深度学习python处理数据脚本

概述

总结深度学习项目中使用到的小工具.

图片与标签xml文件相对应

此脚本文件是为了解决图片与标签xml文件不对应的问题,清理数据,主要体现在图片多出几张或者标签多出几张的情况下.
(这里也可以变成“图片与标签txt文件相对应”，只需要把代码中的xml改成txt就可以了)

#encoding:utf-8
# !/usr/bin/python

'''
USE Method:python move.py --xml xmlpath --pic picpath
//由于path =  os.path.abspath('.')代码，xmlpath取当前路径下的文件夹路径
'''


import os
import sys
import argparse
import os.path
import shutil

path =  os.path.abspath('.')

def parse_args():
    """
    Parse input arguments
    """
    parser = argparse.ArgumentParser(description='cut pic and xml 1 to 2')
    parser.add_argument('--xml', dest='xml', type=str)
    parser.add_argument('--pic', dest='pic',type=str)

    if len(sys.argv) == 1:
        parser.print_help()
        sys.exit(1)

    args = parser.parse_args()
    return args

if __name__ == '__main__':

    args = parse_args()
    print('Called with args:')
    print(args)

    xml = args.xml
    pic = args.pic
    xml_path = path + '/' + xml
    pic_path = path + '/' + pic
    dst_xml_path = path + '/' + "not_pic"
    dst_pic_path = path + '/' + "not_xml"

    if not os.path.exists(dst_pic_path):
        os.mkdir(dst_pic_path)
    else:
        print "此目录已存在！"
        sys.exit(1)

    if not os.path.exists(dst_xml_path):
        os.mkdir(dst_xml_path)
    else:
        print "此目录已存在！"
        sys.exit(1)

    for file in os.listdir(pic_path):
        xml_name = os.path.join(xml_path, os.path.splitext(file)[0] + ".xml")
        if os.path.exists(xml_name):
            print "This file", file, "has xml !"
        elif not os.path.exists(xml_name):
            print "This file", file, "has not xml !"
            shutil.move(os.path.join(pic_path,file), dst_pic_path)

    for file in os.listdir(xml_path):
        pic_name = os.path.join(pic_path, os.path.splitext(file)[0] + ".jpg")
        if os.path.exists(pic_name):
            print "This file", file, "has pic !"
        elif not os.path.exists(pic_name):
            print "This file", file, "has not pic !"
            shutil.move(os.path.join(xml_path, file), dst_xml_path)

    print "Done!"

        #pic_path = os.path.join(pic_path, pic_name)
        #shutil.move(xml_name, dst_path)
        
    #for dirpath, dirnames, filenames in os.walk("f:/"):
     #   for filename in filenames:
      #      if os.path.splitext(filename)[1] == ".txt":
       #        print filepath
        #        copy(filepath, "F:/test/" + filename)
         #       Shutil.move(changeFilePath, dst_path)

修改xml中的标签值

修改标签中的值,这里举例只修改标签中的名称.应用场景:当标签类别需要合并,细标签要合并成粗标签时用到.

#!/usr/bin/python
#-*- coding: utf-8 -*-

'''
USE Method : python modify_xml_label.py
'''
# coding=utf-8
import os
import os.path
import xml.dom.minidom

import sys
reload(sys)
sys.setdefaultencoding("utf-8")

path = "/media/t/disk1/jie_shou_ren_ming/matiansu/output/tail_xml"
outpath = "/media/t/disk1/jie_shou_ren_ming/matiansu/output/tail_output_xml"
files = os.listdir(path)  # 得到文件夹下所有r文件名称
s = []
for xmlFile in files:
    # 遍历文件夹
    portion = os.path.splitext(xmlFile)
    if not os.path.isdir(xmlFile):
        # 判断是否是文件夹,不是文件夹才打开
        # print (xmlFile)
 
        # xml文件读取操作
 
        # 将获取的xml文件名送入到dom解析
	print(xmlFile)
        dom = xml.dom.minidom.parse(os.path.join(path, xmlFile))
        ###最核心的部分os.path.join(path,xmlFile),路径拼接,输入的是具体路径
        root = dom.documentElement
        name = root.getElementsByTagName('name')
        # pose=root.getElementsByTagName('pose')
        # 重命名class name
        for i in range(len(name)):
            # print (name[i].firstChild.data)
            # print(xmlFile)
            # if portion[1] == ".xml":
            #     newname = portion[0] + ".jpg"
            #     print(newname)
            newname ="2"
            if name[i].firstChild.data =="1":
 
                name[i].firstChild.data = newname
                print(name[i].firstChild.data)
 
            # 保存修改到xml文件中
        with open(os.path.join(outpath, xmlFile), 'w') as fh:
            dom.writexml(fh)
            print('修改filename OK!')

