Learning to Caricature via Semantic Shape Transform

Chu, Wenqing; Hung, Wei Chih; Tsai, Yi Hsuan; Chang, Yu; Li, Yijun; Yang, Ming–Hsuan

doi:10.1007/s11263-021-01489-1

Cited by 9 publications

(5 citation statements)

References 40 publications

(71 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As shown in Fig. 7, CariME (Gu et al 2021) and Semantic-CariGAN (Chu et al 2021) can refer to the reference caricatures to provide different exaggeration styles, but the quality of the generated caricatures is low. For example, in the second row, the subject's texture is blurry.…”

Section: Comparison To Caricature Generation Methods Withoutmentioning

confidence: 99%

“…The questions without reference provide a photo and the results of 5 caricature generation methods (CariGANs (Cao, Liao, and Yuan 2018), Warp-GAN (Shi, Deb, and Jain 2019), AutoToon (Gong, Hold-Geoffroy, and Lu 2020), StyleCariGAN (Jang et al 2021), andCariPainter (Huang et al 2022)) which can only generate random geometry styles and the results of our method. The questionnaires with reference provide an additional reference caricature and the results of 3 caricature generation methods (CariME (Gu et al 2021), Semantic-CariGAN (Chu et al 2021), andDualStyleGAN (Yang et al 2022)) and our method. Each question requires users to select an option that is closest to a real hand-drawn caricature.…”

Section: User Perceptual Evaluationmentioning

confidence: 99%

“…Graphicbased methods (Akleman 1997a;Akleman, Palmer, and Logan 2000) pay attention to exaggeration, while image-toimage conversion methods (Huang et al 2018;Zhu et al 2017) mainly focus on color transfer. With the advancement of artificial neural networks, several methods based on generative adversarial networks (GAN) have been proposed (Cao, Liao, and Yuan 2018;Shi, Deb, and Jain 2019;Chu et al 2021;Gu et al 2021), combing both exaggeration and color transfer. However, the quality of caricatures generated by these methods is generally not high, as exaggeration through 2D image warping tends to cause distortion.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

G2L-CariGAN: Caricature Generation from Global Structure to Local Features

Huang,

Bai,

Liang

et al. 2024

AAAI

View full text Add to dashboard Cite

Existing GAN-based approaches to caricature generation mainly focus on exaggerating a character’s global facial structure. This often leads to the failure in highlighting significant facial features such as big eyes and hook nose. To address this limitation, we propose a new approach termed as G2L-CariGAN, which uses feature maps of spatial dimensions instead of latent codes for geometric exaggeration. G2L-CariGAN first exaggerates the global facial structure of the character on a low-dimensional feature map and then exaggerates its local facial features on a high-dimensional feature map. Moreover, we develop a caricature identity loss function based on feature maps, which well retains the character's identity after exaggeration. Our experiments have demonstrated that G2L-CariGAN outperforms the state-of-arts in terms of the quality of exaggerating a character and retaining its identity.

show abstract

Section: Comparison To Caricature Generation Methods Withoutmentioning

confidence: 99%

Section: User Perceptual Evaluationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

G2L-CariGAN: Caricature Generation from Global Structure to Local Features

Huang,

Bai,

Liang

et al. 2024

AAAI

View full text Add to dashboard Cite

show abstract

“…The number of altered values depends on the noise ratio (NR), representing the percentage of noisy values relative to the image size. Alterations in the image values to 0 or 255 are evident in the image histogram [9,10], as illustrated in Figures 2 and 3, where an increased occurrence of 0s and 255s indicates the image has been affected by SAPN. When digital images are contaminated with noise, it becomes imperative to employ image processing techniques to eliminate or minimize the detrimental effects of noise.…”

Section: Figure 1 Color Image Representationmentioning

confidence: 96%

Enhanced Average Filtering Technique for Mitigating Salt and Pepper Noise in High-Resolution Color Images

Khrisat¹

2023

View full text Add to dashboard Cite

Digital color images play a crucial role in various critical applications, necessitating the development of effective noise reduction techniques to preserve image quality and characteristics. Salt and pepper noise, in particular, can significantly degrade digital image quality, with the extent of the impact contingent upon image size and noise ratio. Existing methods reliant on arithmetic mean and median filtering have proven inadequate for addressing high noise ratios. In this study, we propose and implement a novel average filter-an enhancement over traditional mean filters-to efficiently mitigate salt and pepper noise, specifically in cases with high noise ratios. Our results demonstrate the superior performance of the proposed filter in terms of preserving image quality and reducing noise, as evidenced by improved maximum auto-correlation factors and peak signal-to-noise ratios.

show abstract

“…Cartoons are one of the most important visual tools that make teaching interesting (Özer, 1990). However, creating cartoons is a difficult task that requires imagination and artistic skills (Chu et al, 2021).…”

Section: Introductionmentioning

confidence: 99%

The Effect of Cartoon-Enriched Instruction on Primary School Students’ Academic Achievement and Attitude in Turkish Courses

HAYRAN

Şahin

Özüdoğru

2022

Journal of Teacher Education and Lifelong Learning

View full text Add to dashboard Cite

This study aimed to determine the effect of cartoon-enriched instruction on 4th grade primary school students’ academic achievement and attitudes towards Turkish courses. The research was carried out with a total of 62 students studying in two separate fourth grade classes of a public primary school during the 2019-2020 academic year. A quasi-experimental design with pretest-posttest control group was used in the study. While the cartoon-enriched method was implemented with the students in the experimental group, regular curriculum was followed with the students in the control group. Achievement test and the attitude scale were applied to the students as a pre-test and post-test to obtain the research data. Normality tests, descriptive analyses, Mann-Whitney U and Independent Groups t-test were used in data analysis. According to the findings obtained as a result of the research, it was concluded that there was a difference between academic achievement and attitudes towards the course in favor of the experimental group, but this difference was not significant. While there was a difference in academic achievement according to gender in the experimental group students in favor of female students, there was no difference in the attitude towards the course according to gender. The research can also be applied to other courses taught in primary schools. Different humor elements can be used together with cartoons to improve students’ attitude towards humor in a positive way. In the research, some acquisitions were selected and cartoons were created in a limited way. The study can be replicated by creating cartoons for all other acquisitions in Turkish courses.

show abstract

Learning to Caricature via Semantic Shape Transform

Cited by 9 publications

References 40 publications

G2L-CariGAN: Caricature Generation from Global Structure to Local Features

G2L-CariGAN: Caricature Generation from Global Structure to Local Features

Enhanced Average Filtering Technique for Mitigating Salt and Pepper Noise in High-Resolution Color Images

The Effect of Cartoon-Enriched Instruction on Primary School Students’ Academic Achievement and Attitude in Turkish Courses

Contact Info

Product

Resources

About