In the field of computer vision, image classification tasks have been significantly advanced by deep learning techniques, particularly through the use of Convolutional Neural Networks (CNNs). This research paper explores the application of transfer learning and fine-tuning on the pre-trained VGG16 model for the binary classification of images into dogs and cats. Utilizing the popular Dogs vs. Cats dataset from Kaggle, which consists of 20,000 training images and 5,000 validation images, we demonstrate how freezing initial layers and fine-tuning the later convolutional blocks can achieve high accuracy with reduced training time and computational resources. The methodology involves loading the VGG16 base, setting specific layers as trainable, adding custom dense layers, and training the model over five epochs. Results show a validation accuracy of 95.42% after fine-tuning, with detailed analysis of training history, loss curves, accuracy plots, and a confusion matrix. This approach highlights the efficacy of transfer learning in handling limited datasets and provides insights into model performance through visualizations and statistical evaluations. The paper also includes a comprehensive literature survey of related works, drawing from 15 key references, and discusses implications for future research in image classification. The study contributes to the understanding of fine-tuning strategies in deep learning, emphasizing practical implementations in Python with TensorFlow and Keras.
Vishwa Patel (Sat,) studied this question.