生成xml脚本(caffe)

应用场合:通过小批量标注数据,用caffe训练一个初步模型,在通过下面脚本生成xml标签文件,方便后续标注组人员标注.

#!/usr/bin/env python
# set up Python environment: numpy for numerical routines, and matplotlib for plotting
#-*- coding: utf-8 -*-

import glob

from lxml import etree, objectify

import time
import numpy as np
import sys
import cv2
#from pascal_voc_writer import Writer
#import matplotlib.pyplot as plt
# display plots in this notebook
import argparse
# set display defaults
#plt.rcParams['figure.figsize'] = (10, 10)        # large images
#plt.rcParams['image.interpolation'] = 'nearest'  # don't interpolate: show square pixels
#plt.rcParams['image.cmap'] = 'gray'  # use grayscale output rather than a (potentially misleading) color heatmap

# The caffe module needs to be on the Python path;
#  we'll add it here explicitly.
import sys
caffe_root = './'  # this file should be run from {caffe_root}/examples (otherwise change this line)
sys.path.insert(0, caffe_root + 'python')
import os
import caffe
import math
from os import walk
from os.path import join
# If you get "No module named _caffe", either you have not built pycaffe or you have the wrong path.

CLASSES = ('__background__',
           'SmallVehicle', 'bus', 'BigVehicle')

def gen_txt(image,result,filename, path):
    w = image.shape[1]
    h = image.shape[0]
    E = objectify.ElementMaker(annotate=False)
    anno_tree = E.annotation(
        E.folder('JPEGImages'),
        E.filename(filename),
        E.size(
            E.width(w),
            E.height(h),
            E.depth(3)
        ),
        E.segmented(0),
    )
    #etree.ElementTree(anno_tree).write(path[:-4]+".xml", pretty_print=True)
    for i in range(result.shape[1]):
	for j in range(3,7,1):
	    if result[0][i][j]<0:
		result[0][i][j] = 0
	    if result[0][i][j]>1:
		result[0][i][j] = 1
        left = result[0][i][3] * w
        top = result[0][i][4] * h
        right = result[0][i][5] * w
        bot = result[0][i][6] * h
        label = result[0][i][1]
	score = result[0][i][2]
	if(score<0.5) :
	    continue;
        E2 = objectify.ElementMaker(annotate=False)
        anno_tree2 = E2.object(
            E.name(CLASSES[int(label)]),
            E.pose('Unspecified'),
            E.truncated('0'),
            E.difficult('0'),
            E.bndbox(
                E.xmin(int(left)),
                E.ymin(int(top)),
                E.xmax(int(right)),
                E.ymax(int(bot))
            )
        )
	print "i = %d " % i
        anno_tree.append(anno_tree2)
    etree.ElementTree(anno_tree).write(path[:-4]+".xml", pretty_print=True)

def vis_detections(image,result) :
    w = image.shape[1]
    h = image.shape[0]
    for i in range(result.shape[1]):
        left = result[0][i][3] * w
        top = result[0][i][4] * h
        right = result[0][i][5] * w
        bot = result[0][i][6] * h
        score = result[0][i][2]
        label = result[0][i][1]
        box = [left,top,right,bot];
        if(score>0.4) :
            print(left,right,top,bot,score,label)
            cv2.rectangle(image,(int(left), int(top)),(int(right),int(bot)),(0,255,0), 2)

            label = '{:s} {:.3f}'.format(CLASSES[int(label)], score)
            font = cv2.FONT_HERSHEY_SIMPLEX
            size = cv2.getTextSize(label, font, 0.5, 0)[0]
            cv2.rectangle(image,(int(left), int(top)),
                    (int(left+size[0]),int(top+ size[1])),(0,255,0), -1)

