Researchers have developed FocusOVD, a novel open-vocabulary object detection system designed to improve the detection of small objects. FocusOVD incorporates three key components: a Reweighted Cross-scale Feature Fusion module for balancing feature levels, a Residual Pinwheel-shaped Gated Convolution module to enhance geometric details, and a Residual Convolutional Gated Linear Unit module to refine region-level discriminability. Experiments on COCO-OVD and LVIS-OVD datasets show FocusOVD outperforming existing methods, achieving a novel-category AP of 36.3% on COCO-OVD and 33.7% overall AP on LVIS-OVD. AI
IMPACT Introduces new techniques for improving small-object detection in computer vision models.
RANK_REASON Academic paper detailing a new method for object detection. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →