TexGAN: Textile Pattern Generation Using Deep Convolutional Generative Adversarial Network (DCGAN)

Generative Adversarial Networks (GANs) are used for image, video, and 3D scene generation along with their ability to enhance images. In image generation, GANs generate new images by learning features from the existing images. Features are extracted by a convolutional network, and images are generated through a de-convolutional network. GANs have been recently applied to the field of fashion where they have been used to generate new kinds of shoes and clothes. In the field of textile design, designs are printed onto clothes. These designs are mostly made of single repetitive patterns. These patterns need to be created by the designers and then these patterns are photoshopped to make the whole design. Generating textile patterns through GANs can help designers to take inspiration for new ideas. To achieve this goal GANs have been proposed to generate new patterns. Different variants of GANs are available in fields and they helped in many other fields along with the field of fashion. At first, we formulated a dataset of 17k+ images collected via google images. The dataset is available at the GitHub repository of the project (https://github.com/Gnahy/TexGAN.git). We provided an idea of generating textile patterns in general and one can also specify the patterns they want to generate i.e., cheetah, geometric, and abstract. DCGAN has been trained on all the patterns collectively up to a specific point and the weights from the trained model have been saved. After that point model is trained on specific kinds of patterns i.e., cheetah in order to generate that specific kind of new patterns. This technique helps even when the data for the new patterns are small and resources to train the model are fewer.