Publications by authors named "Aoyan Li"
IEEE Trans Image Process
July 2023
Article Synopsis
- Improving boundary segmentation in semantic segmentation is challenging due to unclear boundary cues in feature spaces of existing methods.
- The proposed conditional boundary loss (CBL) focuses on optimizing each boundary pixel based on its surrounding neighbors, enhancing the accuracy of boundary results while maintaining separation between classes.
- Extensive experiments on datasets like ADE20K, Cityscapes, and Pascal Context show that integrating CBL into popular segmentation networks significantly boosts their performance in terms of mean Intersection over Union (mIoU) and boundary F-score.
