Geometric Scattering Attention Networks

Min, Yimeng; Wenkel, Frederik; Wolf, Guy

doi:10.1109/icassp39728.2021.9414557

Cited by 11 publications

(10 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Some studies have proposed to transform the joint space of the human body to enhance different structural features and perform signal filtering or aggregate vertex information based on graph Laplacian feature decomposition [15]. Graph scattering transforms, by changing the bandwidth size [16], using different graph signal filters [17] to obtain the expected spectrum, increasing the richness of the graph. On this basis, the wavelet diffusion method [18], and parametric feature learner [19] have good performance.…”

Section: B Graph Structurementioning

confidence: 99%

A Progressive Decoding Strategy for Action Recognition

Zhao,

Li,

Liu

et al. 2023

IEEE Access

View full text Add to dashboard Cite

In recent years, significant progress has been made in modeling temporal sequences and spatial structures in skeleton-based human action recognition. However, existing methods rely on explicit modeling of the inherent structure of the human body, which may result in reduced joint saliency and poor interpretability due to the sparsity of skeleton data and the relative smoothness of convolutions. This paper proposes a feature matching method based on the progressive decoding strategy. As human movement is a chain process, the strategy progressively decodes human pose features from the center to the periphery, using multi-level graph filters to obtain multi-frequency hierarchical graph features. Then the adaptive convolution kernels are constructed to match the local similarities between graph features. Self-similarity of the query set and the mutual similarity between the query set and the support set samples are calculated to analyze the entire posture of the human body, and similar skeletal features are distinguished according to the similarity of the node spectrum to differentiate between different action categories. Through experimental verification of two public data sets, the proposed method has better recognition accuracy and generalization of small sample behavior. The experiments show that the proposed method outperforms the existing methods on NTU RGB + D and Human36M Dataset.

show abstract

Section: B Graph Structurementioning

confidence: 99%

A Progressive Decoding Strategy for Action Recognition

Zhao,

Li,

Liu

et al. 2023

IEEE Access

View full text Add to dashboard Cite

show abstract

“…[13,66] develop diffusion wavelets. [44,45] integrate designed scattering filters and parameterized feature learners. [47] expands GSTs on the spatio-temporal domain.…”

Section: Graph Representation Learningmentioning

confidence: 99%

“…[43]. 15 45. 26.88 43.51 49.23 88.73 20.17 32.98 42.75 44.65 60.57 20.52 40.58 75.38 90.36 153.12 26.85 48.07 93.50 108.90 162.84 DMGNN [35] 15.57 28.72 59.01 73.05 138.62 5.03 9.28 20.21 26.23 52.04 10.21 20.90 41.55 52.28 111.23 31.97 54.32 96.66 119.92 224.63 Traj-GCN [41] 11.68 21.26 40.99 50.78 97.99 3.33 6.25 13.58 17.98 54.00 6.92 13.69 30.30 39.97 114.16 17.18 32.37 60.12 72.55 127.41 MST-GCN [9] 10.28 18.94 37.68 47.03 86.96 3.03 5.68 12.35 16.26 47.91 5.92 12.09 28.36 38.04 111.04 14.99 28.66 55.86 69.05 124.79 STSGCN [53] 12.56 23.04 41.92 50.33 94.17 4.72 6.69 14.53 17.88 49.52 6.41 12.38 29.05 38.86 109.42 17.52 31.48 58.74 72.06 127.40 SPGSN 10.24 18.54 38.22 48.68 89.58 2.91 5.25 11.31 15.01 47.31 5.52 11.16 25.48 37.06 108.14 14.93 28.16 56.72 71.16 125.20 .…”

