University of Twente Student Theses

Login

A Comparative Study on Pre-trained Classifiers in the Context of Image Classification

Udrea, Mihnea-Adrian (2021) A Comparative Study on Pre-trained Classifiers in the Context of Image Classification.

[img] PDF
766kB
Abstract:Convolutional Neural Networks (ConvNets or CNNs) are nowadays the standard machine learning technique for analyzing visual imagery. As performance in the ILSVRC improves more and more, this paper questions the robustness and generalization abilities of state-of-the-art models. To test the hypotheses that four popular architectures (i.e., GoogLeNet, VGG19BN, ResNet152, and DenseNet161) are not significantly different when classifying images on the ImageNet, ImageNetV2, and ImageNetC benchmarks, the top-1 and top-5 accuracies are calculated and analyzed using the Iman-Davenport, Friedman’s Aligned Ranks and Bergmann-Hommel procedures. Our results show that there is enough evidence to reject the null hypotheses. We conclude that the four pre-trained networks do not have identical performance capabilities.
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:http://purl.utwente.nl/essays/85666
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page