Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
102 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Anisotropic Stroke Control for Multiple Artists Style Transfer (2010.08175v2)

Published 16 Oct 2020 in cs.CV and cs.AI

Abstract: Though significant progress has been made in artistic style transfer, semantic information is usually difficult to be preserved in a fine-grained locally consistent manner by most existing methods, especially when multiple artists styles are required to transfer within one single model. To circumvent this issue, we propose a Stroke Control Multi-Artist Style Transfer framework. On the one hand, we develop a multi-condition single-generator structure which first performs multi-artist style transfer. On the one hand, we design an Anisotropic Stroke Module (ASM) which realizes the dynamic adjustment of style-stroke between the non-trivial and the trivial regions. ASM endows the network with the ability of adaptive semantic-consistency among various styles. On the other hand, we present an novel Multi-Scale Projection Discriminator} to realize the texture-level conditional generation. In contrast to the single-scale conditional discriminator, our discriminator is able to capture multi-scale texture clue to effectively distinguish a wide range of artistic styles. Extensive experimental results well demonstrate the feasibility and effectiveness of our approach. Our framework can transform a photograph into different artistic style oil painting via only ONE single model. Furthermore, the results are with distinctive artistic style and retain the anisotropic semantic information. The code is already available on github: https://github.com/neuralchen/ASMAGAN.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (5)
  1. Xuanhong Chen (16 papers)
  2. Xirui Yan (1 paper)
  3. Naiyuan Liu (5 papers)
  4. Ting Qiu (2 papers)
  5. Bingbing Ni (95 papers)
Citations (11)

Summary

Anisotropic Stroke Control for Multi-Artist Style Transfer

The paper "Anisotropic Stroke Control for Multiple Artists' Style Transfer" introduces a novel approach to artistic style transfer that adeptly addresses both the preservation of semantic information and the ability to transfer multiple styles using a single model. This method represents a significant progression from previous style transfer approaches that required separate models for each target style, often resulting in excessive distortion of semantic details.

The central innovation of the paper is the proposed Anisotropic Stroke Module (ASM) which confers the ability to achieve adaptive semantic-consistency among different styles. This module allows for dynamic adjustment of style strokes between non-trivial regions (such as faces) and more trivial regions (such as backgrounds), based on the content of the image. This is particularly crucial as artists naturally use varying stroke sizes to emphasize important regions and minimize details in less critical areas.

Alongside the ASM, the paper introduces a Multi-Scale Projection Discriminator. This discriminative model significantly enhances the mechanism of style transfer by exploiting multi-scale texture clues to effectively distinguish among a wide range of artistic styles. Unlike a single-scale discriminator, the multi-scale approach allows the model to capture diverse levels of detail and stroke characteristics, thereby improving the discriminator's ability to guide the generator in producing high-fidelity artistic renderings.

In terms of results, the proposed model demonstrates its capacity to transform a single content image into various artistic styles in one unified framework. This flexibility offers considerable advantages in terms of deployment and computational efficiency, requiring only a single model to handle multiple style outputs, in stark contrast to methods necessitating a distinct model for each style.

From a practical standpoint, the implications of the research are twofold: Firstly, it offers a practical tool for digital artists and creators to experiment with different stylistic effects quickly and efficiently. Secondly, it harnesses neural networks' capabilities to replicate complex artistic styles in a computational framework, which may extend into broader applications in computational art and automated design systems.

Theoretically, the research contributes to ongoing discussions around the neural representation of artistic styles, stroke manipulation in style transfer, and the importance of retaining semantic content during image transformation processes. By proposing the ASM and the multi-scale approach to discrimination, the paper offers novel insights into how neural networks can be architecturally adapted to mimic nuanced human artistic practices.

Future developments in AI-driven art generation may look into further refining the integration of semantic understanding and style modulation, potentially leveraging unsupervised learning techniques to improve upon the style transfer processes without extensive style datasets. Moreover, expanding the model's capabilities to include more complex styles or even interactive style adaptation could significantly enhance the user experience in digital art creation tools.

In summary, the anisotropic stroke control framework adeptly delivers an efficient, semantic-preserving, and flexible solution to the challenge of multi-artist style transfer, boasting both practical utility and deep theoretical underpinnings within the artificial intelligence and computational art communities.