell segmentation from wide-field light microscopy images using CNNs
GHAZNAVI, Ali
Image object segmentation allows localising the region of interest in the image (ROI) and separating the foreground from the background. Cell detection and segmentation are the primary and critical steps in microscopy image analysis. Analysing microscopy images allows us to extract vital information about the cells, including their morphology, size, and life cycle. On the other hand, living cell segmentation is challenging due to the complexity of these datasets. This research focused on developing Artificial Intelligence/Machine Learning methods of single- and multi-class segmentation of living cells. For this study, the Negroid cervical epithelioid carcinoma HeLa line was chosen as the oldest, immortal, and most widely used model cell line. Several time-lapse image series of living HeLa cells were captured using a high-resolved wide-field transmitted/reflected light microscope (custom-made for the Institute of Complex System, Nové Hrady, Czech Republic) to observe micro-objects and cells. Employing a telecentric objective with a high-resolution camera with a large sensor size allows us to achieve a high level of detail and sharper borders in large microscopy images. The collected time-lapse images were calibrated and denoised in the pre-processing step. The data sets collected under the transmission microscope setup were analyzed using a simple U-Net, Attention U-Net, and Residual Attention U-Net to achieve the best single-class semantic segmentation result. The data sets collected under the reflection microscope setup were analyzed using hybrid U-Net methods, including Vgg19-Unet, Inception-Unet, and ResNet34-Unet, to achieve the most precise multi-class segmentation result.
|
guest ::
