Leveraging AI-Generated Image for Scene Classification: A transfer Learning Approach

When it comes to scene classification, the lack of labelled data is a significant obstacle that forces researchers to think about innovative solutions. This work presents a revolutionary method to strengthen scene classification models by combining transfer learning with AI-generated images. A key component of this project is the construction of a carefully curated dataset of 3672 augmented AI-generated photos in a total of 6 scenes. This database not only serves as a helpful resource for additional research but also makes a substantial contribution to this area of research. Using this dataset, a VGG16 model is trained, demonstrating the usefulness of synthetic data with a notable increase in accuracy. The model's performance is further improved through subsequent training using realistic photos through transfer learning, thereby surpassing the initial baseline. This work demonstrates the transformative effect of transfer learning on scene categorization accuracy and verifies its effectiveness using synthetic data through rigorous testing and evaluation. The results open the door for AI-generated photos to be used as an effective approach to address data scarcity issues in scene classification, which will help real-world applications improve.