            cv2.putText(image, label,(int(left+0.5), int(top+ size[1]+0.5)),font,0.5,(0,0,0),0)
def write_detections(image,result,writer) :
    w = image.shape[1]
    h = image.shape[0]
    for i in range(result.shape[1]):
        left = result[0][i][3] * w
        top = result[0][i][4] * h
        right = result[0][i][5] * w
        bot = result[0][i][6] * h
        score = result[0][i][2]
        label = result[0][i][1]
        if(score>0.1) :
            print(left,right,top,bot,score,label)
            cv2.rectangle(image,(int(left), int(top)),(int(right),int(bot)),(0,255,0), 2)

            label = '{:s}'.format(CLASSES[int(label)])
            font = cv2.FONT_HERSHEY_SIMPLEX
            size = cv2.getTextSize(label, font, 0.5, 0)[0]
            cv2.rectangle(image,(int(left), int(top)),
                    (int(left+size[0]),int(top+ size[1])),(0,255,0), -1)

            cv2.putText(image, label,(int(left), int(top+ size[1])),font,0.5,(0,0,0),0)
            #writer.addObject(label, int(left+0.5), int(top+0.5), int(right+0.5), int(bot+0.5))
def det(image,transformer,net):
    
    transformed_image = transformer.preprocess('data', image)
    #plt.imshow(image)

    net.blobs['data'].data[...] = transformed_image

    ### perform classification
    output = net.forward()

    res = output['detection_out'][0]  # the output probability vector for the first image in the batch
    #print(res.shape)
    return res

def is_imag(filename):
    return filename[-4:] in ['.png', '.jpg']

def main(args):    
 
    caffe.set_mode_gpu()
    model_def = args.model_def
    model_weights = args.model_weights
    
    net = caffe.Net(model_def,      # defines the structure of the model
                    model_weights,  # contains the trained weights
                    caffe.TEST)     # use test mode (e.g., don't perform dropout)
    
    mu = np.array([0.5, 0.5, 0.5])
    # create transformer for the input called 'data'
    transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})

    transformer.set_transpose('data', (2,0,1))  # move image channels to outermost dimension
    transformer.set_mean('data', mu)            # subtract the dataset-mean value in each channel
    transformer.set_raw_scale('data', 1.0)      # rescale from [0, 1] to [0, 255]
    transformer.set_channel_swap('data', (2,1,0))  # swap channels from RGB to BGR

    net.blobs['data'].reshape(1,        # batch size
                              3,         # 3-channel (BGR) images
                              args.image_resize, args.image_resize)  # image size is 227x227
    
    filenames = os.listdir(args.image_dir)
    images = filter(is_imag, filenames)
    for image in images :
        
        pic = args.image_dir + image
        input = caffe.io.load_image(pic)       
        image_show =cv2.imread(pic)
        start_time=time.time()
        result = det(input,transformer,net)
        end_time=time.time() 
        print("time:%lf"  % (end_time-start_time))
        #vis_detections(image_show,result)
        gen_txt(image_show,result,image, pic)
        if args.write_voc:
            writer = Writer(pic, input.shape[1], input.shape[0])
            write_detections(image_show,result,writer)
            base = os.path.splitext(pic)[0]
            writer.save(base+".xml")
        else :
            cv2.imshow("Image", image_show)
            #cv2.waitKey (1000)
      
def parse_args():
    parser = argparse.ArgumentParser()
    '''parse args'''
    parser.add_argument('--image_dir', default='/media/t/disk1/DukTo0929/14/')
    parser.add_argument('--model_def', default='/media/t/disk1/jie_shou_ren_ming/guopei/MobileNetSSD_deploy.prototxt')
    parser.add_argument('--model_weights', default='/media/t/disk1/jie_shou_ren_ming/guopei/mobilenet_v2_300x300_iter_10000.caffemodel')
    parser.add_argument('--image_resize', default=608, type=int)
    parser.add_argument('--write_voc', default=False)
    return parser.parse_args()
    
if __name__ == '__main__':
    main(parse_args())

删除满足一定条件的标注框

应用场景:例如删除标注数据中标注框的面积小于10000的框.

