CariMe: Unpaired Caricature Generation with Multiple Exaggerations

Gu, Zhongwei; Dong, Chuanqi; Huo, Jing; Li, Wenbin

doi:10.48550/arxiv.2010.00246

Cited by 1 publication

(6 citation statements)

References 37 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Ding et al (2020) proposed an unsupervised constrastive photo-to-caricature translation architecture, which includes two paired encoder-decoder networks and distortion prediction module to achieve style transfer and shape exaggeration. CariMe (Gu et al, 2021) proposes a multi-exaggeration warper network to learn the distribution-level mapping from photos to facial exaggerations and a styler to transfer the caricature style to the warped photo.…”

Section: Gan-based Methodsmentioning

confidence: 99%

“…In this section, we qualitatively and quantitatively evaluate our proposed method on both the WebCaricature and CaVINet datasets. We mainly compare with the GAN-based image translation methods, including CycleGAN (Zhu et al, 2017a) and MUNIT (Huang et al, 2018), and caricature generation methods, i.e., WarpGAN (Shi et al, 2019) and CariMe (Gu et al, 2021). The reason why we choose WarpGAN (Shi et al, 2019) as the representative method for caricature generation is that it does not require the annotation of facial landmarks for caricature images, which is under the same settings as our method.…”

Section: Comparison With Previous Methodsmentioning

confidence: 99%

“…As for the WebCaricature dataset, Figure 5 demonstrates the caricatures generated by our proposed method, WarpGAN (Shi et al, 2019), CariMe (Gu et al, 2021), CycleGAN (Zhu et al, 2017a), and MUNIT (Huang et al, 2018). As visualized in the figure, our proposed method achieves much better performance than the existing methods.…”

Section: Qualitative Evaluationmentioning

confidence: 99%

“…However, AdaIN transfers style by scaling and shifting each feature map with adaptive parameters, which cannot learn the relationship between the caricature style and photo image and is thus not able to appropriately fuse their features. As listed in the sixth column of Zheng et al (2019) CariGAN (Li et al, 2020b) CariGANs (Cao et al, 2018) MW-GAN (Hou et al, 2020) WarpGAN (Shi et al, 2019) Ding et al (2020 CariMe (Gu et al, 2021) Ours Decoupling indicates decoupling style transfer and shape exaggeration.…”

Section: Figurementioning

confidence: 99%

“…The first one, namely conventional methods (Akleman et al, 2000;Fujiwara et al, 2000;Chen et al, 2001;Liang et al, 2002;Gooch et al, 2004;Mo et al, 2004;Liu et al, 2006;Tseng et al, 2012;Yang et al, 2016), is composed of interactive methods, regularitybased methods, and learning-based methods. On the other hand, GAN-based methods (Cao et al, 2018;Shi et al, 2019;Zheng et al, 2019;Hou et al, 2020;Li et al, 2020b;Gu et al, 2021), mainly applies generative adversarial networks (GANs) (Goodfellow et al, 2014) for caricature generation.…”

Section: Related Workmentioning

confidence: 99%

See 4 more Smart Citations

Style attention based global-local aware GAN for personalized facial caricature generation

Zhao

Chen²,

Xie³

et al. 2023

Front. Neurosci.

View full text Add to dashboard Cite

IntroductionCaricature is an exaggerated pictorial representation of a person, which is widely used in entertainment and political media. Recently, GAN-based methods achieved automatic caricature generation through transferring caricature style and performing shape exaggeration simultaneously. However, the caricature synthesized by these methods cannot perfectly reflect the characteristics of the subject, whose shape exaggeration are not reasonable and requires facial landmarks of caricature. In addition, the existing methods always produce the bad cases in caricature style due to the simpleness of their style transfer method.MethodsIn this paper, we propose a Style Attention based Global-local Aware GAN to apply the characteristics of a subject to generate personalized caricature. To integrate the facial characteristics of a subject, we introduce a landmark-based warp controller for personalized shape exaggeration, which employs the facial landmarks as control points to warp image according to its facial features, without requirement of the facial landmarks of caricature. To fuse the facial feature with caricature style appropriately, we introduce a style-attention module, which adopts an attention mechanism, instead of the simple Adaptive Instance Normalization (AdaIN) for style transfer. To reduce the bad cases and increase the quality of generated caricatures, we propose a multi-scale discriminator to both globally and locally discriminate the synthesized and real caricature, which improves the whole structure and realistic details of the synthesized caricature.ResultsExperimental results on two publicly available datasets, the WebCaricature and the CaVINet datasets, validate the effectiveness of our proposed method and suggest that our proposed method achieves better performance than the existing methods.DiscussionThe caricatures generated by the proposed method can not only preserve the identity of input photo but also the characteristic shape exaggeration for each person, which are highly close to the real caricatures drawn by real artists. It indicates that our method can be adopted in the real application.

show abstract

Section: Gan-based Methodsmentioning

confidence: 99%