StepNet: Spatial-temporal Part-aware Network for Sign Language Recognition
Xiaolong ShenZhedong ZhengYi Yang
Xiaolong ShenZhedong ZhengYi Yang
Dec 2022
0被引用
3笔记
读论文，拿好礼活动火爆进行中，iPad、蓝牙耳机、拍立得、键盘鼠标套装等你来拿！
摘要原文
Sign language recognition (SLR) aims to overcome the communication barrierfor the people with deafness or the people with hard hearing. Most existingapproaches can be typically divided into two lines, i.e., Skeleton-based andRGB-based methods, but both the two lines of methods have their limitations.RGB-based approaches usually overlook the fine-grained hand structure, whileSkeleton-based methods do not take the facial expression into account. Inattempts to address both limitations, we propose a new framework namedSpatial-temporal Part-aware network (StepNet), based on RGB parts. As the nameimplies, StepNet consists of two modules: Part-level Spatial Modeling andPart-level Temporal Modeling. Particularly, without using any keypoint-levelannotations, Part-level Spatial Modeling implicitly captures theappearance-based properties, such as hands and faces, in the feature space. Onthe other hand, Part-level Temporal Modeling captures the pertinent propertiesover time by implicitly mining the long-short term context. Extensiveexperiments show that our StepNet, thanks to Spatial-temporal modules, achievescompetitive Top-1 Per-instance accuracy on three widely-used SLR benchmarks,i.e., 56.89% on WLASL, 77.2% on NMFs-CSL, and 77.1% on BOBSL. Moreover, theproposed method is compatible with the optical flow input, and can yield higherperformance if fused. We hope that this work can serve as a preliminary stepfor the people with deafness.
