Answers are grounded in this paper and cite their sources.
To address the challenges in detecting abrasion-resistant color fastness samples including limited sample instances, non-uniform shapes, and insufficiently distinct texture variations that compromise localization accuracy this paper optimizes the detection framework through the integration of three key strategies: Global Attention Mechanism (GAM), Dynamic Sampling (DySample), and Adaptively Spatial Feature Fusion (ASFF), thereby enhancing detection accuracy and efficiency. Initially, Mosaic data augmentation is implemented to enrich dataset diversity and improve model robustness. Subsequently, the GAM attention mechanism is embedded into the backbone network to enhance target feature extraction capabilities. DySample replaces conventional upsampling methods in the neck network to achieve more effective feature reconstruction. Finally, the ASFF module is integrated into the Detect module within the head network to enable adaptively spatial weight learning for multi-scale feature map fusion. Compared with baseline algorithms, the improved framework demonstrates performance gains of 1.2% in Precision, 3.0% in Recall, 1.2% in mAP@0.5, and 13.5% in mAP@0.5:0.95. Experimental results validate the effectiveness of the proposed method, which maintains satisfactory performance across additional datasets, demonstrating strong robustness and superior generalization capability.
No takes yet. Share an insight, caveat, or question.
A 2025 study studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: