对检测出来的结果单独进行softnms操作
1.从官方文档https://github.com/bharatsingh430/soft-nms/blob/master/lib/nms/cpu_nms.pyx，下载cython实现的softnms的代码 记住最后一行的把return keep改为return boxes[keep,:]，具体如下
# --------------------------------------------------------
# Deformable Convolutional Networks
# Copyright (c) 2015 Microsoft
# Licensed under The MIT License [see LICENSE for details]
# Modified from py-faster-rcnn (https://github.com/rbgirshick/py-faster-rcnn)
# --------------------------------------------------------
import numpy as np
cimport numpy as np
cdef inline np.float32_t max(np.float32_t a, np.float32_t b):
return a if a >= b else b
cdef inline np.float32_t min(np.float32_t a, np.float32_t b):
return a if a <= b else b
def cpu_soft_nms(np.ndarray[float, ndim=2] boxes, float sigma=0.5, float Nt=0.3, float threshold=0.001, unsigned int method=0):
cdef unsigned int N = boxes.shape[0]
cdef float iw, ih, box_area
cdef float ua
cdef int pos = 0
cdef float maxscore = 0
cdef int maxpos = 0
cdef float x1,x2,y1,y2,tx1,tx2,ty1,ty2,ts,area,weight,ov
for i in range(N):
maxscore = boxes[i, 4]
maxpos = i
tx1 = boxes[i,0]
ty1 = boxes[i,1]
tx2 = boxes[i,2]
ty2 = boxes[i,3]
ts = boxes[i,4]
pos = i + 1
# get max box
while pos < N:
if maxscore < boxes[pos, 4]:
maxscore = boxes[pos, 4]
maxpos = pos
pos = pos + 1
# add max box as a detection
boxes[i,0] = boxes[maxpos,0]
boxes[i,1] = boxes[maxpos,1]
boxes[i,2] = boxes[maxpos,2]
boxes[i,3] = boxes[maxpos,3]
boxes[i,4] = boxes[maxpos,4]
# swap ith box with position of max box
boxes[maxpos,0] = tx1
boxes[maxpos,1] = ty1
boxes[maxpos,2] = tx2
boxes[maxpos,3] = ty2
boxes[maxpos,4] = ts
tx1 = boxes[i,0]
ty1 = boxes[i,1]
tx2 = boxes[i,2]
ty2 = boxes[i,3]
ts = boxes[i,4]
pos = i + 1
# NMS iterations, note that N changes if detection boxes fall below threshold
while pos < N:
x1 = boxes[pos, 0]
y1 = boxes[pos, 1]
x2 = boxes[pos, 2]
y2 = boxes[pos, 3]
s = boxes[pos, 4]
area = (x2 - x1 + 1) * (y2 - y1 + 1)
iw = (min(tx2, x2) - max(tx1, x1) + 1)
if iw > 0:
ih = (min(ty2, y2) - max(ty1, y1) + 1)
if ih > 0:
ua = float((tx2 - tx1 + 1) * (ty2 - ty1 + 1) + area - iw * ih)
ov = iw * ih / ua #iou between max box and detection box
if method == 1: # linear
if ov > Nt:
weight = 1 - ov
else:
weight = 1
elif method == 2: # gaussian
weight = np.exp(-(ov * ov)/sigma)
else: # original NMS
if ov > Nt:
weight = 0
else:
weight = 1
boxes[pos, 4] = weight*boxes[pos, 4]
# if box score falls below threshold, discard the box by swapping with last box
# update N
if boxes[pos, 4] < threshold:
boxes[pos,0] = boxes[N-1, 0]
boxes[pos,1] = boxes[N-1, 1]
boxes[pos,2] = boxes[N-1, 2]
boxes[pos,3] = boxes[N-1, 3]
boxes[pos,4] = boxes[N-1, 4]
N = N - 1
pos = pos - 1
pos = pos + 1
keep = [i for i in range(N)]
return boxes[keep,:]2.创建setup.py文件
from distutils.core import setup
from Cython.Build import cythonize
setup(
name = 'softnms_module',
ext_modules = cythonize('cpu_nms.pyx'),
)3.开始生成动态链接库so文件和o文件，以及c源代码：
python setup.py build我在这里出现了错误，提示numpy/arrayobject.h: No such file or directory，usr/include/python2.7下面的numpy的软连接，出现错误，然后顺势找到/usr/lib/python2.7/dist-packages发现没有numpy文件夹，然后按照网上的提示 sudo apt-get install python-numpy，提示我已经安装了，然后我在终端下运行cython，import numpy也没有问题。后来pip install numpy，然后发下/usr/lib/python2.7/dist-packages还是发现没有numpy文件夹，之后想起了我是多用户的，在/home/ygx/.local/lib/python2.7/site-packages/numpy，把这个文件夹复制到/usr/lib/python2.7/dist-packages，问题解决。
4.将生成的build文件夹下面的so文件，拷贝到主文件夹下面，删除build文件夹，即可。
