最原始的代码是py-fasterrcnn作者写的，这个使用过程有点繁杂，主要是传递的参数比较多杂，我在原始基础上做了封装和部分重写，方便自己使用，在此也分享一下：
import xml.etree.ElementTree as ET
import os,glob,json
import numpy as np
def pathjoin(prepath,*filenames):
fullpath=prepath
for filename in filenames:
fullpath=os.path.join(fullpath,filename)
return fullpath
def getxextnamefilefromdir(datadir,fileextname='.xml'):
filepaths = glob.glob('{}/*{}'.format(datadir, fileextname))
filepaths.sort()
return filepaths
class VOCDetIndicator:
def __init__(self,testsetdirs,detresulttxtdir,databasename,clsnames,mergeclsnamesdict={},iouthresh=0.5,use_07_metric=False):
self.testsetdirs=testsetdirs
self.clsnames=clsnames
self.mergeclsnamesdict=mergeclsnamesdict
self.detresulttxtdir=detresulttxtdir
self.iouthresh=iouthresh
self.use_07_metric=use_07_metric
def getannotgroundtruth():
def parse_rec(filename):
""" Parse a PASCAL VOC xml file """
tree = ET.parse(filename)
objects = []
for obj in tree.findall('object'):
obj_struct = {}
cls = obj.find('name').text
if cls in self.mergeclsnamesdict.keys():
cls=mergeclsnamesdict[cls]
obj_struct['name'] = cls
# obj_struct['pose'] = obj.find('pose').text
obj_struct['truncated'] = int(obj.find('truncated').text)
obj_struct['difficult'] = int(obj.find('difficult').text)
bbox = obj.find('bndbox')
obj_struct['bbox'] = [int(float(bbox.find('xmin').text)),
int(float(bbox.find('ymin').text)),
int(float(bbox.find('xmax').text)),
int(float(bbox.find('ymax').text))]
objects.append(obj_struct)
# print 'voc_eval.py-->parse_rec()-->objects='+str(objects)+'\n'
return objects
recs = {}
annotcachefilepath = pathjoin(detresulttxtdir, databasename + '_annot.json')
if os.path.exists(annotcachefilepath):
print('reading annotation cache file')
with open(annotcachefilepath, 'r') as f:
recs = json.load(f)
else:
i = 0
for testsetdir in testsetdirs:
annotpaths = getxextnamefilefromdir(testsetdir, fileextname='.xml')
for annotpath in annotpaths:
filename = os.path.basename(annotpath)[:-4]
recs[filename] = parse_rec(annotpath)
i += 1
if i % 500 == 0:
print('Reading annotation for {:d}'.format(i))
# save
print('Saving cached annotations to {:s}'.format(annotcachefilepath))
with open(annotcachefilepath, 'w') as fp:
json.dump(recs, fp,indent=4)
return recs
self.annotgroundtruth=getannotgroundtruth()
def voc_ap(self,rec, prec, use_07_metric=False):
""" ap = voc_ap(rec, prec, [use_07_metric])
Compute VOC AP given precision and recall.
If use_07_metric is true, uses the
VOC 07 11 point method (default: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(self,clsname):
recs = self.annotgroundtruth
# extract gt objects for this class
class_recs = {}
npos = 0 #
for imagename in recs.keys():
R = [obj for obj in recs[imagename] if obj['name'] == clsname]
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[imagename] = {'bbox': bbox,
'difficult': difficult,
'det': det}
# read dets
detresulttxtname = '{}.txt'.format(clsname)
detfile =pathjoin(self.detresulttxtdir,detresulttxtname)
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) #
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 > self.iouthresh: #
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 = self.voc_ap(rec, prec, self.use_07_metric) #
return rec, prec, ap
def gettestsetdetindicator(self):
precisions,recalls,aps={},{},{}
for clsname in self.clsnames:
rec, prec, ap = self.voc_eval(clsname)
precisions[clsname]=round(prec[-1],3)
recalls[clsname]=round(rec[-1],3)
aps[clsname]=round(ap,3)
map=0
for cls in aps.keys():
map+=aps[cls]
map=map/len(aps)
map=round(map,3)
precision_recall_ap_map = {'precision':precisions,'recall':recalls,'ap':aps,'map':map}
return precision_recall_ap_map
if __name__ == '__main__':
testsetdirs=['testset1','testset2']#maybe you have mutiply testset and they are in different locale
detresulttxtdir='detect result txt save dir'
databasename='your database name'#its convenient for remember the data
clsnames=('class1','class2','...')#class names
DetIndicator=VOCDetIndicator(testsetdirs, detresulttxtdir, databasename, clsnames)
本文地址：https://blog.csdn.net/lingyunxianhe/article/details/109628455
