Structured Landmark Detection via Topology-Adapting Deep Graph Learning

CANet: Context Aware Network for Brain Glioma Segmentation

“…Parisot et al [36] constructed a population graph for degenerative disease classification where each node represents the features from an individual patient. Li et al [37] proposed a topology-adaptive graph neural network for landmark detection with applications on X-ray images. Chao et al [38] introduced a graph neural network based framework to model the relationship between inter lymph nodes for gross tumor volume detection.…”

Learning Laplacians in Chebyshev Graph Convolutional Networks

“…Laplacian -or equivalently graph -inference is generally ill-posed, NP-hard [1][2][3] and most of the existing approaches rely on constraints (including similarity, smoothness, sparsity, band-limitedness, etc.) for a better conditioning [4,5,[73][74][75][76][77][78][79][80][81][82][83][84][85][86]. Particularly in GCNs, early methods [6,49,52] rely on predetermined node-to-node relationships using similarities or the inherent properties of the targeted applications in order to define Laplacian operators.…”

When Liebig's Barrel Meets Facial Landmark Detection: A Practical Model

“…To localize landmarks more precisely and robustly, coordinate regressions can be used in a coarse-to-fine manner through cascaded structures [27,40,31,21,8]. DAG [17] is a recent work that applied graph convolutional networks to landmark detection, enabling the model to dynamically leverage global and local features with graph signals. Despite the superior performance, DAG is computationally inefficient as it relies on high-resolution feature maps and multi-stage regressions.…”

“…Jin et al [11] improved generalization capability of PIPNet by designing a generalizable semi-supervised learning method. DAG [17] performed well on cross-domain evaluation due to the dynamic mechanism from graph message passing.…”