Positional Encodings Anchor Spatial Structure in Vision Transformers: A Geometric Perspective on Robustness 文章

ArXiv CS.CV2026-06-02NEWSen作者: Mahmoud Mannes
查看原文 →
关系图谱

Positional Encodings Anchor Spatial Structure in Vision Transformers: A Geometric Perspective on Robustness · 相关技术

相关技术

rotarysinusoidal PElearned absoluteSSDCrotary positional embeddingsinusoidal positional embeddinglearned absolute positional embeddingSpatial Similarity Distance Correlationvision transformersPositional embeddings