.png){kind=icon}
返回
Position attention optimized deep semantic segmentation
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-023-16022-4.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Semantic segmentation can be applied in various fields of computer vision such as scene understanding. In order to assist intelligent machines to detect and recognize the objects, an optimal model to realize accurate semantic segmentation is urgently needed. However, labelling semantic objects in a complex scene is a challenging task due to the lack of more discriminative features. In this work, we solve the semantic segmentation task by proposing a Position Attention Optimized Deep Semantic Segmentation (PADSS) framework in the popular Discriminative Feature Network (DFN), which comprises the Smooth Network (SN) and Border Network (BN). Considering that different levels of feature possess distinct discriminative power, our PADSS framework embeds two types of attention blocks in the smooth network to select more discriminative features. Specifically, in the low stage of the network, we introduce the Position Attention Block (PAB) to obtain more effective features and strengthen the dependencies among similar features. Moreover, in the high stage of the network, the Channel Attention Block (CAB) is leveraged to gain richer context information. We incorporate PAB with CAB to make the prediction of semantic labels more accurate. The proposed model is validated on the two public benchmark datasets and the experimental results show that our PADSS outperforms its competitors and achieves the state-of-the-art (SOTA) performance 87.8% Mean IOU on PASCAL VOC 2012 and 80.8% Mean IOU on Cityscapes dataset. Results suggest that the proposed PADSS is robust enough to learn a variety of discriminative features over various semantic segmentation tasks.
Keyword:
Semantic segmentation
PADSS
Position Attention Block
Channel Attention Block
期刊
IF:
3
论文数:
2.0W
被引数:
3.2W
机构
引用论文
In vitro and in vivo binding of neuroactive steroids to the sigma‐1 receptor as measured with the positron emission tomography radioligand [18F]FPS.png)
Synapse
SynapseIF0
PREAI
分享收藏Recent Development of Dual-Dictionary Learning Approach in Medical Image Analysis and Reconstruction
OAAI
OAAI分享收藏Generation of hydroxyl radicals by urban suspended particulate air matter. The role of iron ions城市悬浮颗粒物产生羟基自由基。铁离子的作用:
PREAI
分享收藏
查看更多内容