preTrained Convolutional Neural Networks Performance Assessment under Different Color Variability Indexes
Alejandro Garcia Magaña, Jaime Cerda, Salvador Ramírez-Zavala · Procedia Computer Science · 2022
Convolutional Neural Networks (CNN) are currently widely used in image classification, so efficiency is key at the time of their implementation. Actually there are several pretrainned CNNs (ptCNN) for image classification e.g. Vgg19, ResNet50, InceptionV3, Xception, DenseNet, EfficientNet-B0, and so on. This document shows that the performance in the classification of images using ptCNN can be affected if images with different levels of color variability, as well as the brightness and standard deviation in the images are used. However, even that some changes in efficiency were found, we can conclude that the ptCNN are stable under changes in those indexes.