To provide semantic image style transfer results which are consistent with human perception, transferring styles of semantic regions of the style image to their corresponding semantic regions of the ...content image is necessary. However, when the object categories between the content and style images are not the same, it is difficult to match semantic regions between two images for semantic image style transfer. To solve the semantic matching problem and guide the semantic image style transfer based on matched regions, we propose a novel semantic context-aware image style transfer method by performing semantic context matching followed by a hierarchical local-to-global network architecture. The semantic context matching aims to obtain the corresponding regions between the content and style images by using context correlations of different object categories. Based on the matching results, we retrieve semantic context pairs where each pair is composed of two semantically matched regions from the content and style images. To achieve semantic context-aware style transfer, a hierarchical local-to-global network architecture, which contains two sub-networks including the local context network and the global context network, is proposed. The former focuses on style transfer for each semantic context pair from the style image to the content image, and generates a local style transfer image storing the detailed style feature representations for corresponding semantic regions. The latter aims to derive the stylized image by considering the content, the style, and the intermediate local style transfer images, so that inconsistency between different corresponding semantic regions can be addressed and solved. The experimental results show that the stylized results using our method are more consistent with human perception compared with the state-of-the-art methods.
Transfer learning-based methods hold promise for enhancing classification task performance. However, a transfer learning mechanism for hard-to-classify time series classification tasks caused by ...limited samples in training set is still to be accomplished. Drawing inspiration from domain adaptation and style transfer methods in computer vision, we aim to address sample scarcity and enhance classifier generalization by leveraging abundant unrelated time series datasets. In this paper, a transfer learning mechanism called feature-level style transfer, where the feature alignment is conducted in a domain adaptation-inspired manner, has been proposed to target initially hard-to-classify time series tasks with limited training samples. The proposed mechanism eliminates constraints on source dataset selection, enabling the utilization of weakly-related or unrelated datasets to enhance classifier performance. Furthermore, a voting mechanism has been formulated to achieve more accurate predictions by taking transferable information learnt from different source domains into comprehensive consideration.
•The TSC transfer learning using weakly-related datasets has been achieved.•Focusing on TSC datasets with limited training samples.•Accommodating the difference in label sets, lengths and channel numbers.•A voting mechanism has been proposed for multi-source transfer learning.
We propose a novel framework to transfer the portrait image into its correspondence with photo-realistic and cartoon style. The existing work on neural style transfer conducts impressive results on ...artistic style transfer; however, the lack of semantic clues will lead to the color artifacts in photo-realistic style transfer because of the complex background and noise issues. In this work, we re-define the semantics as the pixel motion field according to the color displacement between adjacent animation frames along the optical direction and initiatively propose the self-supervised semantic network (SSNet) to learn semantic maps without human inference or any priories. The SSNet shares parameters with the style transfer network; thus, the superior alternatives can preserve the semantic completeness in the styled image. To solve the content missing and blur problems common in NST, we propose the bilateral convolution block (B-block) and feature fusion strategy (F-block) for visual smoothness to meet the perceptive satisfaction. The ablation studies are provided to validate the effectiveness, and comparative experiments with the state-of-the-art baselines demonstrate the advantages of the proposed method.
Authorship attribution is an NLP task to identify the author of a text based on stylometric analysis. On the other hand, authorship obfuscation aims to protect against authorship attribution by ...modifying a text’s style. The main challenge in authorship obfuscation is how to keep the content of the text despite the text modification. In this research, we are applying text style transfer methods for modifying the writing style while preserving the content of the input text. We implemented two unsupervised text style transfer: dictionary-based and back translation methods to change the formality level of the text. Experiment results shows that the back-translation method outperformed the dictionary-based method. The authorship attribution performance decreased up to 16.15% and 23.66% on F1-score for 3 and 10 authors respectively using back-translation. While for dictionary-based method the F1-score dropped up to 1.99% and 11.56% for 3 and 10 authors respectively. Evaluation on sensibleness and soundness factors show that the back-translation method can preserve the semantic of the obfuscated texts. Moreover, the modified texts are well-formed and inconspicuous.
In this paper, we propose a new cross-lingual font style transfer model, FCAGAN, which enables font style transfer between different languages by observing a small number of samples. Most previous ...work has been on style transfer of different fonts for single language content, but in our task we can learn the font style of one language and migrate it to another. We investigated the drawbacks of related studies and found that existing cross-lingual approaches cannot perfectly learn styles from other languages and maintain the integrity of their own content. Therefore, we designed a new full-domain convolutional attention (FCA) module in combination with other modules to better learn font styles, and a multi-layer perceptual discriminator to ensure character integrity. Experiments show that using this model provides more satisfying results than the current cross-lingual font style transfer methods. Code can be found at https://github.com/jtlxlf/FCAGAN.
•We propose a full-domain convolutional attention module to extract information from feature maps.•We propose adaptive fusion to improve the performance of feature fusion.•We design a channel-dependent multi-level feature extraction module to extract style encodings from style feature maps at different scales.•We design a multi-layer perceptual discriminator that more accurately identifies whether the samples are real or not.•We use the quadratic smoothing hinge loss function for the first time in the field of font style transfer.
Scene cartoonization aims to convert photos into stylized cartoons. While generative adversarial networks (GANs) can generate high‐quality images, previous methods focus on individual images or ...single styles, ignoring relationships between datasets. We propose a novel multi‐style scene cartoonization GAN that leverages multiple cartoon datasets jointly. Our main technical contribution is a multi‐branch style encoder that disentangles representations to model styles as distributions over entire datasets rather than images. Combined with a multi‐task discriminator and perceptual losses optimizing across collections, our model achieves state‐of‐the‐art diverse stylization while preserving semantics. Experiments demonstrate that by learning from inter‐dataset relationships, our method translates photos into cartoon images with improved realism and ion fidelity compared to prior arts, without iterative re‐training for new styles.
We introduce a multi‐style scene cartoonization GAN aiming to enhance the technique of photo‐to‐cartoon conversion. By amalgamating multiple cartoon datasets and employing innovative encoding methods, our model achieves more realistic and cartoon effects, surpassing previous approaches. By capturing relationships between datasets, we can provide high‐quality cartoon images without the need for tedious iterative retraining, marking a subtle but significant advancement in the field.
Neural Style Transfer (NST) has exerted algorithms to generate animation images in computer vision for decades. The Convolution Neural Network (CNN) applied to image content and styles in the NST has ...improved the extraction of functionalities and the calculation of the convergence speed to recognize and generate high quality structure images, but unpredictable loss elements are inadequate to iterate human learning ability of unique artists’ paintings or styles. This paper offers a chaotic VGG10 NST model based on CNN, ReLU and Lee-Oscillator. The proposed ReLU-Oscillator dynamically relies on activation functions in a chaotic state that dynamically improves high-frequency iterative training of high-quality image and high-speed time optimization. In addition, the best parameters recognizing and determining a personalized painting style from the search for ReLU-Oscillator by corresponding to a set of parameters from proposed Optimal Oscillator Parameter Search Algorithm. Experimental results showed that the stylized image generated by the Chaotic VGG 10 model with high-frequency oscillation succeeded in reducing the training time in magnitude models with the smallest Params and FLOPs in model performance and image quality with the lowest content loss for preserving semantic information and moderate style loss for style similarity balanced with comparison to 8 state-of-the-art models in visual perception evaluation. Chaotic NST has a unique identification for each artist supplemented with a set of oscillator parameters to evaluate the loss performance based on the relative error between the famous painting and its imitation, indicating that the authentication of paintings can be detected through specific ReLU-Oscillator parameters for each stylization, estimated from the loss value performance.