A Novel Meta-Heuristic Tuned Hybrid GAN Models for Automatic Art Image Generation

Generative adversarial Networks (GAN) have attracted the brighter light of success on producing the high quality arts and images. Though, image generation procedure includes sampling from a latent space on learned arts, allows limited control across input space. These models nearly mimic established artistic styles but also synthesis the novel patterns with blending creativity. However, the existing methods still suffers from the several problems such as vanishing gradients, over-fitting problems and high computational cost that affects the performance of accurate art generation. In this context, this research paper proposes the Walrus Optimized Style2GAN model for achieving the better generation of arts. The extensive experimentation is performed by Kaggle's Art Images with 80k images and 27 classes. In this evaluation process, both quantitative metrics such as Frechet Inception Distance (FID), accuracy, precision and qualitative assessments are carried out using human perception studies. Experimental results demonstrates the performance of the model has outperformed the other models by producing the low FID scores and high generation accuracy. The results demonstrates the suggested model produce high attractive art images by intervention of human' opinion and feedbacks.