InkGAN: Generative Adversarial Networks for Ink-And-Wash Style Transfer of Photographs

Abstract

In this work, we present a novel approach for Chinese Ink-and-Wash style transfer using a GAN structure. The proposed method incorporates a specially designed smooth loss tailored for this style transfer task, and an end-to-end framework that seamlessly integrates various components for efficient and effective image style transferring. To demonstrate the superiority of our approach, comparative results against other popular style transfer methods such as CycleGAN is presented. The experimentation showcased the notable improvements achieved with our proposed method in terms of preserving the intricate details and capturing the essence of the Chinese Ink-and-Wash style. Furthermore, an ablation study is conducted to evaluate the effectiveness of each loss component in our framework. We conclude in the end and anticipate that our findings will inspire further advancements in this domain and foster new avenues for artistic expression in the digital realm.

ConIS: Controllable Text-Driven Image Stylization with Semantic Intensity

“…Nevertheless, the auto-encoder architecture is unsuitable for diffusion models. Zuo et al [20] introduce the style scaling injection module and the style degree interpretation module to the existing GAN-based style transfer methods [29,30]. This allows for controlling the degree of image stylization through fine-tuning the network with datasets.…”

Partial Modal Conditioned GANs for Multi-modal Multi-label Learning with Arbitrary Modal-Missing

“…Some other relevant works in this category are (Gygli et al, 2014), (Fu et al, 2017) -which uses the chat data as supplementary information to determine the location of highlights -and (Song, 2016) -which used the games special effects as a marker for the location of video highlights. Some inspiring work like (Yu et al, 2023) gives insights from another perspective for image treatment.…”

“…In (Ford et al, 2017) it was shown that online chat language exhibits unique linguistic patterns or rules that are "short", "sudden", or even "convulsive", in addition to the irregular use of words and non-words (emojis, acronyms, internet slang and jargons) that don't conform to canonical English grammar. This presents a serious challenge for NLP neural network models (Zhang et al, 2021;Chen et al, 2023;Yang et al, 2023) that are trained on clean English language.…”