MISC: Multi-Condition Injection and Spatially-Adaptive Compositing for Conditional Person Image Synthesis
Abstract
In this paper, we explore synthesizing person images with multiple conditions for various backgrounds. To this end, we propose a framework named "MISC" for conditional image generation and image compositing. For conditional image generation, we improve the existing condition injection mechanisms by leveraging the inter-condition correlations. For the image compositing, we theoretically prove the weaknesses of the cutting-edge methods, and make it more robust by removing the spatially-invariance constraint, and enabling the bounding mechanism and the spatial adaptability. We show the effectiveness of our method on the Video Instance-level Parsing dataset, and demonstrate the robustness through controllability tests.
Cite
Text
Weng et al. "MISC: Multi-Condition Injection and Spatially-Adaptive Compositing for Conditional Person Image Synthesis." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020. doi:10.1109/CVPR42600.2020.00776Markdown
[Weng et al. "MISC: Multi-Condition Injection and Spatially-Adaptive Compositing for Conditional Person Image Synthesis." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020.](https://mlanthology.org/cvpr/2020/weng2020cvpr-misc/) doi:10.1109/CVPR42600.2020.00776BibTeX
@inproceedings{weng2020cvpr-misc,
title = {{MISC: Multi-Condition Injection and Spatially-Adaptive Compositing for Conditional Person Image Synthesis}},
author = {Weng, Shuchen and Li, Wenbo and Li, Dawei and Jin, Hongxia and Shi, Boxin},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
year = {2020},
doi = {10.1109/CVPR42600.2020.00776},
url = {https://mlanthology.org/cvpr/2020/weng2020cvpr-misc/}
}