Pyramid Attention Network for Semantic Segmentation

By Hanchao Li, Pengfei Xiong, Jie An, Lingxue Wang. 2018

1,011 citations recorded by the source corpus.

A Pyramid Attention Network(PAN) is proposed to exploit the impact of global contextual information in semantic segmentation. Different from most existing works, we combine attention mechanism and spatial pyramid to extract precise dense features for pixel labeling instead of complicated dilated convolution and artificially designed decoder networks. Specifically, we introduce a Feature Pyramid Attention module to perform spatial pyramid attention structure on high-level output and combining global pooling to learn a better feature representation, and a Global Attention Upsample module on each decoder layer to provide global context as a guidance of low-level features to select category localization details. The proposed approach achieves state-of-the-art performance on PASCAL VOC 2012 and Cityscapes benchmarks with a new record of mIoU accuracy 84.0% on PASCAL VOC 2012, while training without COCO dataset.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection (2019, methodology): “We follow Pyramid Attention Networks [20] except removing convolution layers in the original design.”
Deep High-Resolution Representation Learning for Visual Recognition (2019, neutral): “PAN [82] D-ResNet-101 78.”
Hierarchical Multi-Scale Attention for Semantic Segmentation (2020, methodology): “Furthermore, our method is orthogonal to other attention or pyramid pooling methods such as [22, 25, 26, 9, 27, 10, 28] as these methods use single scale image and perform attention to better combine multi-level features for generating high-resolution predictions.”
Image Segmentation Using Deep Learning: A Survey (2020, methodology): “[90] developed a Pyramid Attention Network for semantic segmentation, which exploits global contextual information for semantic segmentation.”
ResUNet++: An Advanced Architecture for Medical Image Segmentation (2019, methodology): “Moreover, it has been employed in semantic segmentation tasks, for example, [31], for pixel-wise prediction.”
Optical Remote Sensing Image Change Detection Based on Attention Mechanism and Image Difference (2021, methodology): “Recently, the attention mechanism has been widely used in natural language processing [38], speech recognition, computer vision [39], [40], and other ﬁelds.”
CPFNet: Context Pyramid Fusion Network for Medical Image Segmentation (2020, background): “From the input image, the encoder can learn global context information including the object’s surroundings and the category characteristics of the object [26], [32].”
Asymmetric Contextual Modulation for Infrared Small Target Detection (2020, methodology): “re-customizing the feature fusion approach: Recent works fuse cross-layer features in a one-directional, topdown manner [18, 32], aiming to select the right lowlevel features based on high-level semantics.”
Attentional Local Contrast Networks for Infrared Small Target Detection (2020, methodology): “Compared with GAU-FPN, the proposed ALCNet achieves a higher detection rate and a lower false-alarm rate simultaneously.”
LoveDA: A Remote Sensing Land-Cover Dataset for Domain Adaptive Semantic Segmentation (2021, positive): “Specifically, the selected networks were: UNet[32], UNet++[54], LinkNet[3], DeepLabV3+[5], PSPNet[52], FCN8S[19], PAN[17], Semantic-FPN[16], HRNet[45], and FarSeg[53].”
DFANet: Deep Feature Aggregation for Real-Time Semantic Segmentation (2019, methodology): “Based on this, PSPNet[34], DeepLab series[7][5][4], PAN[16] apply an additional operation to combine more context information and multi-scale feature representation.”
CHAOS Challenge - Combined (CT-MR) Healthy Abdominal Organ Segmentation (2020, methodology): “nnel maps reveal better performance (DICE: 0.87 0.05) compared to some other models such as U-net (Ronneberger et al., 2015) (DICE: 0.81 0.08), DANet (Fu et al., 2019) (DICE: 0.83 0.10), PAN(ResNet) (Li et al., 2018) (DICE: 0.84 0.06), and UNet Attention (Schlemper et al., 2019) (DICE: 0.85 0.05). D”
ISNet: Shape Matters for Infrared Small Target Detection (2022, methodology): “SK [18] GAU [17] ACM [6] SK GAU TBCNet [41] ALCNet [7] TOAA”
Multi-Scale Self-Guided Attention for Medical Image Segmentation (2019, background): “tion, which have been recently proposed for natural scene [19], [20] and medical image [5], [23], [25] segmentation.”
Deep semantic segmentation of natural and medical images: a review (2019, background): “Deep Layer Cascade (Li et al. (2017)) Architecture PASCAL VOC, Cityscapes 82.”