Unsupervised Coherent Video Cartoonization with Perceptual Motion Consistency

Authors: Zhenhuan Liu, Liang Li, Huajie Jiang, Xin Jin, Dandan Tu, Shuhui Wang, Zheng-Jun Zha1846-1853

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Qualitative and quantitative experiments demonstrate our method is able to generate highly stylistic and temporal consistent cartoon videos. We conduct detailed qualitative and quantitative experiments and demonstrate our method achieves both stylistic cartoon effect and temporal consistency.
Researcher Affiliation Collaboration Zhenhuan Liu1,2, Liang Li1 , Huajie Jiang3 Xin Jin3, Dandan Tu3, Shuhui Wang1, Zheng-Jun Zha4 1 Institute of Computing Technology, 2 University of Chinese Academy of Sciences 3 Huawei Technologies, 4 University of Science and Technology of China
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No Our code will be released on github.
Open Datasets Yes For real-world photos, we adopt 10000 human face images from FFHQ dataset and 6227 landscape images from Cycle GAN dataset (Zhu et al. 2017). For cartoon images, we use images from Whitebox GAN (Wang and Yu 2020), including 10000 images of cartoon faces from P.A.Works, Kyoto animation and 14615 images from cartoon movies produced by Shinkai Makoto, Hosoda Mamoru, and Miyazaki Hayao.
Dataset Splits No The paper mentions training and test sets, but does not provide explicit details on validation dataset splits, such as percentages or sample counts for a distinct validation set.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies Yes We implement our model using Pytorch Lightning (Falcon 2019).
Experiment Setup Yes The local patch width of SSA is set as R = 3. Patch discriminator with spectral normalization (Miyato et al. 2018) is adopted to identify each patch s distribution. We use Adam (Kingma and Ba 2015) optimizer with momentums 0.5 and 0.99. The learning rate is set to 0.0002. The loss weight are set as λ1 = 0.1, λ2 = 1, λ3 = 200, λ4 = 200, λ5 = 20000, λ6 = 0.1.