欢迎您访问程序员文章站本站旨在为大家提供分享程序员计算机编程知识!
您现在的位置是: 首页  >  IT编程

绘制PR曲线

程序员文章站 2022-03-28 17:58:43
一、 获取txt文件运行darknet官方代码中的darknet detector valid data cfg weight指令(例如:darknet.exe detector valid data/koujian/koujian.data cfg/yolov3-tiny11.cfg backup/yolov3-tiny11_last.weights),可以在result/目录下得到网络检测的输出txt文件:包括检测的图像名字、类别、概率、边界框位置(左上角和右下角):二.新建两个文件:rev...

一、 获取txt文件

运行darknet官方代码中的darknet detector valid data cfg weight指令(例如:
darknet.exe detector valid data/koujian/koujian.data cfg/yolov3-tiny11.cfg backup/yolov3-tiny11_last.weights),可以在result/目录下得到网络检测的输出txt文件:包括检测的图像名字、类别、概率、边界框位置(左上角和右下角):
绘制PR曲线

二.新建两个文件:

  1. reval_voc_py3.py
#!/usr/bin/env python

import os, sys, argparse
import numpy as np
import _pickle as cPickle
from voc_eval_py3 import voc_eval
import matplotlib.pyplot as plt

def do_python_eval(label_path, valid_file, classes, output_dir = 'results'):
    cachedir = os.path.join('./', 'annotations_cache')
    aps = []
    use_07_metric = False
    print('VOC07 metric? ' + ('Yes' if use_07_metric else 'No'))
    
    if not os.path.isdir(output_dir):
        os.mkdir(output_dir)
    for i, cls in enumerate(classes):
        if cls == '__background__':
            continue
        rec, prec, ap = voc_eval(
            label_path,
            valid_file,  cls, cachedir, ovthresh=0.5,
            use_07_metric=use_07_metric)
        aps += [ap]
        print('AP for {} = {:.4f}'.format(cls, ap))
        with open(os.path.join(output_dir, cls + '_pr.pkl'), 'wb') as f:
            cPickle.dump({'rec': rec, 'prec': prec, 'ap': ap}, f)
    print('Mean AP = {:.4f}'.format(np.mean(aps)))
    print('~~~~~~~~')
    print('Results:')
    for ap in aps:
        print('{:.3f}'.format(ap))
    print('{:.3f}'.format(np.mean(aps)))
  

    fr = open(cls + '_pr.pkl','rb')
    inf = cPickle.load(fr)
    fr.close()

    x=inf['rec']
    y=inf['prec']
    fig = plt.figure(1, dpi=160)
    ax = fig.add_subplot(1,1,1)
    ax.plot(x, y, label='PR')
    #ax.plot(result['Avg Recall'].values, label='Avg Recall')
    #plt.grid()
    ax.legend(loc='upper right')
    ax.set_ylim([0.6, 1.04])
    ax.set_xlim([0.0, 1.05])
    ax.set_title('PR curves')
    ax.set_xlabel('recall')
    ax.set_ylabel('precision')
    ax.spines['top'].set_visible(False)     #去掉上边框
    ax.spines['right'].set_visible(False)   #去掉右边框
    fig.savefig('PR')
    plt.savefig("PR.svg", format="svg")
'''
    x=inf['rec']
    y=inf['prec']
    plt.figure()
    plt.xlabel('recall')
    plt.ylabel('precision')
    plt.title('PR cruve')
    plt.ylim([0.6, 1.05])
    plt.plot(x,y, label='PR')
    plt.legend(loc= 'upper right')
    plt.show()
    print('AP:',inf['ap'])
'''
if __name__ == '__main__':
    label_path = r'D:\darknet\darknet-master\build\darknet\x64\data\koujian\valid'  		#label文件夹(验证集所在位置)
    valid_file = r'comp4_det_test_koujian.txt'						#valid命令生成的txt文件,在result/目录下。	
    name_path = r'D:\darknet\darknet-master\build\darknet\x64\data\koujian\koujian.names' 	#name文件
    output_dir = os.path.abspath('./')			 		#pkl保存路径
    with open(name_path, 'r') as f:	     
        lines = f.readlines()
    classes = [t.strip('\n') for t in lines]
    print('Evaluating detections')
    do_python_eval(label_path, valid_file, classes, output_dir)

其中需要修改:
(1). label_path # label文件夹,标注txt和图像应在同一目录下
(2). valid_file # valid命令生成的txt文件,在result/目录下。
(3). name_path # name文件
(4). output_dir # 生成的pkl保存路径

2 voc_eval_py3.py

import xml.etree.ElementTree as ET
import os
import _pickle as cPickle
import numpy as np
import cv2

def parse_rec(label_path, label_name):
    objects = []
    label_file = os.path.join(label_path, label_name + '.txt')
    img_file = os.path.join(label_path, label_name + '.jpg')
    height, width, _ = cv2.imread(img_file).shape
    with open(label_file) as f:
        for line in f.readlines():
            obj_struct = {}
            obj_struct['name'] = 'koujian'	#需要修改成自己的names
            obj_struct['difficult'] = int(0)
            center_x, center_y, width_b, height_b =[float(x) for x in line.split()[1:]]
            obj_struct['bbox'] = [int(center_x * width - width * width_b / 2.0),
                                  int(center_y * height - height * height_b / 2.0),
                                  int(center_x * width + width * width_b / 2.0),
                                  int(center_y * height + height * height_b / 2.0)]
            objects.append(obj_struct)
    return objects

def voc_ap(rec, prec, use_07_metric=False):
    if use_07_metric:
        # 11 point metric
        ap = 0.
        for t in np.arange(0., 1.1, 0.1):
            if np.sum(rec >= t) == 0:
                p = 0
            else:
                p = np.max(prec[rec >= t])
            ap = ap + p / 11.
    else:
        # correct AP calculation
        # first append sentinel values at the end
        mrec = np.concatenate(([0.], rec, [1.]))
        mpre = np.concatenate(([0.], prec, [0.]))

