University of Twente Student Theses

Login

Performance Comparison of CNN-based Semantic Segmentation on Indoor and Outdoor Scenes

Irokosu, Anthony (2023) Performance Comparison of CNN-based Semantic Segmentation on Indoor and Outdoor Scenes.

[img] PDF
361kB
Abstract:Semantic segmentation is an important task in computer vision. It involves the assignment of class labels to regions in an image. The use of Convolutional Neural Networks has been successful for semantic segmentation; however, the performance of CNN-based semantic segmentation models can vary depending on the properties of the input data. Such properties are found in the indoor or outdoor environment. This paper presents a comprehensive performance comparison of CNN-based semantic segmentation models on indoor and outdoor datasets. The goal is to highlight the environment’s impact on the effectiveness of CNN-based segmentation models. To facilitate the conduct of the research, a DeepLabv3Plus based on the ResNet-101 CNN architecture was evaluated on datasets specific to indoor and outdoor environments. The datasets were from the popular benchmark dataset, Pascal Visual Object Classes 2012. The performance of the CNN-based semantic segmentation model is assessed using well-known evaluation metrics such as mean over intersection union (mIoU), frequency-weighted mIoU, pixel accuracy, and class accuracy. The evaluation shows better performance of the CNN-based semantic segmentation in outdoor environments compared to indoor environments across all metrics.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science BSc (56964)
Link to this item:https://purl.utwente.nl/essays/96027
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page