Classification accuracy won't improve when running Keras example Image Classification

Question

I have been trying to train image classification networks (ResNet, EfficientNet, etc.) from the Keras Applications, both from scratch and only fine-tuning the final layers.