Iskanje po katalogu digitalne knjižnice

Opis: Diplomsko delo opisuje implementacijo upodabljanja digitalnega modela reliefa s štiriškim drevesom, trikotniško mrežo, z Blinn-Phongovo osvetlitvijo in sivinsko barvo. Cilj je pohitriti prikaz digitalnega modela reliefa s štiriškim drevesom. Ugotovili smo, da je prikaz s štiriškim drevesom hitrejši od prikaza celotnega modela reliefa, vendar ima večjo pomnilniško zahtevnost. Hitrost izrisa je odvisna tudi od števila oglišč, kar smo ugotovili z uporabo treh digitalnih modelov reliefa z različnim številom oglišč.
Ključne besede: štiriško drevo, digitalni model reliefa, trikotniška mreža, Blinn-Phongov osvetlitveni model, OpenGL
Objavljeno v DKUM: 08.05.2025; Ogledov: 0; Prenosov: 3
Celotno besedilo (1,61 MB)

Opis: Moderno globoko učenje pogosto vključuje nevronske mreže, ki imajo nepotrebno veliko število parametrov oziroma povezav. Posledica tega je višja časovna in prostorska zahtevnost pri delu z mrežami. Možna rešitev problema je hipoteza loterijskih srečk, ki pravi, da v množici povezav vsake naprej povezane nevronske mreže obstaja podmnožica, ki je manjša in ohranja enako in v nekaterih primerih tudi večjo uspešnost. Imenujemo jo zmagovita srečka. V magistrskem delu je predstavljena in dokazana hipoteza loterijske srečke ter implementiran lasten algoritem za iterativno rezanje parametrov, ki je bil testiran in analiziran na različnih arhitekturah, podatkovnih zbirkah in hiperparametrih.
Ključne besede: strojno učenje, redka nevronska mreža, hipoteza loterijskih srečk, rezanje
Objavljeno v DKUM: 04.03.2025; Ogledov: 0; Prenosov: 31
Celotno besedilo (2,88 MB)

Opis: After a boom that coincided with the advent of the internet, digital cameras, digital video and audio storage and playback devices, the research on data compression has rested on its laurels for a quarter of a century. Domain-dependent lossy algorithms of the time, such as JPEG, AVC, MP3 and others, achieved remarkable compression ratios and encoding and decoding speeds with acceptable data quality, which has kept them in common use to this day. However, recent computing paradigms such as cloud computing, edge computing, the Internet of Things (IoT), and digital preservation have gradually posed new challenges, and, as a consequence, development trends in data compression are focusing on concepts that were not previously in the spotlight. In this article, we try to critically evaluate the most prominent of these trends and to explore their parallels, complementarities, and differences. Digital data restoration mimics the human ability to omit memorising information that is satisfactorily retrievable from the context. Feature-based data compression introduces a two-level data representation with higher-level semantic features and with residuals that correct the feature-restored (predicted) data. The integration of the advantages of individual domain-specific data compression methods into a general approach is also challenging. To the best of our knowledge, a method that addresses all these trends does not exist yet. Our methodology, COMPROMISE, has been developed exactly to make as many solutions to these challenges as possible inter-operable. It incorporates features and digital restoration. Furthermore, it is largely domain-independent (general), asymmetric, and universal. The latter refers to the ability to compress data in a common framework in a lossy, lossless, and near-lossless mode. COMPROMISE may also be considered an umbrella that links many existing domain-dependent and independent methods, supports hybrid lossless–lossy techniques, and encourages the development of new data compression algorithms
Ključne besede: data compression, data resoration, universal algorithm, feature, residual
Objavljeno v DKUM: 04.02.2025; Ogledov: 0; Prenosov: 11
Celotno besedilo (1,13 MB)

Opis: Point clouds have become a popular training data for many practical applications of machine learning in the fields of environmental modeling and precision agriculture. In order to reduce high space requirements and the effect of noise in the data, point clouds are often transformed to a structured representation such as a voxel grid. Storing, transmitting and consuming voxelized geometry, however, remains a challenging problem for machine learning pipelines running on devices with limited amount of on-chip memory with low access latency. A viable solution is to store the data in a compact encoded format, and perform on-the-fly decoding when it is needed for processing. Such on-demand expansion must be fast in order to avoid introducing substantial additional delay to the pipeline. This can be achieved by parallel decoding, which is particularly suitable for massively parallel architecture of GPUs on which the majority of machine learning is currently executed. In this paper, we present such method for efficient and parallelizable encoding/decoding of voxelized geometry. The method employs multi-level context-aware prediction of voxel occupancy based on the extracted binary feature prediction table, and encodes the residual grid with a pointerless sparse voxel octree (PSVO). We particularly focused on encoding the datasets of voxelized trees, obtained from both synthetic tree models and LiDAR point clouds of real trees. The method achieved 15.6% and 12.8% reduction of storage size with respect to plain PSVO on synthetic and real dataset, respectively. We also tested the method on a general set of diverse voxelized objects, where an average 11% improvement of storage space was achieved.
Ključne besede: voxel grid, feature prediction, tree models, prediction-based encoding, key voxels, residuals, sparse voxel octree
Objavljeno v DKUM: 09.01.2025; Ogledov: 0; Prenosov: 5
Celotno besedilo (20,93 MB)

Opis: This paper introduces a new method for the region segmentation of images. The approach is based on the raster-scan paradigm and builds the segments incrementally. The pixels are processed in the raster-scan order, while the construction of the segments is based on a distance metric in regard to the already segmented pixels in the neighbourhood. The segmentation procedure operates in linear time according to the total number of pixels. The proposed method, named the RSM (raster-scan segmentation method), was tested on selected images from the popular benchmark datasets MS COCO and DIV2K. The experimental results indicate that our method successfully extracts regions with similar pixel values. Furthermore, a comparison with two of the well-known segmentation methods—Watershed and DBSCAN—demonstrates that the proposed approach is superior in regard to efficiency while yielding visually similar results.
Ključne besede: segment, image analysis, distance metric, Watershed, DBSCAN
Objavljeno v DKUM: 05.12.2024; Ogledov: 0; Prenosov: 4
Povezava na datoteko