Semantička segmentacija harmoničkim gusto povezanim modelima

undergraduate thesis

• Title (english): Semantic segmentation with harmonic densely connected models
• Author: Anto Matanović
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Petra Bevandić (član povjerenstva)
• Committee member: Marin Kačan (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2022-07-15, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Tehnološki napredak omogućuje treniranje dubokih modela u razumnom vremenu. Modeli imaju raznolike upotrebe u raznim područjima poput medicine i prometa. Ovaj rad bavi se analizom prometnih scena te određivanjem objekata na slici, a to spada u područje semantičke segmentacije. Za rješavanje tog problema koristi se FC-HarDNet 70 arhitektura koja je pažljivo osmišljena tako da se smanji memorijsko opterećenje modela. FC-Hardnet 70 se pokazao puno efikasnijim od drugih sličnih modela poput DenseNet-a. Za treniranje i evaluaciju korišten je skup podataka CamVid. Opisani su osnovni pojmovi vezani za duboko učenje, korištene tehnologije i komentirani dobiveni rezultati.

Abstract (english)

Technological advance allows us to train deep models in real-time. These models
have a variety of uses in many different fields like medicine or traffic. This paper is
focused on interpreting and understanding scenes from traffic. That problem is known
as semantic segmentation and the main goal is to classify every pixel in the picture to
its label. To solve this problem, we used architecture FC-HarDNet 70 which was designed carefully to reduce dynamic random memory access(DRAM). FC-HarDNet 70
has proven to be more efficient than previous models like DenseNets. For training and
evaluation purposes we used the CamVid dataset. In this paper, we explained fundamental
concepts in deep learning as well as used technologies. In the end, we published and
interpreted the given results of the experiment.

• Keywords: Duboko učenje; konvolucijske neuronske mreže; semantička segmentacija; HardNet; CamVid
• Keywords (english): Deep learning; convolutional neural networks; semantic segmentation; HarDNet; CamVid
• Language: croatian
• URN:NBN: urn:nbn:hr:168:236302
• 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: 2023-01-15 10:47:20