Paper
8 April 2024 An efficient online hard instance mining algorithm for semantic segmentation combining category and instance information
Ke Li, Qing Song
Author Affiliations +
Proceedings Volume 13090, International Conference on Computer Application and Information Security (ICCAIS 2023); 130902E (2024) https://doi.org/10.1117/12.3026344
Event: International Conference on Computer Application and Information Security (ICCAIS 2023), 2023, Wuhan, China
Abstract
Semantic image segmentation has been greatly improved since fully convolution network (FCN) is proposed. Recent work has further optimized the results of semantic segmentation, especially DeepLab, RefineNet and PSPNet. In this paper, we exploit Online Hard Instance Mining (OHIM) algorithm that enables the networks to combine category and instance information for semantic segmentation. Online and automatic mining of the hard instance can make training procedure purposive and efficient. Implementation of the proposed algorithm is simple yet effective, and can be combined with any previous semantic segmentation networks. It is worth mentioning that OHIM does not increase the computation cost in testing phase. We evaluate the proposed algorithm on PASCAL VOC 2012 and Cityscapes dataset, yielding a non-negligible improvement for semantic segmentation.
(2024) Published by SPIE. Downloading of the abstract is permitted for personal use only.
Ke Li and Qing Song "An efficient online hard instance mining algorithm for semantic segmentation combining category and instance information", Proc. SPIE 13090, International Conference on Computer Application and Information Security (ICCAIS 2023), 130902E (8 April 2024); https://doi.org/10.1117/12.3026344
Advertisement
Advertisement
RIGHTS & PERMISSIONS
Get copyright permission  Get copyright permission on Copyright Marketplace
KEYWORDS
Image segmentation

Semantics

Education and training

Mining

Feature extraction

Object detection

Classification systems

Back to Top