A Technique for Building 3D Game Characters from Cartoon Character Images Using Deep 3D Face Reconstruction
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
We propose a deep learning-based 3D face reconstruction technique for webtoon character images. Our method consists of a default module and a detail module. The default module incorporates the Albedo module to calculate the amount of incoming light on the face and generate an albedo map for the input webtoon character image. Additionally, the FLAME module, another component of the default module, generates the basic 3D face shape for the input webtoon character image. Simultaneously, the detail module is applied to extract the detailed features such as facial expressions and depth of the transformed webtoon character image, which differ from real human faces. As a result of the detail module a detail map is generated and combined with the 3D face shape generated by the FLAME module to generate the detail face shape. Finally, the Albedo map generated by the Albedo module is applied to complete the 3D face reconstruction for webtoon character images. We generate various results including webtoon characters and stylized animation characters, and compare them with those from previous research to demonstrate our superiority.
Abstract
We propose a deep learning-based 3D face reconstruction technique for webtoon character images. Our method consists of a default module and a detail module. The default module incorporates the Albedo module to calculate the amount of incoming light on the face and generate an albedo map for the input webtoon character image. Additionally, the FLAME module, another component of the default module, generates the basic 3D face shape for the input webtoon character image. Simultaneously, the detail module is applied to extract the detailed features such as facial expressions and depth of the transformed webtoon character image, which differ from real human faces. As a result of the detail module a detail map is generated and combined with the 3D face shape generated by the FLAME module to generate the detail face shape. Finally, the Albedo map generated by the Albedo module is applied to complete the 3D face reconstruction for webtoon character images. We generate various results including webtoon characters and stylized animation characters, and compare them with those from previous research to demonstrate our superiority.
참고문헌 (12)
- Deng, Y., "Accurate 3d face reconstruction with weakly-supervised learning: From single image to image set", Proc. of CVPRW 2019, pp.285~295 (2019)
- Feng, Y., "Learning an animatable detailed 3d face model from in-the-wild images", ACM ToG, 40(4), pp.88:1~88:13 (2021)
- Blanz, V., "A morphable model for the synthesis of 3D faces", Proc. of SIGGRAPH 1999, pp.187~194 (1999)
- S. Romdhani, "Estimating 3d shape and texture using pixel intensity, edges, specular highlights, texture constraints and a prior", IEEE Conference on CVPR 2005, 2, pp.986~993 (2005)
- Cai, H., "Landmark detection and 3D face reconstruction for caricature using a nonlinear parametric model", Graphical Models, 115, pp.101~103 (2021)
- Kemelmacher-Shlizerman, I., "3D face reconstruction from a single image using a single reference face shape", IEEE transactions on PAMI, 33(2), pp.394~405 (2010)
- Zhu, X., "Face alignment across large poses: A 3d solution", Proc. of CVPR 2016, pp.146~155 (2016)
- Tran, L., "On learning 3d face morphable model from in-the-wild images", IEEE transactions on PAMI, 43(4), pp.157~171 (2019)
- Trần, A. T., "Extreme 3d face reconstruction: Seeing through occlusions", Proc. of CVPR 2018, pp.3935~3944 (2018)
- T. Hassner, "Example based 3d reconstruction from sinle 2d images", IEEE Conference on CVPRW 2006, pp.15~ (2006)
- E. Richardson, "3d face reconstruction by learning from synthetic data", International Conference on 3DV, pp.460~469 (2016)
- E. Richardson, "Learning detailed face reconstruction from a single image", CVPR 2017, pp.5553~5562 (2017)