Prepoznavanje objekata na slikama

undergraduate thesis

• Title (english): Object Recognition
• Author: Hrvoje Čukman
• Mentor: Tomislav Pribanić (mentor)
• Committee member: Tomislav Pribanić (predsjednik povjerenstva)
• Committee member: Kristijan Bartol (član povjerenstva)
• Committee member: David Bojanić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-01-27, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U ovom radu opisan je postupak treniranja modela na 20 različitih tipova hrane. Treniranje je napravljeno pomoću prijenosnog učenja. Od već unaprijed istreniranog modela na jako velikom broju slika pomoću Create ML-a se prenjelo to znanje na moj model podešavajući samo određene klasifikacijske slojeve. Slike koje su preuzete i na kojima se treniralo su slike iz skupa \textit{Google Open Images Dataset}, koji sadrži više od 9 milijuna slika organiziranih u tisuće kategorija. Iz tog skupa su odabrane 20 kategorija te su se preuzele bilješka koje sadrže metapodatke za svaku od njih. Cilj ovog rada je pokazati način na koji se model trenira na osobnom računalu (konretno Mac uređaju) te primjenu istreniranog modela pri prepoznavanju objekata na slikama u stvarnom. Nakon dovršetka treniranja idući korak je dobiveni model prebaciti u aplikaciju te prilagoditi aplikaciju da u stvarnom vremenu radi predviđanja i prikazuje ih na ekran. Aplikacija je napravljena u programskom jeziku Swift te je kompatabilna za iOS uređaje.

Abstract (english)

This paper describes the procedure of training models on 20 different types of food. The training was done using transfer learning. From a pre-trained model on a very large number of images using Create ML, knowledge was transferred to my model by adjusting only certain classification layers. The images downloaded and trained on are images from the Google Open Images Dataset, which contains more than 9 million images organized into thousands of categories. From this set, 20 categories were selected and notes containing metadata for each of them were downloaded. The aim of this paper is to show the way in which the model is trained on a personal computer (specifically a Mac device) and the application of a trained model in recognizing objects in real images. After completing the training, the next step is to transfer the resulting model to the application and adjust the application to make predictions in real-time and display them on the screen. The application is made in the Swift programming language and is compatible with iOS devices.

• Keywords: Strojno učenje; Prijenosno učenje; Create ML; Treniranje klasifikatora slike; iOS Aplikacija; Prepoznavanje hrane na slikama
• Keywords (english): Machine learning; Transfer learning; Create ML; Image classifier training; iOS app; Food recognition in images
• Language: croatian
• URN:NBN: urn:nbn:hr:168:070703
• Study programme: Title: 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 resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-14 08:39:51