Researchers have developed RACANet, a novel framework for RGB-Thermal crowd counting that improves accuracy by explicitly modeling local spatial discrepancies and modality reliability. The method employs a two-stage approach, beginning with cross-modal alignment pretraining and followed by a Local Anchor Fusion Module. This module leverages reliable regions to generate semantic anchors and adaptively redistributes features using attention mechanisms. Experiments on benchmark datasets show RACANet surpasses existing methods. AI
Summary written by gemini-2.5-flash-lite from 3 sources. How we write summaries →
IMPACT Introduces a new method for improving crowd counting accuracy by integrating visible and thermal data.
RANK_REASON Academic paper introducing a new method for RGB-T crowd counting.