Performance Optimization of GAN-based Image Style Transfer on Indoor Geometric Shaped Data

As the applications of image processing is increasing day by day, the requirement of image enhancing techniques are also increasing. Generative Adversarial Network is a type of deep learning model created by Ian Goodfellow in 2014 for new image generation with existing datasets/images. Since then, various models have been created by using these network models. The main applications of these models include image generation, image enhancement, image segmentation and restoration, object detection, new character generation, image augmentation, image style transfer etc. This paper proposed a different approach with existing deep learning models for indoor image style transfer on geometric shapes only and calculate the effective output in terms of adversarial loss, perceptual loss, and diversity. The model is specifically designed to transfer the style on the selected area or geometric shape of the image of indoor, that may be an image of office, home, or some other building. Each loss is calculated separately to minimize the impact of loss on the output image. The paper describes a detailed analysis of losses of proposed model using vgg19 deep learning model. The proposed model can be used for commercial purposes in interior designing of homes, offices, buildings, fashion designing and in augmented reality and virtual reality to generate real-time images with different styles.