#!/usr/bin/python
#-*- coding: utf-8 -*-

'''
USE Method : python modify_xml_label.py
'''
# coding=utf-8
import os
import os.path
from lxml import etree, objectify
import xml.dom.minidom

import sys
reload(sys)
sys.setdefaultencoding("utf-8")

path = "/home/t/data/VOCdevkit/MyDataSet/Annotations"
outpath = "/home/t/data/VOCdevkit/MyDataSet/Annotations_output"

def gen_txt(w,h,result,filename,path):
    E = objectify.ElementMaker(annotate=False)
    anno_tree = E.annotation(
        E.folder('JPEGImages'),
        E.filename(filename),
	#E.path(path),
        E.size(
            E.width(w),
            E.height(h),
            E.depth(3)
        ),
        E.segmented(0),
    )
    for i in range(len(result)):
        E2 = objectify.ElementMaker(annotate=False)
        anno_tree2 = E2.object(
            E.name(result[i][0]),
            E.pose('Unspecified'),
            E.truncated('0'),
            E.difficult('0'),
            E.bndbox(
                E.xmin(result[i][1]),
		E.xmax(result[i][2]),
                E.ymin(result[i][3]),
                E.ymax(result[i][4])
            )
        )
	print "i = %d " % i
        anno_tree.append(anno_tree2)
    etree.ElementTree(anno_tree).write(path, pretty_print=True)

files = os.listdir(path)  # 得到文件夹下所有r文件名称

for xmlFile in files:
    # 遍历文件夹
    portion = os.path.splitext(xmlFile)
    rect_coordinates = []
    if not os.path.isdir(xmlFile):
        # 判断是否是文件夹,不是文件夹才打开
        # print (xmlFile)
 
        # xml文件读取操作
 
        # 将获取的xml文件名送入到dom解析
	print(xmlFile)
        dom = xml.dom.minidom.parse(os.path.join(path, xmlFile))
        ###最核心的部分os.path.join(path,xmlFile),路径拼接,输入的是具体路径
        root = dom.documentElement
        xmin = root.getElementsByTagName('xmin')
        ymin = root.getElementsByTagName('ymin')
        xmax = root.getElementsByTagName('xmax')
        ymax = root.getElementsByTagName('ymax')
	width = root.getElementsByTagName('width')
	height = root.getElementsByTagName('height')
	name = root.getElementsByTagName('name')

        # pose=root.getElementsByTagName('pose')
        # 重命名class name
	if(0 == len(xmin)):
	    continue
        for i in range(len(xmin)):
            # print (name[i].firstChild.data)
            # print(xmlFile)
            # if portion[1] == ".xml":
            #     newname = portion[0] + ".jpg"
            #     print(newname)
	    rect_coordinate = []
            area  = ((int)(xmax[i].firstChild.data)-(int)(xmin[i].firstChild.data))*((int)(ymax[i].firstChild.data)-(int)(ymin[i].firstChild.data))
            
            if area < 10000:
 		print area
                continue
	    else:
		rect_coordinate.append(name[i].firstChild.data)
		rect_coordinate.append(xmin[i].firstChild.data)
		rect_coordinate.append(xmax[i].firstChild.data)
		rect_coordinate.append(ymin[i].firstChild.data)
		rect_coordinate.append(ymax[i].firstChild.data)
                rect_coordinates.append(rect_coordinate)
    	gen_txt(width[0].firstChild.data,height[0].firstChild.data,rect_coordinates,xmlFile,os.path.join(outpath, xmlFile))
            # 保存修改到xml文件中
        #with open(os.path.join(outpath, xmlFile), 'w') as fh:
            #dom.writexml(fh)
            #print('修改filename OK!')

caffe测试python脚本

caffe前向推理脚本:

import numpy as np  
import sys,os  
import cv2
caffe_root = './caffe_ssd/'  # this file should be run from {caffe_root}/examples (otherwise change this line)
sys.path.insert(0, caffe_root + 'python')
import caffe  

net_file= './MobileNetSSD_deploy.prototxt'  
caffe_model='./mobilenet_v2_300x300_iter_2000.caffemodel'  

test_dir = "/media/t/disk1/chengyun/vehicle_data_v2/kakou/JPEGImages"

if not os.path.exists(caffe_model):
    print(caffe_model + " does not exist")
    exit()
if not os.path.exists(net_file):
    print(net_file + " does not exist")
    exit()
net = caffe.Net(net_file, caffe_model, caffe.TEST)  

