Object Detection evaluation tools

Object Detection evaluation tools

Introduction

Lots of object detection algorithms used mAp(mean Average precision) as evaluation metric, and yet their mAp calculation code embedded in algorithm implementation code, which is not easy to be separated. Here is offering an independent mAp calculation tool, you can do:

• Calculating mAp for your object detection algorithm testing(11 points method supported)
• Get recall and precision in any confidence threshold
• Drawing PR curve
• Drawing ROC curve(Done)
• Get FPPI(False Positive Per Image) and FPPW(False Positive Per Window) in any threshold(Done)

Update

• fix some bugs
• sematic label name is supported now (e.g. "Person", "Car")

Requirements

• python 2.X OR python 3.X
• python-opencv (any version)
• matplotlib 2.2.3

Input format

• TXT format, see 'sample' folder
• XML format, only Pascal VOC style supported

Usage

python test.py --dir ./sample/prediction ./sample/test_annos --ratio 0.5 --thre 0.7

Anything you want to know about usage can be found with typing:

python test.py --h

or

python test.py --help