5.之后下面是softnms.py的文件，该文件作用是单独执行softnms操作，可调参数有sigma和method，method如果是0就和普通的nms没差。注意我这里的的nms_thresh其实就是置信度的阈值，和普通的不一样。你把它设置成和原先检测时置信度的阈值就可以了，因为重叠区域高的置信度不会变，会将相对低的变的更低，比如0.4变成0.1这样。
#!/usr/bin/env python
# -*- coding: utf-8 -*-
"""
Created on Mon Apr 9 21:09:09 2018
对每一类生成的prebbox.txt进行softnms操作
1.读入文本，读入bbox
@author: ygx
"""
import sys
defaultencoding = 'utf-8'
if sys.getdefaultencoding() != defaultencoding:
reload(sys)
sys.setdefaultencoding(defaultencoding)
import os
import numpy as np
from cpu_nms import cpu_soft_nms
nms_thresh = 0.45 #这个和你之前设置的置信度的阈值一样就好了
def custombasename(fullname):
return os.path.basename(os.path.splitext(fullname)[0])
def GetFileFromThisRootDir(dir,ext = None):
allfiles = []
needExtFilter = (ext != None)
for root,dirs,files in os.walk(dir):
for filespath in files:
filepath = os.path.join(root, filespath)
extension = os.path.splitext(filepath)[1][1:]
if needExtFilter and extension in ext:
allfiles.append(filepath)
elif not needExtFilter:
allfiles.append(filepath)
return allfiles
def cpu_soft_nms_ygx(dets, sigma=0.5, Nt=0.3, threshold=0.001, method=1):
keep = cpu_soft_nms(np.ascontiguousarray(dets, dtype=np.float32),
np.float32(sigma), np.float32(Nt),
np.float32(threshold),
np.uint8(method))
return keep
def nmsbynamedict(nameboxdict, nms):
nameboxnmsdict = {x: [] for x in nameboxdict}
for imgname in nameboxdict:
#print('imgname:', imgname)
#keep = py_cpu_nms(np.array(nameboxdict[imgname]), thresh)
#print('type nameboxdict:', type(nameboxnmsdict))
#print('type imgname:', type(imgname))
#print('type nms:', type(nms))
keep = nms(np.array(nameboxdict[imgname]))
#======================ygx=====================
#keep = nms(np.array(keep))#又多进行了一层的softnms
##############################################
#print('keep:', keep)
outdets = []
#print('nameboxdict[imgname]: ', nameboxnmsdict[imgname])
#======================ygx=====================
all_num = len(keep)
keep_soft = []
for i in range(all_num):
#可以调试的时候看一下
if keep[i][4]>nms_thresh:
keep_soft.append(i)
else:
continue
for index in keep_soft:
# print('index:', index)
outdets.append(keep[index])
#千万 注意在softnms之后索引已经发生了变化。。。
#outdets.append(nameboxdict[imgname][index])
nameboxnmsdict[imgname] = outdets
return nameboxnmsdict
#我要将他改成softnms，对15类最终的txt进行softnms
def softnmsbase(srcpath, dstpath, nms):
filelist = GetFileFromThisRootDir(srcpath)
for fullname in filelist:
name = custombasename(fullname)
#print('name:', name)
dstname = os.path.join(dstpath, name + '_softnms_'+str(nms_thresh)+'.txt')
with open(fullname, 'r') as f_in:
nameboxdict = {}
lines = f_in.readlines()
splitlines = [x.strip().split(' ') for x in lines]
for splitline in splitlines:
oriname = splitline[0]
confidence = splitline[1]
poly = list(map(float, splitline[2:]))
det = poly
#det.append(classname)
det.append(confidence)
det = list(map(float, det))
if (oriname not in nameboxdict):
nameboxdict[oriname] = []
nameboxdict[oriname].append(det)
nameboxnmsdict = nmsbynamedict(nameboxdict, nms)
with open(dstname, 'w') as f_out:
for imgname in nameboxnmsdict:
for det in nameboxnmsdict[imgname]:
#print('det:', det)
confidence = det[-1]
bbox = det[0:-1]
outline = str(imgname) + ' ' + ' '.join(map(str, bbox))
#outline = imgname + ' ' + str(confidence) + ' ' + ' '.join(map(str, bbox))
#print('outline:', outline)
f_out.write(outline + '\n')
if __name__ == '__main__':
#mergebypoly()
srcpath = r'/home/ygx/666/nms之前'
dstpath = r'/home/ygx/666/nms之后'
softnmsbase(srcpath,
dstpath,
cpu_soft_nms_ygx)
参考博客：http://www.cnblogs.com/king-lps/p/9031568.html