CLASSES = ('background',
           'car')


def preprocess(src):
    img = cv2.resize(src, (300,300))
    #img = img - np.array([104.0, 117.0, 123.0])
    img = img - 127.5
    img = img * 0.007843
    return img

def postprocess(img, out):   
    h = img.shape[0]
    w = img.shape[1]
    box = out['detection_out'][0,0,:,3:7] * np.array([w, h, w, h])

    cls = out['detection_out'][0,0,:,1]
    conf = out['detection_out'][0,0,:,2]
    return (box.astype(np.int32), conf, cls)

def detect(imgfile):
    origimg = cv2.imread(imgfile)
    img = preprocess(origimg)
    
    img = img.astype(np.float32)
    img = img.transpose((2, 0, 1))

    net.blobs['data'].data[...] = img
    out = net.forward()  
    box, conf, cls = postprocess(origimg, out)

    for i in range(len(box)):
       p1 = (box[i][0], box[i][1])
       p2 = (box[i][2], box[i][3])
       width = box[i][2] - box[i][0]
       height = box[i][3] - box[i][1]
       #cv2.rectangle(origimg, p1, p2, (255,0,0), 2)
       p3 = (max(p1[0], 15), max(p1[1], 15))
       cv2.rectangle(origimg, p1, p2, (255,0,0), 2)
       title = "%s:%.2f" % (CLASSES[int(cls[i])], conf[i])
       cv2.putText(origimg, title, p3, cv2.FONT_ITALIC, 1, (0, 255, 0), 2)
    #cv2.imshow("SSD", origimg)
    cv2.imwrite(os.path.join("imgs_res", os.path.basename(imgfile)), origimg)
    #k = cv2.waitKey(0) & 0xff
        #Exit if ESC pressed
    #if k == 27 : return False
    return True

for f in os.listdir(test_dir):
    if detect(test_dir + "/" + f) == False:
       break

将图片分成白天和晚上

#!/usr/bin/env python
# set up Python environment: numpy for numerical routines, and matplotlib for plotting
#-*- coding: utf-8 -*-

import glob

import time
import numpy as np
import sys
import cv2
#from pascal_voc_writer import Writer
#import matplotlib.pyplot as plt
# display plots in this notebook
import argparse
# set display defaults
#plt.rcParams['figure.figsize'] = (10, 10)        # large images
#plt.rcParams['image.interpolation'] = 'nearest'  # don't interpolate: show square pixels
#plt.rcParams['image.cmap'] = 'gray'  # use grayscale output rather than a (potentially misleading) color heatmap

# The caffe module needs to be on the Python path;
#  we'll add it here explicitly.
import sys
import os
#import python
import math
from os import walk
from os.path import join
# If you get "No module named _caffe", either you have not built pycaffe or you have the wrong path.

def get_image_list(image_dir, suffix=['jpg','JPG','png']):
    '''get all image path ends with suffix'''
    if not os.path.exists(image_dir):
        print("PATH:%s not exists" % image_dir)
        return []
    imglist = []
    filenames = []
    for root, sdirs, files in os.walk(image_dir):
        if not files:
            continue
        for filename in files:
            filepath = os.path.join(root, filename)
            if filename.split('.')[-1] in suffix:
                imglist.append(filepath)
		filenames.append(filename)
    return imglist,filenames

def is_night_or_day(img,thres0=40,thres1=100):
    img_BGR = img.copy()
    img_HSV = cv2.cvtColor(img_BGR,cv2.COLOR_BGR2HSV)
    img_H,img_S,img_V = cv2.split(img_HSV)
    average_V = np.sum(np.reshape(img_V,(img_V.size,))) / img_V.size
    if (average_V < thres0):
	return 0
    elif (average_V > thres1):
        return 1
    else:
	return 2

if __name__ == '__main__':
    image_src = '/media/t/disk1/jie_shou_ren_ming/chengyun/vehicle_data_v2/kakou/JPEGImages'
    image_src_path , filenames = get_image_list(image_src, suffix=['jpg','JPG','png'])
    for i in range(len(image_src_path)):
        image_show =cv2.imread(image_src_path[i])
	dst_img_name = filenames[i]
	if (is_night_or_day(image_show) == 0):
	    cv2.imwrite('/media/t/disk1/jie_shou_ren_ming/chengyun/vehicle_data_v2/kakou/output'+'/night/'+dst_img_name,image_show)
	elif (is_night_or_day(image_show) == 1):
	    cv2.imwrite('/media/t/disk1/jie_shou_ren_ming/chengyun/vehicle_data_v2/kakou/output'+'/day/'+dst_img_name,image_show)
	else:
	    cv2.imwrite('/media/t/disk1/jie_shou_ren_ming/chengyun/vehicle_data_v2/kakou/output'+'/others/'+dst_img_name,image_show)

