This paper identifies a robustness fading phenomenon where shallow layers spontaneously develop robust representations and flat loss landscapes in early training, yet these properties are not preserved during standard convergence, and proposes a framework that performs strategic interventions on training dynamics to st...
Jian-Gang Yang, Wen-ku Shi, Lunyuan Hu et al.· 0 citations
This work identifies a progressive decay of robust features across network layers and establishes a functional dependency between the prevalence of these features and model performance, and proposes Suppress and Diversify (S\&D), a non-intrusive refinement approach that enhances robustness by dynamically selecting robu...
Jian-Gang Yang, Wen-ku Shi, Xiao-Ran Xu et al.· 0 citations
KNA-SG, a framework for constructing open-vocabulary 3D scene graphs from RGB sequences with explicit keyframe–node associations, is proposed and Experimental results show that KNA-SG outperforms existing methods on open-vocabulary 3D semantic segmentation and 3D object grounding tasks.
Yang Xu, Wen-ku Shi, Jing Xing et al.· Technologies· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.