prikaz prve stranice dokumenta Destiliranje semantičke segmentacije
No public access
undergraduate thesis
Destiliranje semantičke segmentacije
Zagreb: University of Zagreb, Faculty of Electrical Engineering and Computing, 2024. urn:nbn:hr:168:927246
Sučić, Filip
University of Zagreb
Faculty of Electrical Engineering and Computing

Institutional repository: FER Repository

Cite this document

Sučić, F. (2024). Destiliranje semantičke segmentacije (Undergraduate thesis). Zagreb: University of Zagreb, Faculty of Electrical Engineering and Computing. Retrieved from https://urn.nsk.hr/urn:nbn:hr:168:927246

Sučić, Filip. "Destiliranje semantičke segmentacije." Undergraduate thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, 2024. https://urn.nsk.hr/urn:nbn:hr:168:927246

Sučić, Filip. "Destiliranje semantičke segmentacije." Undergraduate thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, 2024. https://urn.nsk.hr/urn:nbn:hr:168:927246

Sučić, F. (2024). 'Destiliranje semantičke segmentacije', Undergraduate thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, accessed 29 November 2024, https://urn.nsk.hr/urn:nbn:hr:168:927246

Sučić F. Destiliranje semantičke segmentacije [Undergraduate thesis]. Zagreb: University of Zagreb, Faculty of Electrical Engineering and Computing; 2024 [cited 2024 November 29] Available at: https://urn.nsk.hr/urn:nbn:hr:168:927246

F. Sučić, "Destiliranje semantičke segmentacije", Undergraduate thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, Zagreb, 2024. Available at: https://urn.nsk.hr/urn:nbn:hr:168:927246

Metadata
TitleDestiliranje semantičke segmentacije
Title (english)Distillation of semantic segmentation
AuthorFilip Sučić
MentorSiniša Šegvić (mentor)
Committee memberSiniša Šegvić (predsjednik povjerenstva)
Committee memberJosip Šarić (član povjerenstva)
Committee memberIvan Martinović (član povjerenstva)
GranterUniversity of Zagreb
Faculty of Electrical Engineering and Computing
Zagreb
Defense date and country2024-07-01, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Computing
Abstract
Problem kod modela dubokog učenja koji postižu zapanjujuće rezultate je što su često glomazni i ne možemo ih koristiti u stvarnom vremenu. U ovom radu se istražuju metode kako učinkovito prenijeti znanje takvih modela na manje, efikasnije modele. Prikazane su osnovne metode, uključujući destilaciju pomoću tvrdih i mekih oznaka, kao i kombinaciju stvarnih tvrdih oznaka i mekih oznaka učitelja. Eksperimenti su provedeni na skupu podataka Cityscapes, pri čemu je model učitelj Mask2Former, a model student Swiftnet. Rezultati pokazuju da je moguće prenijeti znanje s većih modela na manje, uz zadržavanje zadovoljavajuće razine performansi.
Abstract (english)
The issue with deep learning models that achieve outstanding results is that they are often too cumbersome to be used in real-time applications. This paper explores methods to effectively transfer the knowledge of such models to smaller, more efficient ones. Basic methods are presented, including distillation using hard and soft labels, as well as a combination of true hard labels and soft teacher labels. Experiments were conducted on the Cityscapes dataset, with Mask2Former as the teacher model and Swiftnet as the student model. The results demonstrate that it is possible to transfer knowledge from larger models to smaller ones while maintaining a satisfactory level of performance.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:168:927246
Study programmeTitle: Computing
Study programme type: university
Study level: undergraduate
Academic / professional title: sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva (sveučilišni/a prvostupnik/ prvostupnica (baccalaureus/baccalaurea) inženjer/inženjerka računarstva)
Type of resourceText
File originBorn digital
Access conditionsClosed access
Terms of useLicence In copyright icon
Public note
Created on2024-10-12 22:10:54