截取ROI区域的图片

#encoding:utf-8
import cv2
import argparse
import sys
import os
import xml.dom.minidom

path =  os.path.abspath('.')

def parse_args():
    """
    Parse input arguments
    """
    parser = argparse.ArgumentParser(description='save classes to folder')
    parser.add_argument('--xml', dest='xml', type=str)
    parser.add_argument('--pic', dest='pic', type=str)

    if len(sys.argv) == 1:
        parser.print_help()
        sys.exit(1)

    args = parser.parse_args()
    return args

def ReadXml(xmlfile):
    dom = xml.dom.minidom.parse(xmlfile)        #打开xml文档
    root = dom.documentElement               #得到xml文档对象

    name = root.getElementsByTagName('name')

    xmin = root.getElementsByTagName('xmin')
    ymin = root.getElementsByTagName('ymin')
    xmax = root.getElementsByTagName('xmax')
    ymax = root.getElementsByTagName('ymax')

    results = []

    for i in range(len(name)):
        imgbox = [int(xmin[i].firstChild.data),int(ymin[i].firstChild.data),int(xmax[i].firstChild.data),int(ymax[i].firstChild.data)]
        results.append({'name':name[i].firstChild.data.encode("utf-8"), 'imgbox':imgbox})

    return results

def saveimg(dir,picfile):
    img = cv2.imread(picfile)
 #   if img == None :
 #       print None
 #       return "0"
    xml_class = dir['name']

    xml_class_path = os.path.join(path , 'classes1',xml_class)
    pic_class = os.path.join(xml_class_path,os.path.splitext(x)[0] + ".jpg")

    if not os.path.exists(xml_class_path):
        os.mkdir(xml_class_path)
        class_img = img[dir['imgbox'][1]:dir['imgbox'][3], dir['imgbox'][0]:dir['imgbox'][2]]
        cv2.imwrite(pic_class,class_img)
    else:
        class_img = img[dir['imgbox'][1]:dir['imgbox'][3], dir['imgbox'][0]:dir['imgbox'][2]]
        cv2.imwrite(pic_class, class_img)

if __name__ == '__main__':

    args = parse_args()
    print('Called with args:')
    print(args)

    pic_folder = args.pic
    xml_folder = args.xml

    pic_folder_path = os.path.join(path,pic_folder)
    xml_folder_path = os.path.join(path,xml_folder)

    if not os.path.exists(os.path.join(path, 'classes1')):
        os.mkdir(os.path.join(path, 'classes1'))
    else:
        print "此目录已存在！"
        sys.exit(1)

    for x in os.listdir(xml_folder_path) :

        xml_file_path = os.path.join(xml_folder_path,x)
        pic_file_path = os.path.join(pic_folder_path,os.path.splitext(x)[0]  + ".jpg")
        results = ReadXml(xml_file_path)

        print xml_file_path
        print pic_file_path

        for num in xrange(len(results)) :
            saveimg(results[num],pic_file_path)
            
        print "This file has done! "
        print "_______________________________________"

    print 'Done'

xml转txt文件的脚本

我们知道使用labelimg标注好的文件格式是.xml文件 如果我们要使用yolov5就需要把它转换成yolov5需要的文件格式 .txt （这是重点） 代码附下：

# -*- coding: utf-8 -*-
import os
import xml.etree.ElementTree as ET
 
dirpath = r'D:\pythonProject1\yolov5-6.0\bottle_dataset\stronger\xml'  # 原来存放xml文件的目录
newdir = r'D:\pythonProject1\yolov5-6.0\bottle_dataset\stronger\labels'  # 修改label后形成的txt目录
 
if not os.path.exists(newdir):
    os.makedirs(newdir)
 
dict_info = {'green': 0, 'transparent': 1, 'white': 2, 'blue': 3, 'unknown': 4, 'orange': 5}  # 有几个 类别 填写几个label names
 
for fp in os.listdir(dirpath):
    if fp.endswith('.xml'):
        root = ET.parse(os.path.join(dirpath, fp)).getroot()
 
        xmin, ymin, xmax, ymax = 0, 0, 0, 0
        sz = root.find('size')
        width = float(sz[0].text)
        height = float(sz[1].text)
        filename = root.find('filename').text
        for child in root.findall('object'):  # 找到图片中的所有框
 
