National Repository of Grey Literature 215 records found  1 - 10nextend  jump to record: Search took 0.00 seconds. 
Document Information Extraction
Janík, Roman ; Špaňhel, Jakub (referee) ; Hradiš, Michal (advisor)
S rozvojem digitalizace přichází potřeba analýzy historických dokumentů. Důležitou úlohou pro extrakci informací a dolování dat je rozpoznávání pojmenovaných entit. Cílem této práce je vyvinout systém pro extrakci informací z českých historických dokumentů, jako jsou noviny, kroniky a matriční knihy. Byl navržen systém pro extrakci informací, jehož vstupem jsou naskenované historické dokumenty zpracované OCR algoritmem. Systém je založen na modifikovaném modelu RoBERTa. Extrakce informací z českých historických dokumentů přináší výzvy v podobě nutnosti vhodného korpusu pro historickou Češtinu. Pro trénování systému byly použity korpusy Czech Named Entity Corpus (CNEC) a Czech Historical Named Entity Corpus (CHNEC), spolu s mým vlastním vytvořeným korpusem. Systém dosahuje úspěšnosti 88,85 F1 skóre na CNEC a 87,19 F1 skóre na CHNEC. Toto je zlepšení o 1,36 F1 u CNEC a 5,19 F1 u CHNEC a tedy nejlepší známé výsledky.
GAN Generated Data for CNN Age Estimation
Venkrbec, Tomáš ; Herout, Adam (referee) ; Hradiš, Michal (advisor)
The goal of this thesis is to implement one of the state-of-the-art methods of generative adversarial networks and to propose its extension to conditional generation. This has been used to generate photorealistic images of human faces with specified characteristics such as age and gender. For this purpose, a highly diverse dataset of over 230,000 samples was created by merging and cleaning existing annotated face datasets. All ages, genders and different ethnic groups are well represented in it. StyleGAN2 generator trained on this dataset achieved a FID of 7.14. The synthetic data ratio was then experimented with during age classifier training. For the test subset of the dataset, the addition of synthetic data achieved a reduction in the mean absolute error from 3.499 years to 3.294 years. For the independent test dataset, a reduction in mean error from 4.012 years to 3.875 years was achieved.
Deep Neural Networks for Text Recognition
Kavuliak, Daniel ; Hradiš, Michal (referee) ; Kišš, Martin (advisor)
The aim of this work is to build a model for handwritten text recognition, which will use non-autoregressive decoder. This type of decoder calculates character predictions independently of other predicted characters, which can be advantageous in terms of inference speed, but the quality of the prediction is worse. The motivation is to design a non-autoregressive decoder, which will have the task of refining the encoder's predictions. The task was solved with the help of decoders, which mask the encoder's predictions or partially suppress the information due to the use of information about unmasked symbols or using input sequence information. Subsequently, a series of experiments was performed, where the best model reached a character error rate of 8.92 %. But the assignment was not fulfilled, because the encoder itself reached 6.38 %.
Identifikace osob pomocí obrazu duhovky
Žákovic, Marek ; Hradiš, Michal (referee) ; Vaško, Marek (advisor)
The goal of this bachelor’s thesis was to create a system for person identification using iris images. The thesis describes existing methods and procedures for iris recognition. The proposed method utilizes a convolutional neural network trained to extract features, which are then used to compare whether the image belongs to the same person or not. The experiments involve training and evaluating the neural network. For the purposes of this thesis, freely available datasets were used, which were modified for specific use.
Graph Neural Networks for Document Analysis
Patrik, Nikolas ; Španěl, Michal (referee) ; Hradiš, Michal (advisor)
V tejto práci sa zameriaveme na analýzu dokumentov pomocou grafových neuronových sietí. Na začiatok si predstavíme ako tieto grafove konvolučné siete fungujú a predstavíme si koncept na základe ktorého sa dajú naimplementovať. Ďalej rozoberieme súčasné riešenia ktoré sa zaoberajú semantickým označovaním entít v skenovaných dokumentoch, čo je aj cieľom tejto práce. Následne si predstavíme navrh riešenie ktoré by malo riešiť túto problematiku spolu s ďaľším problémom na ktorý sa chceme zamariať v tejto práci a tým je výber textových entít z dokumentov pomocou aktívneho učenia. Postupne si predstavíme ako bolo toto riešenie implementované a aké nástroje sme pritom použili. Pred koncom si predstavíme dataset ktorý sme annotovali pre vyhodnotenie a tréning našeho riešenia. Na záver si predstavíme výsledky tejto práce, porovnáme vysledky s ostatnými prístupmi ktoré sa zamerievajú na podobný problém a ešte vyhodnotíme ako náš model zvládol extrakciu informácii pomocou aktívneho učenia.
