Publication

Abstract : Neural Style Transfer (NST) has emerged as a powerful technique for artistic image synthesis by fusing the base image with style source. In this study, we present a comparative analysis of NST using popular convolutional neural network (CNN) architectures, using pre-trained models such as VGG19, DenseNet121, ResNet50, and MobileNetV2. The methodology involves leveraging the deep features extracted by the models to balance content and style in synthesized images. We delve into the specifics of the model’s layer utilization, loss function definition, and TensorFlow implementation details. Experimental results demonstrate the model’s ability to produce visually compelling images that seamlessly merge content and style characteristics. Densenet121 has demonstrated superior performance compared to the other models in the evaluation. This work not only contributes to the field of NST but also offers a user-friendly TensorFlow framework, fostering further exploration and advancements in the realm of computer vision and image processing.