            sub = child.find('bndbox')  # 找到框的标注值并进行读取
            label = child.find('name').text
            label_ = dict_info.get(label)
            if label_:
                label_ = label_
            else:
                label_ = 0
            xmin = float(sub[0].text)
            ymin = float(sub[1].text)
            xmax = float(sub[2].text)
            ymax = float(sub[3].text)
            try:  # 转换成yolov3的标签格式，需要归一化到（0-1）的范围内
                x_center = (xmin + xmax) / (2 * width)
                x_center = '%.6f' % x_center
                y_center = (ymin + ymax) / (2 * height)
                y_center = '%.6f' % y_center
                w = (xmax - xmin) / width
                w = '%.6f' % w
                h = (ymax - ymin) / height
                h = '%.6f' % h
            except ZeroDivisionError:
                print(filename, '的 width有问题')
            with open(os.path.join(newdir, fp.split('.xml')[0] + '.txt'), 'a+') as f:
                f.write(' '.join([str(label_), str(x_center), str(y_center), str(w), str(h) + '\n']))
print('ok')

代码只需要更改5，6行的文件路径和第11行你所标注的类别即可。

注：.xml转化成.txt文件放入labels内。

注：我们用到的所有照片放入JPEGImage内。

生成的.txt内容如下（以我的数据为例）第一列是设定的标签，后面是坐标位置

yolo的txt标签文件转换成voc的xml文件脚本

网上很多xml转txt格式的，却没有txt转xml格式的，自己写了脚本如下：
txt标签文件格式

xml文件格式

转换代码如下：

#coding=utf-8
#makexml("txt所在文件夹","xml保存地址","图片所在地址")

from xml.dom.minidom import Document
import os
import cv2

def makexml(txtPath,xmlPath,picPath): #读取txt路径，xml保存路径，数据集图片所在路径
        dict = {'0': "person",#字典对类型进行转换
                '1': "car",
                '2': "bus",
                '3': "truck"}
        files = os.listdir(txtPath)
        for i, name in enumerate(files):
          xmlBuilder = Document()
          annotation = xmlBuilder.createElement("annotation")  # 创建annotation标签
          xmlBuilder.appendChild(annotation)
          txtFile=open(txtPath+name,'r',encoding='iso8859-1')
          txtList = txtFile.readlines()
          img = cv2.imread(picPath+name[0:-4]+".jpg")
          print(picPath+name[0:-4]+".jpg")
          Pheight,Pwidth,Pdepth=img.shape
          #for i in txtList:
          #oneline = i.strip().split(" ")

          folder = xmlBuilder.createElement("folder")#folder标签
          folderContent = xmlBuilder.createTextNode("VOC2007")
          folder.appendChild(folderContent)
          annotation.appendChild(folder)

          filename = xmlBuilder.createElement("filename")#filename标签
          filenameContent = xmlBuilder.createTextNode(name[0:-4]+".jpg")
          filename.appendChild(filenameContent)
          annotation.appendChild(filename)

          size = xmlBuilder.createElement("size")# size标签
          width = xmlBuilder.createElement("width")  # size子标签width
          widthContent = xmlBuilder.createTextNode(str(Pwidth))
          width.appendChild(widthContent)
          size.appendChild(width)
          height = xmlBuilder.createElement("height") # size子标签height
          heightContent = xmlBuilder.createTextNode(str(Pheight))
          height.appendChild(heightContent)
          size.appendChild(height)
          depth = xmlBuilder.createElement("depth") # size子标签depth
          depthContent = xmlBuilder.createTextNode(str(Pdepth))
          depth.appendChild(depthContent)
          size.appendChild(depth)
          annotation.appendChild(size)