Model Compression of Denoising Diffusion Probabilistic Models for Image Generation
Dobiš, Lukáš ; Kišš, Martin (referee) ; Hradiš, Michal (advisor)
Táto diplomová práca sa zameriava na optimalizáciu výpočtovej efektívnosti generatívnych difúznych modelov skrz vyhodnotenie konvenčných metód komprimácie neurónovych sieti na architektúre Denoising Diffusion Probabilistic Model (DDPM). Modelová komprimácia bola vykonaná na parametroch predtrénovanéj sieti DDPM niekoľkými kvantizačnými a prerezávacími metódami. Tieto metódy boli vyhodnotené na troch rôznych obrázkových dátových sadách. Výsledky potvrdzujú, že implementované kompresné metódy sú vhodne pre nasadenie difúznych modelov na malých zariadeniach s obmedzenými zdrojmi alebo na zníženie ich výpočetnych prevádzkových nákladov.
Znovupoužitelný 2D editor pro webové aplikace
Schneider, Martin ; Hradiš, Michal (referee) ; Herout, Adam (advisor)
The goal of this work is to implement a 2D editor as a library for web applications together with a test application. The library primarily targets segmentation neural network and remote use environments with optimizations for data transfer. First, the reader is introduced to frontend application development with a focus on 2D graphics editing and image data display. Next, an analysis of the needs for a 2D editor with applications to annotation editing and displaying results in computer vision tasks is presented. The result of the analysis is then converted into a system design, which is subsequently implemented and tested.
Reidentifikace automobilů v obraze
Ohradzanská, Karolína ; Hradiš, Michal (referee) ; Herout, Adam (advisor)
Vehicle re-identification is a helpful technic for tracking and monitoring traffic in various situations. This thesis deals with the issue of re-identification cars in the image to track vehicles using camera systems. Specifically, it focuses on the multi-camera vehicle tracking task from the international AI City Challenge competition. In this work were trained five types of convolutional networks and one transformer model. It investigated how successfully different convolutional networks worked compared to the transformer model in the re-identification task. Several experiments were performed with these networks on several datasets, while the resNeXt model achieved a success rate of up to 86.35~\% on the VeRi dataset. Participation in the AI City Challenge in 2023 required creating a dataset with people for the re-identification task.
Vehicle Re-Identification Using Vision Transformers
Jelínek, Zdeněk ; Hradiš, Michal (referee) ; Špaňhel, Jakub (advisor)
The main objective of this thesis was to investigate the potential of vision transformers in vehicle re-identification. Convolutional neural networks have so far dominated this area of computer vision. In total, two models have been tested - TransReID and CMT. TransReID is a model based purely on vision transformers and was created specifically for vehicle re-identification. The main part of the experiments with this model was devoted to the use of key points on the vehicle. With proper extraction of the regions around the key points and the use of post-processing, I achieved state-of-the-art results. The CMT model is a combination of convolutional networks and transformers that was not designed for vehicle re-identification. I modified the model and conducted extensive experiments with it to obtain the best configuration for vehicle re-identification. I evaluated the models on the standard datasets VeRi-776, VehicleID, CityFlowV2-ReID and CarsReId74k and compared with state-of-the-art models. With the CMT model, I achieved the best result of 0.860 on the mAP metric on the VeRi-776 dataset and the best result of 97.6% on the Rank5 metric on the VehicleID dataset.
Understanding of Badminton Videos
Mašláň, Vojtěch ; Hradiš, Michal (referee) ; Herout, Adam (advisor)
The aim of this thesis was to experiment with machine learning models for the understanding of badminton videos. The thesis maps the current state of using computer vision and machine learning for sport videos analysis. In the experimental part of the work, several models for badminton stroke detection were made. All the models are predicting the strokes based on player poses extracted from a pose estimation model. Developed models achieved an accuracy of 80.1 % for detecting 7 different strokes and 84.0 % for detecting 4 different strokes. Trained models were then used to create a simple web application for short badminton video analysis.

National Repository of Grey Literature : 215 records found   1 - 10nextend  jump to record:
Interested in being notified about new results for this query?
Subscribe to the RSS feed.