        # compute the precision envelope
        for i in range(mpre.size - 1, 0, -1):
            mpre[i - 1] = np.maximum(mpre[i - 1], mpre[i])

        # to calculate area under PR curve, look for points
        # where X axis (recall) changes value
        i = np.where(mrec[1:] != mrec[:-1])[0]

        # and sum (\Delta recall) * prec
        ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1])
    return ap

def voc_eval(label_path,
             detpath,
             classname,
             cachedir,
             ovthresh=0.5,
             use_07_metric=False):
    # first load gt
    if not os.path.isdir(cachedir):
        os.mkdir(cachedir)
    cachefile = os.path.join(cachedir, 'annots.pkl')

    label_file = []
    for f in os.listdir(label_path):
        file, tmp = f.split('.')
        if  tmp == 'txt':
            label_file.append(file)
    
    if not os.path.isfile(cachefile):
        # load annots
        recs = {}
        for label_name in label_file:
            recs[label_name] = parse_rec(label_path, label_name)
        with open(cachefile, 'wb') as f:
            cPickle.dump(recs, f)
    else:
        # load
        print('!!! cachefile = ',cachefile)
        with open(cachefile, 'rb') as f:
            recs = cPickle.load(f)

    # extract gt objects for this class
    class_recs = {}
    npos = 0   #修改
    for label_name in label_file:
    # for imagename in imagenames:
        R = [obj for obj in recs[label_name] if obj['name'] == classname]
        bbox = np.array([x['bbox'] for x in R])
        difficult = np.array([x['difficult'] for x in R]).astype(np.bool)
        det = [False] * len(R)
        npos = npos + sum(~difficult)
        class_recs[label_name] = {'bbox': bbox,
                                 'difficult': difficult,
                                 'det': det}

    # read dets
    detfile = detpath
    # detfile = detpath.format(classname)
    with open(detfile, 'r') as f:
        lines = f.readlines()

    splitlines = [x.strip().split(' ') for x in lines]
    image_ids = [x[0] for x in splitlines]
    confidence = np.array([float(x[1]) for x in splitlines])
    BB = np.array([[float(z) for z in x[2:]] for x in splitlines])

    # sort by confidence
    sorted_ind = np.argsort(-confidence)
    sorted_scores = np.sort(-confidence)
    BB = BB[sorted_ind, :]
    image_ids = [image_ids[x] for x in sorted_ind]
    # go down dets and mark TPs and FPs
    nd = len(image_ids)
    # print(image_ids)
    # print(nd)
    tp = np.zeros(nd)
    fp = np.zeros(nd)
    for d in range(nd):
        R = class_recs[image_ids[d]]
        bb = BB[d, :].astype(float)
        ovmax = -np.inf
        BBGT = R['bbox'].astype(float)

        if BBGT.size > 0:
            # compute overlaps
            # intersection
            ixmin = np.maximum(BBGT[:, 0], bb[0])
            iymin = np.maximum(BBGT[:, 1], bb[1])
            ixmax = np.minimum(BBGT[:, 2], bb[2])
            iymax = np.minimum(BBGT[:, 3], bb[3])
            iw = np.maximum(ixmax - ixmin + 1., 0.)
            ih = np.maximum(iymax - iymin + 1., 0.)
            inters = iw * ih

            # union
            uni = ((bb[2] - bb[0] + 1.) * (bb[3] - bb[1] + 1.) +
                   (BBGT[:, 2] - BBGT[:, 0] + 1.) *
                   (BBGT[:, 3] - BBGT[:, 1] + 1.) - inters)

            overlaps = inters / uni
            ovmax = np.max(overlaps)
            jmax = np.argmax(overlaps)

        if ovmax > ovthresh:
            if not R['difficult'][jmax]:
                if not R['det'][jmax]:
                    tp[d] = 1.
                    R['det'][jmax] = 1
                else:
                    fp[d] = 1.
        else:
            fp[d] = 1.

    # compute precision recall
    fp = np.cumsum(fp)
    tp = np.cumsum(tp)
    rec = tp / float(npos)
    # avoid divide by zero in case the first detection matches a difficult
    # ground truth
    prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps)
    ap = voc_ap(rec, prec, use_07_metric)

    return rec, prec, ap

三、运行

将以上valid命令生成的txt文件和py文件放在同一个文件夹下,在python中终端运行reval_voc_py.py,可在该文件夹下得到一个pkl文件,名字为你检测的物体名字。如果没错的话应该会直接在终端生成一个PR曲线图。
当然,你也可以根据上面生成的这个pkl文件,再新建一个PR_draw.py文件:

import _pickle as cPickle
import matplotlib.pyplot as plt
fr = open('koujian_pr.pkl','rb')#这里open中第一个参数需要修改成自己生产的pkl文件
#fr1= open('quexian_pr.pkl','rb')
inf = cPickle.load(fr)

fr.close()

 
x=inf['rec']
y=inf['prec']

plt.figure()
plt.xlabel('recall')
plt.ylabel('precision')
plt.title('PR cruve')
plt.plot(x,y, label='PR')
plt.legend(loc='upper right')
plt.show()
 
print('AP:',inf['ap'])

运行该文件,同样可以得到PR曲线图。

参考链接

https://blog.csdn.net/Mr_kuilei/article/details/105641774
https://blog.csdn.net/qq_33350808/article/details/83178002

本文地址:https://blog.csdn.net/tzwsg/article/details/107677231