Fino ugađanje konvolucijskih modela za lokalizaciju objekata

master's thesis

• Title (english): Fine-tuning convolutional models for object localization
• Author: Vjeran Hanžek
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Zoran Kalafatić (član povjerenstva)
• Committee member: Tomislav Hrkać (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2020-07-14, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U okviru ovog rada razmatrane su metode za detekciju objekata zasnovane na predlaganju
područja. Na početku rada ukratko se prolazi kroz osnovne koncepte korištene
u konvolucijskih mrežama. Zatim su detaljnije opisane četiri metode za detekciju
objekta zasnovanih na predlaganju područja - R-CNN modeli. Svaki od modela
dao je bolje rezultate od svojih prethodnika u pogledu brzine i preciznosti. Nadalje,
objašnjene su osnovne mjere korištene kod evaluacije modela za detekciju objekata.
Provedeni su eksperimenti nad vlastitim skupom podataka koji koriste Faster i Mask
R-CNN kao bazne modele. Za skup podataka korištene su sličice iz snimke nogometne
utakmice gdje su objekti igrači obje momčadi i sudac. Za kraj, modeli su evaluirani
osnovnim mjerama uspješnosti, prikazana su poboljšanja i krajnji rezultati detekcije.

Abstract (english)

This paper deals with object detection methods based on region proposal. At the
start of the paper, we briefly go through the basic concepts used in convolutional
networks. Then we describe four object detection methods in more detail known as
R-CNN family. Each of the models gave better results than its predecessors in terms
of speed and precision. Furthermore, we explain basic metrics used in object detection
models. Experiments were performed using Faster and Mask R-CNN as base models
over our own dataset. Extracted frames from a football match footage are used as dataset
in which objects for detection are players of both teams and the referee. Finally, we
evaluate models using basic metrics and present its improvements and final detection
results.

• Keywords: duboko učenje; detekcija objekata; konvolucijske neuronske mreže; Faster R-CNN; Mask R-CNN
• Keywords (english): deep learning; object detection; convolutional neural networks; Faster R-CNN; Mask R-CNN
• Language: croatian
• URN:NBN: urn:nbn:hr:168:758497
• 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 10:02:53