mentioning

confidence: 99%

Skeleton Graph Scattering Networks for 3D Skeleton-based Human Motion Prediction

Chen

Liu

et al. 2021

2021 IEEE/CVF International Conference on Computer Vision Workshops (ICCVW)

View full text Add to dashboard Cite

Graph convolutional network based methods that model the body-joints' relations, have recently shown great promise in 3D skeletonbased human motion prediction. However, these methods have two critical issues: first, deep graph convolutions filter features within only limited graph spectrums, losing sufficient information in the full band; second, using a single graph to model the whole body underestimates the diverse patterns on various body-parts. To address the first issue, we propose adaptive graph scattering, which leverages multiple trainable bandpass graph filters to decompose pose features into richer graph spectrum bands. To address the second issue, body-parts are modeled separately to learn diverse dynamics, which enables finer feature extraction along the spatial dimensions. Integrating the above two designs, we propose a novel skeleton-parted graph scattering network (SPGSN). The cores of the model are cascaded multi-part graph scattering blocks (MPGSBs), building adaptive graph scattering on diverse body-parts, as well as fusing the decomposed features based on the inferred spectrum importance and body-part interactions. Extensive experiments have shown that SPGSN outperforms state-of-the-art methods by remarkable margins of 13.8%, 9.3% and 2.7% in terms of 3D mean per joint position error (MPJPE) on Human3.6M, CMU Mocap and 3DPW datasets, respectively 1 .

show abstract

“…Similarly, these works prevent the over-smoothing issue by replacing the aggregation strategy of the entire network, lacking an understanding of node-specific differentiations. There are also some works that make efforts to theoretically propose methods for training deep GNNs (Xu et al, 2021;Min et al, 2020). However, these works are limited to specific types of GNNs, lacking generalizability and practical significance.…”

Section: Details Of Datasets and Backbonesmentioning

confidence: 99%

Exploration and Practice of the Summer International Scientific Research Training on Chemistry for the Top Talents of USTC

Wang¹,

Guang-shui²,

Li³

et al. 2019

University Chemistry

View full text Add to dashboard Cite

Despite Graph Neural Networks (GNNs) demonstrating considerable promise in graph representation learning tasks, GNNs predominantly face significant issues with over-fitting and over-smoothing as they go deeper as models of computer vision (CV) realm. Given that the potency of numerous CV and language models is attributable to that support reliably training very deep architectures, we conduct a systematic study of deeper GNN research trajectories. Our findings indicate that the current success of deep GNNs primarily stems from (I) the adoption of innovations from CNNs, such as residual/skip connections, or (II) the tailor-made aggregation algorithms like DropEdge. However, these algorithms often lack intrinsic interpretability and indiscriminately treat all nodes within a given layer in a similar manner, thereby failing to capture the nuanced differences among various nodes. In this paper, we introduce the Snowflake Hypothesis -a novel paradigm underpinning the concept of "one node, one receptive field". The hypothesis draws inspiration from the unique and individualistic patterns of each snowflake, proposing a corresponding uniqueness in the receptive fields of nodes in the GNNs. We employ the simplest gradient and node-level cosine distance as guiding principles to regulate the aggregation depth for each node, and conduct comprehensive experiments including: (1) different training scheme; (2) various shallow and deep GNN backbones, especially on deep frameworks such as JKNet, ResGCN, PairNorm etc. (3) various numbers of layers (8, 16, 32, 64) on multiple benchmarks (six graphs including dense graphs with millions of nodes); (4) compare with different aggregation strategies. The observational results demonstrate that our framework can serve as a universal operator for a range of tasks, and it displays tremendous potential on deep GNNs. It can be applied to various GNN frameworks, enhancing its effectiveness when operating in-depth, and guiding the selection of the optimal network depth in an explainable and generalizable way.

show abstract

Geometric Scattering Attention Networks

Cited by 11 publications

References 12 publications

A Progressive Decoding Strategy for Action Recognition

A Progressive Decoding Strategy for Action Recognition

Skeleton Graph Scattering Networks for 3D Skeleton-based Human Motion Prediction

Exploration and Practice of the Summer International Scientific Research Training on Chemistry for the Top Talents of USTC

Contact Info

Product

Resources

About