摘 要：现阶段深度学习作为一种实现机器学习的技术，在分析模型问题和评估模型的方法上基本一致。文章从评估模型的角度，以混淆矩阵为基础，通过常用的 Accuracy，Precision 以及 Recall 等衡量模型的预测能力。研究结合深度学习近几年的竞赛任务分析样本均衡与非均衡下几种评估模型方法的差异，从几种评估指标之间的联系讨论 P-R 曲线评估模型之间的相关性，以及 P-R 曲线在目标检测任务中作为评估模型方法的合理性。
关键词：评估标准；Accuracy；P-R 曲线；mAP
中图分类号：TP181；TP311.1 文献标识码：A 文章编号：2096-4706（2020）04-0023-03
Research on P-R Curve and Model Evaluation
ZHANG Chao
（Nanjing University of Information Science and Technology，Nanjing 210044，China）
Abstract：At present，deep learning，as a technology to realize machine learning，is basically consistent in analyzing model problems and evaluating model methods. From the perspective of the evaluation model，based on the confusion matrix，this paper measures the prediction ability of the model from the commonly used Accuracy，Precision and Recall. This paper analyzes the differences of several evaluation models under the condition of sample equilibrium and non-equilibrium，discusses the correlation between the evaluation models of P-R curve from the relationship between several evaluation indexes，and discusses the rationality of P-R curve as the evaluation model method in the target detection task.
Keywords：model performance evaluation；Accuracy；P-R curve；mAP
参考文献：
[1] DENG J，DONG W，SOCHER R，et al.ImageNet：a LargeScale Hierarchical Image Database [C]//2009 IEEE Computer Society Conference on Computer Vision and Pattern Recognition （CVPR 2009），USA.IEEE，2009.
[2] DAVIS J，GOADRICH M.The Relationship Between Precision-Recall and ROC Curves [C]//ICML’06：Proceedings of the 23rd international conference on Machine learning，2006：233-240.
[3] EVERINGHAM M，GOOL L V，WILLIAMS C K I，et al.The Pascal Visual Object Classes （VOC） Challenge [J].International Journal of Computer Vision，2010，88（2）：303-338.
[4] CHEN K，LI J G，LIN W Y，et al.Towards Accurate OneStage Object Detection with AP-Loss [J].[2019-12-26].https://arxiv.org/abs/1904.06373?context=cs.CV.
作者简介：张超（1992.06-），男，汉族，河南固始人，硕士在读，研究方向：模式识别。