ColorwAI: Generative Colorways of Textiles through GAN and Diffusion Disentanglement

TL;DR

It is suggested that disentanglement can solicit a creative system for colorway creation, and evaluate it through expert questionnaires and creativity theory.

Abstract

Colorway creation is the task of generating textile samples in alternate color variations maintaining an underlying pattern. The individuation of a suitable color palette for a colorway is a complex creative task, responding to client and market needs, stylistic and cultural specifications, and mood. We introduce a modification of this task, the "generative colorway" creation, that includes minimal shape modifications, and propose a framework, "ColorwAI", to tackle this task using color disentanglement on StyleGAN and Diffusion. We introduce a variation of the InterfaceGAN method for supervised disentanglement, ShapleyVec. We use Shapley values to subselect a few dimensions of the detected latent direction. Moreover, we introduce a general framework to adopt common disentanglement methods on any architecture with a semantic latent space and test it on Diffusion and GANs. We interpret the color representations within the models' latent space. We find StyleGAN's W space to be the most aligned with human notions of color. Finally, we suggest that disentanglement can solicit a creative system for colorway creation, and evaluate it through expert questionnaires and creativity theory.

Figures (7)

• Figure 1: Example of a colorway in cold and warm tones. Pattern generated using StyleGAN2-ADA and colorway generated using InterfaceGAN.
• Figure 2: Schematization of the ColorwAI framework. We show the disentanglement steps for the two architectures, their latent spaces, and the inference procedure. Steps 1 and 4 are different for GANs and Diffusion, while 2 and 3 are common.
• Figure 3: Colorways in three colors for three patterns using InterfaceGAN, ShapleyVec, and StyleSpace on StyleGAN and DDM. The examples are chosen because of their representativity of general observed trends.
• Figure 4: Relationship between color representations for StyleGAN and DDM using ShapleyVec. The relationships are shown in terms of similarity between the disentangled color vectors, in terms of which dimensions of the latent space encode which color(s), and of how often the modifications using such color directions result in other colors. For clarity, we only show results for a subset of color directions, which cover the spectrum of colors available, sacrificing some shades of brown and blue.
• Figure 5: Predominant color annotation. From left to right: palette extraction, codebook identification, color quantization. The final annotated color is the lower box of each rightmost image.