NTU RGB+D 120: A Large-Scale Benchmark for 3D Human Activity Understanding
PulseAugur coverage of NTU RGB+D 120: A Large-Scale Benchmark for 3D Human Activity Understanding — every cluster mentioning NTU RGB+D 120: A Large-Scale Benchmark for 3D Human Activity Understanding across labs, papers, and developer communities, ranked by signal.
3 day(s) with sentiment data
-
New DyGCN Model Enhances Skeleton-Based Action Recognition
Researchers have developed a new model called DyGCN, a Geometry-Constrained Dynamic Hypergraph Convolutional Network, designed to improve skeleton-based action recognition in computer vision. Unlike traditional methods …
-
StickFigure: Pose-Only IoT Cameras for Privacy-Preserving Edge Vision
Researchers have developed StickFigure, a novel pose-only IoT camera system designed to enhance privacy in edge vision applications. This system processes raw frames locally, exporting only skeleton streams for human ac…
-
New Neuro-Symbolic Graph Transformer enhances human-robot intention recognition
Researchers have developed a Neuro-Symbolic Graph Transformer (NSGT) designed to improve hierarchical intention recognition in human-robot collaboration. This new architecture integrates a Human Reasoning Graph (HR-G) i…
-
HuC-VideoMAE uses synthetic data for ethical human-centric video pretraining
Researchers have developed HuC-VideoMAE, a novel approach to pretraining video transformers using synthetic human-motion data. This method addresses ethical concerns associated with using real-world videos by employing …
-
New Neuro-Symbolic Model Anticipates Human Intentions with High Accuracy
Researchers have developed a Neuro-Symbolic Hierarchical Intention Anticipation (HPD) system designed to predict human goals from partially observed multimodal data. The HPD attaches to a recognition encoder and forecas…
-
NeuroPath: Brain-Inspired Dual-Pathway Networks Enhance Action Recognition
Researchers have developed NeuroPath, a novel dual-pathway graph convolutional network designed for skeleton-based action recognition. This architecture, inspired by the human brain's ventral and dorsal pathways, separa…
-
New RGB-only action recognition model for edge devices unveiled
Researchers have developed a new, resource-efficient RGB-only action recognition network designed for deployment on edge devices. This model, which incorporates several architectural innovations like temporal shift and …
-
New framework KineMIC enhances few-shot action synthesis for HAR
Researchers have developed KineMIC, a novel transfer learning framework designed to improve few-shot action synthesis for skeletal-based Human Activity Recognition (HAR). This method adapts text-to-motion diffusion mode…
-
New hypergraph framework boosts action recognition with partial skeleton data
Researchers have developed PartialVisGraph, a new hypergraph framework designed to improve skeleton-based action recognition in scenarios with limited field-of-view. This approach addresses the performance degradation t…
-
New Attack Method Preserves Motion Quality in Skeleton Action Recognition
Researchers have developed a new method for creating imperceptible adversarial attacks on skeleton-based human action recognition systems. This approach aims to preserve the naturalness of motion data, unlike previous m…
-
BioVid generates videos with natural behavioral timing
Researchers have developed BioVid, a novel autoregressive video generation framework that learns to generate videos reflecting the natural temporal structure of biological behaviors. Unlike existing methods that rely on…
-
New framework uses hyperbolic geometry for skeleton-based action recognition
Researchers have developed SkelHCC, a new framework for one-shot action recognition using skeleton data. This approach utilizes hyperbolic geometry to better model the hierarchical structure of human motion and align it…
-
New framework generates counterfactual explanations for video classifiers
Researchers have developed a new framework called Back To The Feature (BTTF) to generate counterfactual explanations for video classifiers. Unlike previous methods focused on images, BTTF addresses the unique challenges…
-
New method improves zero-shot action recognition with multi-view motion and text
Researchers have developed a new method for zero-shot action recognition that improves robustness to domain changes. The approach combines motion data from multiple camera viewpoints with textual descriptions of actions…
-
Diffusion models enhanced for zero-shot skeleton action recognition
Researchers have developed a new method called Frequency-Aware Diffusion for Skeleton-Text Matching (FDSM) to improve zero-shot skeleton action recognition. This approach addresses the spectral bias in diffusion models …
-
New contrastive learning method improves skeleton-based action localization
Researchers have developed a new self-supervised pretraining method called Skeleton-Snippet Contrastive Learning for improving temporal action localization in skeleton-based data. This approach uses a snippet discrimina…