arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 1702.02138cs.CV
收藏

快速RCNN的实现及区域抽样研究

An Implementation of Faster RCNN with Study for Region Sampling

Carnegie Mellon University(卡内基梅隆大学)

Xinlei Chen, Abhinav Gupta

摘要:

我们将Faster RCNN框架的联合训练方案从Caffe调整为TensorFlow,作为目标检测的基线实现。我们的代码是公开的。本报告记录了对原始管道进行的简化,以及PASCAL VOC 2007和COCO 2014消融分析的依据。我们进一步研究了非最大抑制(NMS)在选择区域分类的感兴趣区域(ROI)中的作用,发现对小区域的偏置采样有助于性能,并且在充分收敛时可以实现基于NMS的采样的平均mAP。

英文摘要:

We adapted the join-training scheme of Faster RCNN framework from Caffe to TensorFlow as a baseline implementation for object detection. Our code is made publicly available. This report documents the simplifications made to the original pipeline, with justifications from ablation analysis on both PASCAL VOC 2007 and COCO 2014. We further investigated the role of non-maximal suppression (NMS) in selecting regions-of-interest (RoIs) for region classification, and found that a biased sampling toward small regions helps performance and can achieve on-par mAP to NMS-based sampling when converged sufficiently.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