Research on the Interpretability and Controllability of Deep Learning-Based Exhibition Design Style Generation and Transfer

With the rapid development of deep learning technology, Generative Adversarial Networks (GANs) have achieved significant results in the field of image generation. However, traditional GAN models are often seen as “black boxes,” with their internal decision-making processes lacking transparency. This limits their trustworthiness and controllability in practical applications. This study addresses the needs for generating and transferring exhibition hall design styles by proposing a deep learning model that combines Explainable Artificial Intelligence (XAI) technology and a user preference adaptation system. By introducing an attention mechanism to design interpretability modules, the transparency of the generation process is enhanced. A personalized user input interface is developed, enabling controllability and flexibility in style generation. Moreover, a multimodal information fusion method is employed to improve the accuracy of text-to-image generation. The experimental results show that the improved model not only enhances user trust and satisfaction with the generated results but also demonstrates high accuracy and diversity in the tasks of generating and transferring exhibition hall design styles.