PSR-GAN: a product concept sketch rendering method based on generative adversarial network and colour tags

Looking for the fit of colour and form is an important goal of product design, but conceptual visualisation based on hand-drawn sketches, a necessary way in the early stage, consumes a lot of designers' time and energy, resulting in missing fleeting design inspirations and clues. We propose a product concept sketch rendering method based on generative adversarial network and colour tags to assist designers in quickly capturing the harmony and balance between product colour and form, and improving design efficiency. This method encodes colour semantic information as the condition and combines ACGAN to achieve controllable rendering of line sketches based on specified colour tags. The generator is a hybrid of CNN and Transformer, further guided to optimise by combining pixel-wise loss and perceptual loss, while the discriminator adopts a convolution-based spatial-channel attention structure. Results show that PSR-GAN outperforms existing methods in terms of generation quality, and it also demonstrates excellent rendering results compared to professional manuscripts. Designers can use this method not only to obtain real-time comprehensive conceptual feedback but also to effectively narrow the colour search space for product details, accelerating the convergence of their design ideas during the sketch phase.