Download PDFOpen PDF in browserExperimentally Enhancing ResNet50 Performance on the Intel Dataset Through Architectural Modifications10 pages•Published: August 6, 2024AbstractThis study focuses on enhancing the performance of the ResNet50 model on the Intel dataset, a collection of images depicting diverse natural scenes under various environmental conditions. While ResNet50 has shown remarkable performance in image classification tasks, its application to the Intel dataset reveals certain limitations in accurately discern- ing subtle features within scenes. To address this, proposed architectural modifications to ResNet50 aimed at capturing intricate features specific to the Intel dataset. Four distinct modifications are introduced, tailored to exploit different aspects of scene complexity present in the dataset. Through extensive experimentation and evaluation, we demonstrate the effectiveness of these modifications in improving the model’s classification accuracy on the Intel dataset. the findings not only contribute to advancing deep learning methodologies for image analysis but also underscore the importance of tailored model design for specific task domains.Keyphrases: architectural modifications, neural networks, resnet50, scene classification, spatial pyramid pooling, transfer learning In: Rajakumar G (editor). Proceedings of 6th International Conference on Smart Systems and Inventive Technology, vol 19, pages 445-454.
|
