PYSKL: Towards Good Practices for Skeleton Action Recognition

Duan, Haodong; Wang, Jiaqi; Chen, Kai; Lin, Dahua

doi:10.1145/3503161.3548546

Cited by 87 publications

(31 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Theoretically, GCDAE can be integrated with any skeleton based action recognition model to improve its robustness to noisy data. We combine it with five SOTA recognition models mentioned in Section 2.1, that is, EfficientGCN [35], ST‐GCN++ [36], CTR‐GCN [21], AAGCN [13] and MS‐G3D [14] for experiments. As different methods utilize high‐order features in different ways, we use only the first‐order joint stream for simplicity and fair comparison.…”

Section: Methodsmentioning

confidence: 99%

“…We retrain the EfficientGCN‐B0 as their data preprocessing on CS benchmark is different from ours. For other four backbone models, pretrained weights provided by PYSKL [36] are directly used. Comparison of these backbones are shown in Table 1.…”

Section: Methodsmentioning

confidence: 99%

“…EfficientGCN [35] adopts separable graph convolutions and an early fusion strategy to reduce parameters. Duan et al [36] propose a simple yet effective model ST-GCN++ without any complicated attention mechanism and present PYSKL, an open-source toolbox which implements several SOTA recognition backbones under a unified framework. In this work, we compare robustness of some of these backbones towards noisy skeletons and show how GCDAE helps improve them.…”

Section: Skeleton-based Action Recognitionmentioning

confidence: 99%

“…mentioned in Section 2.1, that is, EfficientGCN [35], ST-GCN++ [36], CTR-GCN [21], AAGCN [13] and MS-G3D [14] for experiments. As different methods utilize high-order features in different ways, we use only the first-order joint stream for simplicity and fair comparison.…”

Section: Backbone Modelsmentioning

confidence: 99%

See 3 more Smart Citations

Overcomplete graph convolutional denoising autoencoder for noisy skeleton action recognition

Guo,

Ji,

Shan

2023

IET Image Processing

View full text Add to dashboard Cite

Current skeleton‐based action recognition methods usually assume the input skeleton is complete and noise‐free. However, it is inevitable that the captured skeletons are incomplete due to occlusions or noisy due to changes in the environment. When dealing with these data, even State Of The Art (SOTA) recognition backbones experience significant degradation in recognition accuracy. Though a few methods have been proposed to address this issue, they still lack flexibility, efficiency and interpretability. In this work, an overcomplete Graph Convolutional Denoising Autoencoder (GCDAE) is proposed which can act as a flexible preprocessing module for pretrained recognition backbones and improve their robustness. Taking advantages of the overcomplete and fully graph convolutional structure, GCDAE is able to rectify noisy joints while keeping information of unspoiled details efficiently. On two large scale skeleton datasets NTU RGB+D 60 and 120, the introducing of GCDAE brings significant robustness improvements to SOTA backbones towards different types of noises.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Skeleton-based Action Recognitionmentioning

confidence: 99%

Section: Backbone Modelsmentioning

confidence: 99%

See 2 more Smart Citations

Overcomplete graph convolutional denoising autoencoder for noisy skeleton action recognition

Guo,

Ji,

Shan

2023

IET Image Processing

View full text Add to dashboard Cite

show abstract

“…The key to processing unorganized motion datasets is label acquisition. We demonstrate that an action recognition network [Duan et al 2022] pretrained on NTURGB+D 120 dataset [Liu et al 2020] can produce reliable skill labels. We test with raw unstructured motion clips from CMU Mocap dataset [CMU 2002] by finetuning the network on the aforementioned Composite Skills dataset, where we set the learning rate to be 0.1 with momentum being 0.9 and weight decay being 5×10 −4 .…”

Section: A2 Skill Labelmentioning

confidence: 99%

C·ASE: Learning Conditional Adversarial Skill Embeddings for Physics-based Characters

Dou,

Chen,

Fan

et al. 2023

SIGGRAPH Asia 2023 Conference Papers

View full text Add to dashboard Cite

Figure 1: Our framework enables physically simulated characters to master highly varied and extensive skills with high efficiency and effectiveness. Notably, it offers an explicit control handle for directly specifying the desired skill from a diverse and extensive set of skills. Here, a character is instructed by the user to perform a sequence of skills, including kick, jump attack, sword bash, shield bash, and finally, roaring.

show abstract

Continuous Action Recognition in Manufacturing Contexts by Deep Graph Convolutional Networks

Maselli,

Marani,

Cicirelli

et al. 2024

Lecture Notes in Networks and Systems

View full text Add to dashboard Cite

PYSKL: Towards Good Practices for Skeleton Action Recognition

Cited by 87 publications

References 22 publications

Overcomplete graph convolutional denoising autoencoder for noisy skeleton action recognition

Overcomplete graph convolutional denoising autoencoder for noisy skeleton action recognition

C·ASE: Learning Conditional Adversarial Skill Embeddings for Physics-based Characters

Continuous Action Recognition in Manufacturing Contexts by Deep Graph Convolutional Networks

Contact Info

Product

Resources

About