FASSD:基于特征扩展融合的SSD优化模型

FASSD:SSD with feature fusion atrous

SSD(Single Shot MultiBox Detector)是目标检测领域内重要的算法之一,在检测精度与速度上都取得了较好的结果。针对SSD单层感受野不足,特征分辨能力差的问题,提出了FASSD优化模型(SSD with Feature Fusion Atrous)。FASSD模型通过对特征金字塔与预测器的改造,使模型在扩展感受野的同时能更充分的学习特征金字塔所带来的多尺度特征。在PASCAL VOC数据集上,FASSD取得了82.5%的mAP,相比于原始SSD提高了7.6%。

SSD(Single Shot Multibox Detector)is one of the important algorithms in the field of object detection,which has achieved better results in terms of detection accuracy and speed.The FASSD optimization model(SSD with Feature Fusion Atrous)is proposed to address the problem of inadequate single-layer perceptual field and poor feature discrimination ability of SSD.On the PASCAL VOC dataset,FASSD achieves 82.5%mAP,which is a 7.6%improvement compared to the original SSD.