Abstract: This paper investigates the performance of widely used pre-trained CNN architectures (VGG16, MobileNetV3, DenseNet121, and RegNet040) across diverse datasets, particularly focusing on ...