| 引用本文: | 庾晨龙1,杨 天2,唐贝贝3.基于双调制策略的真实感风格迁移方法(J/M/D/N,J:杂志，M：书，D：论文，N：报纸).期刊名称,2026，43（1）：20-27 |
| CHEN X. Adap tive slidingmode contr ol for discrete2ti me multi2inputmulti2 out put systems[ J ]. Aut omatica, 2006, 42(6): 4272-435 |
|
| 摘要: |
| :目的 针对现有真实感风格迁移方法因缺乏有效的风格表示而导致内容失真和风格化强度不足的问题,提
出了一种名为 DMPST(Dual Modulation Photorealistic Style Transfer)的真实感风格迁移方法。 方法 为了有效地捕捉
风格信息,DMPST 利用了浅层和深层风格信息互补的优势,对多个尺度的风格特征进行渐进式融合。 然后,通过
双调制策略将融合后的风格特征变换为两组风格信号,并分别对内容特征进行特征调制和滤波调制,从而实现全
局性和局部性的风格化指导。 此外,为弥补风格化强度提高带来的内容信息丢失,设计了自适应空间插值模块,在
解码阶段进行多尺度的内容细节修复。 结果 通过定性和定量实验验证,DMPST 能够生成内容信息保存良好和高
度风格化的结果,在 DPST 数据集上,Content loss、Style loss、SSIM、PSNR 和 LPIPS 指标分别达到了 0. 595、0. 405、
0. 725、16. 894 和 0. 758。 结论 所提出的方法在保持源图像内容细节的前提下,实现了高度风格化的迁移效果,对
后续提高真实感风格化强度的研究具有参考价值。 |
| 关键词: 真实感风格迁移 双调制策略 渐进式融合 自适应空间插值 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Photorealistic Style Transfer Method Based on Dual Modulation Strategy |
|
YU Chenlong1，YANG Tian2，TANG Beibei3
|
|
1. School of Big Data and Artificial Intelligence Ma?? anshan University Ma?? anshan 243100 Anhui China
2. School of Computer Science and Engineering Anhui University of Science and Technology Huainan 232001 Anhui
China
3. School of Artificial Intelligence Anhui University of Science and Technology Huainan 232001 Anhui China
|
| Abstract: |
| Objective Aiming at the problems of content distortion and insufficient stylization intensity caused by the lack
of effective style representation in existing photorealistic style transfer methods a photorealistic style transfer method
named dual modulation photorealistic style transfer DMPST is proposed. Methods To efficiently capture the style
information DMPST utilized the complementarity of the shallow and deep style information to progressively fuse stylistic
features at multiple scales. Then through a dual-modulation strategy the fused style features were transformed into two
sets of style signals which were used for feature modulation and filter modulation of content features respectively so as to
achieve global and local stylization guidance. In addition to make up for the loss of content information due to the
increase in stylization intensity an adaptive spatial interpolation module was designed to repair multi-scale content details during the decoding stage. Results Qualitative and quantitative experiments verified that DMPST was able to generate
results with well-preserved content information and a high degree of stylization. On the DPST dataset the Content loss
Style loss SSIM PSNR and LPIPS indicators achieved 0. 595 0. 405 0. 725 16. 894 and 0. 758 respectively.
Conclusion The proposed method achieves a highly stylized transfer effect while maintaining the content details of the
source image and it has reference value for subsequent research on improving the intensity of photorealistic stylization. |
| Key words: photorealistic style transfer dual-modulation strategy progressive fusion adaptive spatial interpolation |
