Return
SALA: Semantic alignment and localization alignment for visual grounding
DOI:10.1016/j.neucom.2025.131265.png)
Abstract
En 中文
• Combining semantic alignment and localization alignment (SALA) to solve two prevalent issues in visual grounding. • Semantic alignment captures consistent semantics across and within modalities through coarse-grained guidance. • Localization alignment enhances foreground while suppressing background through fine-grained supervision. • Experimental results demonstrate SALA’s superior grounding performance.
Keywords:
semantic alignment
localization alignment
visual grounding
multimodal learning
fine-grained supervision
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W
Organization
No organization information available

