Spatial Pyramid Based Graph Reasoning for Semantic Segmentation

Li, Xia; Yang, Yibo; Zhao, Qijie; Shen, Tiantian; Lin, Zhouchen; Liu, Hong

doi:10.1109/cvpr42600.2020.00897

Cited by 173 publications

(72 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In recent years, graph-based models have been applied to various computer vision tasks including graph matching [36], image segmentation [28], image alignment [48], action recognition [35] and object tracking [47] and many graph neural networks (GNNs) [49,55] have been designed to perform calculations on graph data. Inspired by those works, we extend the graph network block framework [1] to explore the relational reasoning on the graph and develop the classification task of the proposals in object detection.…”

Section: The Proposed Methodsmentioning

confidence: 99%

Object detection by crossing relational reasoning based on graph neural network

Liu

Wang

Feng

et al. 2021

Machine Vision and Applications

View full text Add to dashboard Cite

Utilizing relational representations to facilitate object detection has attracted growing research attention in recent years. However, previous studies mainly focus on relationships within the region proposals or within the label embeddings and pay less attention to the relationships between them. To fill this gap, we propose a novel object detection framework that fully explores the relationships across visual feature space and label embedding space to facilitate the proposal classification in object detection. Specifically, we model the region proposals and class labels into a uniform relation graph, where the extracted proposals and labels are regarded as nodes and each pair of them is associated by an assignment edge, and convert the problem of classifying proposals to the problem of selecting reliable edges from the constructed relation graph. Furthermore, a graph convolutional module is developed to perform relational reasoning on the graph, which finally predicts a label for each assignment edge to indicate whether the classification is reliable or not. The updated relational representations for proposals are used for bounding box regression. Embedding our framework into state-of-the-art baselines, we perform extensive comparison experiments on two public benchmarks, i.e., Pascal VOC and COCO2017. And the experimental results demonstrate the flexibility and effectiveness of the proposed framework.

show abstract

Section: The Proposed Methodsmentioning

confidence: 99%

Object detection by crossing relational reasoning based on graph neural network

Liu

Wang

Feng

et al. 2021

Machine Vision and Applications

View full text Add to dashboard Cite

show abstract

“…In the segmentation task, Li [18] applies graph convolution into the semantic segmentation task use Laplacian to perform reasoning directly to the feature space. Lu [19] generates a neighborhood graph that shows the relationship for each point's neighboring points and then filters the neighborhood graph using Chebyshev polynomials.…”

Section: Related Workmentioning

confidence: 99%

Improvement for Single Image Super-resolution and Image Segmentation by Graph Laplacian Regularizer Based on Differences of Neighboring Pixels

2022

IJIES

View full text Add to dashboard Cite

The neighboring relationship of the pixels is fundamental information in the image. Utilizing neighboring pixel information provides an advantage over using only pixel-to-pixel information. This study presents a regularization term based on the differences of neighboring pixels. We define the differences of neighboring pixels by using the Graph Theory approach. We then propose an effective graph Laplacian regularization to enforce the differences between pixels as the nodes and differences between nodes as edges. In our scheme, Graph Laplacian constructs from the output of the convolutional neural network and ground-truth image. The generated differences matrices from the output and ground-truth are combined as a Laplacian regularization term used as the deep convolutional neural network's new objective function. Experiments show that our scheme successfully captures pixel neighbor relations and improves the performance of the convolutional neural network better than the baseline without a regularization term. Qualitative and quantitative results show that our developed regularizer proved to enhance the boundary structure on image super-resolution and image segmentation tasks, which is achieved Structural Similarity Index (SSIM) of 0.9604 and 0.8263 on Cartoon set and Manga109 datasets and area under the curve (AUC) of 0.9740 and 0.9561 on DRIVE and STARE datasets, respectively.

show abstract

“…CRFs in the form of fully connected networks are separated from the previous CNN module when used as a post-processing module for segmentation results, and the previous CNN module is usually fixed during training of CRFs without information interaction with the features extracted by the CNN. Li et al [71] proposed the feature pairwise conditional random field (FPCRF) based on the graph convolutional network (GCN) [115][116][117][118][119], which is a CRF for pairs of potential pixels with local constraints, incorporating the feature maps extracted by the CNN. FPCRF module can be added to the building segmentation network as a plug-and-play component to improve the segmentation performance of the model without significantly increasing the training and inference time.…”

Section: Boundary Contour Refinementmentioning

confidence: 99%

Deep Learning-Based Building Extraction from Remote Sensing Images: A Comprehensive Review

Luo

Yan

2021

Energies

View full text Add to dashboard Cite

Building extraction from remote sensing (RS) images is a fundamental task for geospatial applications, aiming to obtain morphology, location, and other information about buildings from RS images, which is significant for geographic monitoring and construction of human activity areas. In recent years, deep learning (DL) technology has made remarkable progress and breakthroughs in the field of RS and also become a central and state-of-the-art method for building extraction. This paper provides an overview over the developed DL-based building extraction methods from RS images. Firstly, we describe the DL technologies of this field as well as the loss function over semantic segmentation. Next, a description of important publicly available datasets and evaluation metrics directly related to the problem follows. Then, the main DL methods are reviewed, highlighting contributions and significance in the field. After that, comparative results on several publicly available datasets are given for the described methods, following up with a discussion. Finally, we point out a set of promising future works and draw our conclusions about building extraction based on DL techniques.

show abstract

Spatial Pyramid Based Graph Reasoning for Semantic Segmentation

Cited by 173 publications

References 34 publications

Object detection by crossing relational reasoning based on graph neural network

Object detection by crossing relational reasoning based on graph neural network

Improvement for Single Image Super-resolution and Image Segmentation by Graph Laplacian Regularizer Based on Differences of Neighboring Pixels

Deep Learning-Based Building Extraction from Remote Sensing Images: A Comprehensive Review

Contact Info

Product

Resources

About