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

Li, Maosen; Chen, Siheng; Liu, Zihui; Zhang, Zijing; Xie, Lingxi; Zhang, Ya

doi:10.1109/iccvw54120.2021.00101

Cited by 24 publications

(3 citation statements)

References 49 publications

(113 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Skeleton-based HAP and HAR use neural networks methods to extract the skeleton joints position in the 3D scene from recorded RGB-D images. Some works [26][27][28] used RGB-D data for HAP and HAR, others used only the depth data [29][30][31] or the skeleton data [1,32]. Vision-based systems are advantageous in the sense that they are not invasive.…”

Section: Vision-based Devices (Camera)mentioning

confidence: 99%

Predicting Human Activity – State of the Art

Awong¹,

Zielińska²

2023

PAR

View full text Add to dashboard Cite

Predicting human actions is a very actual research field. Artificial intelligence methods are commonly used here. They enable early recognition and classification of human activities. Such knowledge is extremely needed in the work on robots and other interactive systems that communicate and cooperate with people. This ensures early reactions of such devices and proper planning of their future actions. However, due to the complexity of human actions, predicting them is a difficult task. In this article, we review state-of-the-art methods and summarize recent advances in predicting human activity. We focus in particular on four approaches using machine learning methods, namely methods using: artificial neural networks, support vector machines, probabilistic models and decision trees. We discuss the advantages and disadvantages of these approaches, as well as current challenges related to predicting human activity. In addition, we describe the types of sensors and data sets commonly used in research on predicting and recognizing human actions. We analyze the quality of the methods used, based on the prediction accuracy reported in scientific articles. We describe the importance of the data type and the parameters of machine learning models. Finally, we summarize the latest research trends. The article is intended to help in choosing the right method of predicting human activity, along with an indication of the tools and resources necessary to effectively achieve this goal.

show abstract

Section: Vision-based Devices (Camera)mentioning

confidence: 99%

Predicting Human Activity – State of the Art

Awong¹,

Zielińska²

2023

PAR

View full text Add to dashboard Cite

show abstract

“…Li et al 10 presents a method to extract the structural features in different scale of human body component, and organizes a network forecasting the human pose based on GRU. In Li et al 16 , the problem of over-smoothing is solved by extracting comprehensive pose information from multiple spectral bands using graph scattering. Observing the repetition of human behaviour, Wei 17 utilize motion attention to capture the similarity from context.…”

Section: Related Workmentioning

confidence: 99%

Reciprocal collaboration network for 3D skeleton-based human motion prediction

Zhang

Cao

2022

Third International Conference on Computer Science and Communication Technology (ICCSCT 2022)

View full text Add to dashboard Cite

Noticing that the human motion in both the positive and inverted time sequence are predictable, we design a reciprocal network which contains a forward and a backward prediction network to learn this information respectively, and then train the two networks in a collaborative and iterative way with the guidance of consistency constraint. Based on the reciprocal network, we establish an error compensation network to approximate the error of forward prediction network for making the forecast closer to the real pose. Extensive experiments show that our approach outperforms most recent methods in both short and long term motion predictions on Human 3.6M and CMU Mocap.

show abstract

“…Furthermore, since all wavelet coefficients in GSTs are constructed analytically, GST is computationally more efficient and requires less training data compared to GNNs when combined with the same trainable classifiers. As a result, GSTs are frequently used in practice Min et al (2020); Bianchi et al (2021); Li et al (2021); Pan et al (2021), especially in the limited training data regime. Furthermore, it also allows for deriving theoretical analysis on nonlinear models for graph embedding, and the related conclusions could potentially shed some light on the design of GNNs that are more suitable for data removals.…”

Section: Introductionmentioning

confidence: 99%

Unlearning Nonlinear Graph Classifiers in the Limited Training Data Regime

Pan¹,

Chien²,

Milenković³

2022

Preprint

View full text Add to dashboard Cite

As the demand for user privacy grows, controlled data removal (machine unlearning) is becoming an important feature of machine learning models for data-sensitive Web applications such as social networks and recommender systems. Nevertheless, at this point it is still largely unknown how to perform efficient machine unlearning of graph neural networks (GNNs); this is especially the case when the number of training samples is small, in which case unlearning can seriously compromise the performance of the model. To address this issue, we initiate the study of unlearning the Graph Scattering Transform (GST), a mathematical framework that is efficient, provably stable under feature or graph topology perturbations, and offers graph classification performance comparable to that of GNNs. Our main contribution is the first known nonlinear approximate graph unlearning method based on GSTs. Our second contribution is a theoretical analysis of the computational complexity of the proposed unlearning mechanism, which is hard to replicate for deep neural networks. Our third contribution are extensive simulation results which show that, compared to complete retraining of GNNs after each removal request, the new GST-based approach offers, on average, a 10.38x speed-up and leads to a 2.6% increase in test accuracy during unlearning of 90 out of 100 training graphs from the IMDB dataset (10% training ratio). * Equal contribution.1 Existing laws on user data privacy include the European Union General Data Protection Regulation (GDPR), the California Consumer Privacy Act (CCPA), the Canadian Consumer Privacy Protection Act (CPPA), Brazilian General Data Protection Law (LGPD), and many others.

show abstract

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

Cited by 24 publications

References 49 publications

Predicting Human Activity – State of the Art

Predicting Human Activity – State of the Art

Reciprocal collaboration network for 3D skeleton-based human motion prediction

Unlearning Nonlinear Graph Classifiers in the Limited Training Data Regime

Contact Info

Product

Resources

About