Pixel Contrastive-Consistent Semi-Supervised Semantic Segmentation
Abstract
We present a novel semi-supervised semantic segmentation method which jointly achieves two desiderata of segmentation model regularities: the label-space consistency property between image augmentations and the feature-space contrastive property among different pixels. We leverage the pixel-level L2 loss and the pixel contrastive loss for the two purposes respectively. To address the computational efficiency issue and the false negative noise issue involved in the pixel contrastive loss, we further introduce and investigate several negative sampling techniques. Extensive experiments demonstrate the state-of-the-art performance of our method (PC2Seg) with the DeepLab-v3+ architecture, in several challenging semi-supervised settings derived from the VOC, Cityscapes, and COCO datasets.
Cite
Text
Zhong et al. "Pixel Contrastive-Consistent Semi-Supervised Semantic Segmentation." International Conference on Computer Vision, 2021. doi:10.1109/ICCV48922.2021.00718Markdown
[Zhong et al. "Pixel Contrastive-Consistent Semi-Supervised Semantic Segmentation." International Conference on Computer Vision, 2021.](https://mlanthology.org/iccv/2021/zhong2021iccv-pixel/) doi:10.1109/ICCV48922.2021.00718BibTeX
@inproceedings{zhong2021iccv-pixel,
title = {{Pixel Contrastive-Consistent Semi-Supervised Semantic Segmentation}},
author = {Zhong, Yuanyi and Yuan, Bodi and Wu, Hong and Yuan, Zhiqiang and Peng, Jian and Wang, Yu-Xiong},
booktitle = {International Conference on Computer Vision},
year = {2021},
pages = {7273-7282},
doi = {10.1109/ICCV48922.2021.00718},
url = {https://mlanthology.org/iccv/2021/zhong2021iccv-pixel/}
}