3 Jan 2018 | Shifeng Zhang, Longyin Wen, Xiao Bian, Zhen Lei, Stan Z. Li
This paper proposes RefineDet, a single-shot object detection framework that combines the high accuracy of two-stage detectors with the efficiency of one-stage detectors. RefineDet consists of two interconnected modules: the anchor refinement module (ARM) and the object detection module (ODM). The ARM filters out negative anchors to reduce search space for the classifier and coarsely adjusts anchor locations and sizes for better initialization. The ODM uses refined anchors to improve regression and predict multi-class labels. A transfer connection block (TCB) transfers features from the ARM to the ODM for object detection. The multi-task loss function enables end-to-end training. RefineDet achieves state-of-the-art detection accuracy on PASCAL VOC 2007, PASCAL VOC 2012, and MS COCO with high efficiency. It outperforms existing methods in accuracy and runs at 40.2 FPS and 24.1 FPS on NVIDIA Titan X GPUs. RefineDet is efficient, accurate, and effective in handling various object scales and aspect ratios. It achieves 85.8% mAP on VOC 2007, 86.8% mAP on VOC 2012, and 41.8% AP on MS COCO test-dev. The method is validated through extensive experiments and shows strong performance in object detection tasks.This paper proposes RefineDet, a single-shot object detection framework that combines the high accuracy of two-stage detectors with the efficiency of one-stage detectors. RefineDet consists of two interconnected modules: the anchor refinement module (ARM) and the object detection module (ODM). The ARM filters out negative anchors to reduce search space for the classifier and coarsely adjusts anchor locations and sizes for better initialization. The ODM uses refined anchors to improve regression and predict multi-class labels. A transfer connection block (TCB) transfers features from the ARM to the ODM for object detection. The multi-task loss function enables end-to-end training. RefineDet achieves state-of-the-art detection accuracy on PASCAL VOC 2007, PASCAL VOC 2012, and MS COCO with high efficiency. It outperforms existing methods in accuracy and runs at 40.2 FPS and 24.1 FPS on NVIDIA Titan X GPUs. RefineDet is efficient, accurate, and effective in handling various object scales and aspect ratios. It achieves 85.8% mAP on VOC 2007, 86.8% mAP on VOC 2012, and 41.8% AP on MS COCO test-dev. The method is validated through extensive experiments and shows strong performance in object detection tasks.