Konvolucijski modeli za optički tok

master's thesis

• Title (english): Convolutional models for optical flow
• Author: Antonio Anđelić
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Slobodan Ribarić (član povjerenstva)
• Committee member: Vlado Glavinić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2020-07-08, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Procjenjivanje optičkog toka neriješen je problem računalnog vida s mnogim zanimljivim primjenama. U posljednje vrijeme najbolja rješenja tog problema dobivaju se dubokim konvolucijskim modelima. Ovaj rad razmatra nadzirane pristupe kod kojih svaki par slika iz skupa za učenje mora biti označena gustim poljem točnog optičkog toka.

U okviru rada, proučene su konvolucijske arhitekture za procjenu optičkog toka. Oblikovan je odgovarajući model i naučen na javno dostupnim skupovima. Mogućnost izvedbe na ugradbenom računalnom sustavu je procijenjena i opisana u sklopu rada. Hiperparametri su validirani, prikazani uz ostvarene rezultate koji su uspoređeni s rezultatima iz literature. Predložen je pravac budućeg razvoja.

Uz rad je priložen izvorni kod razvijenih postupaka uz potrebna objašnjenja i dokumentaciju.

Abstract (english)

Optical flow estimation is an unsolved problem in computer vision with many interesting applications. Lately, the best results are achieved using deep convolutional models. This paper analyses approaches using supervised learning where each pair of images from training set has correctly annotated optical flow.

In this paper, different convolutional models for optical flow estimation are described. One of those models was implemented and trained on publicly available datasets. Possibility of implementing that model on an embedded device was researched and described. Hyperparameters were validated and presented alongside the achieved results that were also compared to the results from the literature. Suggestion for future work is also described.

• Keywords: optički tok; duboko učenje; PWCNet; TensorRT; CUDA; konvolucijski modeli; duboki modeli; računalni vid
• Keywords (english): optical flow; deep learning; PWCNet; TensorRT; CUDA; convolutional models; deep models; computer vision
• Language: croatian
• URN:NBN: urn:nbn:hr:168:956901
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 11:40:42