Aim

To study transfer learning in deep CNNs by fine-tuning pretrained models such as VGG19 and MobileNetV2 on the Oxford Flowers dataset, and to analyse how layer freezing, fine-tuning depth, learning rate, and dataset size influence feature extraction, training dynamics, representation quality, gradient flow, overfitting, and final inference performance, while also implementing and executing the models step by step to develop a clear understanding of their architecture, transfer learning workflow, training process, and inference behaviour.