Artistic Style Novel View Synthesis
Based on A Single Image

Kuan-Wei Tseng, Yao-Chih Lee, Chu-Song Chen

CVPR 2022 Workshop

Method Overview

ArtNV pipeline: view synthesis followed by spatially consistent neural style transfer

ArtNV is a two-stage approach: a view synthesis stage followed by style transfer. Before neural style transfer, we perform occlusion-aware dense matching with optical flow to improve spatial consistency.

VR Application

Stylized stereoscopic scene synthesized from a monocular image for viewing in VR

We model unreachable regions in VR using stylized novel view synthesis as an alternative to full 3D models or 360 images. Users can view a stereoscopic pair synthesized from monocular input. See the related poster paper here.

Citation

If you find our paper or code helpful, please cite:

@InProceedings{Tseng_2022_CVPRW,
    author    = {Tseng, Kuan-Wei and Lee, Yao-Chih and Chen, Chu-Song},
    title     = {Artistic Style Novel View Synthesis Based on A Single Image},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    month     = {June},
    year      = {2022},
}

For VR application, please cite:

@InProceedings{Tseng_2022_SIGGRAPH,
    author    = {Tseng, Kuan-Wei and Huang, Jing-Yuan and Chen, Yang-Sheng and Chen, Chu-Song and Hung, Yi-Ping},
    title     = {Pseudo-3D Scene Modeling for Virtual Reality Using Stylized Novel View Synthesis},
    booktitle = {ACM SIGGRAPH 2022 Posters},
    month     = {July},
    year      = {2022},
}