          for i in txtList:
             oneline = i.strip().split(" ")
             object = xmlBuilder.createElement("object")
             picname = xmlBuilder.createElement("name")
             if oneline[0]>='4':
                continue
             nameContent = xmlBuilder.createTextNode(dict[oneline[0]])
             picname.appendChild(nameContent)
             object.appendChild(picname)
             pose = xmlBuilder.createElement("pose")
             poseContent = xmlBuilder.createTextNode("Unspecified")
             pose.appendChild(poseContent)
             object.appendChild(pose)
             truncated = xmlBuilder.createElement("truncated")
             truncatedContent = xmlBuilder.createTextNode("0")
             truncated.appendChild(truncatedContent)
             object.appendChild(truncated)
             difficult = xmlBuilder.createElement("difficult")
             difficultContent = xmlBuilder.createTextNode("0")
             difficult.appendChild(difficultContent)
             object.appendChild(difficult)
             bndbox = xmlBuilder.createElement("bndbox")
             xmin = xmlBuilder.createElement("xmin")
             mathData=int(((float(oneline[1]))*Pwidth+1)-(float(oneline[3]))*0.5*Pwidth)
             if mathData<0:
                mathData= 0
             xminContent = xmlBuilder.createTextNode(str(mathData))
             xmin.appendChild(xminContent)
             bndbox.appendChild(xmin)
             ymin = xmlBuilder.createElement("ymin")
             mathData = int(((float(oneline[2]))*Pheight+1)-(float(oneline[4]))*0.5*Pheight)
             if mathData<0:
                mathData= 0
             yminContent = xmlBuilder.createTextNode(str(mathData))
             ymin.appendChild(yminContent)
             bndbox.appendChild(ymin)
             xmax = xmlBuilder.createElement("xmax")
             mathData = int(((float(oneline[1]))*Pwidth+1)+(float(oneline[3]))*0.5*Pwidth)
             if mathData > Pwidth:
                mathData= Pwidth
             xmaxContent = xmlBuilder.createTextNode(str(mathData))
             xmax.appendChild(xmaxContent)
             bndbox.appendChild(xmax)
             ymax = xmlBuilder.createElement("ymax")
             mathData = int(((float(oneline[2]))*Pheight+1)+(float(oneline[4]))*0.5*Pheight)
             if mathData > Pheight:
                mathData= Pheight
             ymaxContent = xmlBuilder.createTextNode(str(mathData))
             ymax.appendChild(ymaxContent)
             bndbox.appendChild(ymax)
             object.appendChild(bndbox)

             annotation.appendChild(object)

          f = open(xmlPath+name[0:-4]+".xml", 'w')
          xmlBuilder.writexml(f, indent='\t', newl='\n', addindent='\t', encoding='utf-8')
          f.close()

if __name__ == "__main__":
	txt_path = '/root/zhangsong/huawei/data/zsmodify_data/txt/'
	xml_path = '/root/zhangsong/huawei/data/zsmodify_data/Annotations/'
	image_path = '/root/zhangsong/huawei/data/zsmodify_data/JPEGImages/'
	
	makexml(txt_path,xml_path,image_path)

python遍历获取文件：for (root, dirs, files) in walk(roots)：

注意事项：这是个遍历类似于生成器的感觉，一层一层遍历直接上代码，帮助大家清晰理解：
测试代码如下

path = 'F:/KuGou'
def file_name(file_dir):
    a = 1
    for root, dirs, files in os.walk(file_dir):#遍历文件夹、根目录、目录文件夹、目录里的文件
        print(root)
        print("*****************")
        print(dirs)
        print("************")
        print(files)
        print(a)
        a +=1
file_name(path)

结果如下

F:/KuGou
*****************
['01', '02']
************
[]
1
F:/KuGou\01
*****************
[]
************
['1.png', '2.png']
2
F:/KuGou\02
*****************
[]
************
['2018120618453717793.pdf']
3

删除爬虫出来的小图片

爬虫爬了大量的图片，但的图片大小残次不齐，下面的代码目的是删除小于20K的图片
这里图片size小于250*250的就会删除，最后计算一个小于20K文件的百分比。但第一次运行的时候先不要运行remove语句，毕竟20K的闸值需要你自己设定的

import os

'''
DirList = [
    '/home/king/PycharmProjects/nsfw_data_scrapper/raw_data/drawings',
    '/home/king/PycharmProjects/nsfw_data_scrapper/raw_data/hentai',
    '/home/king/PycharmProjects/nsfw_data_scrapper/raw_data/neutral',
    '/home/king/PycharmProjects/nsfw_data_scrapper/raw_data/porn',
    '/home/king/PycharmProjects/nsfw_data_scrapper/raw_data/sexy']
'''
DirList = [
    'E:\\projection\\fair\\data\\dataset\\open_fire']


for path in DirList:
    print(path)
    tall = 0
    small = 0
    for filename in os.listdir(path):
        fullName = os.path.join(path, filename)
        size = os.path.getsize(fullName)
        if size < 250 * 250:
            small = small + 1
            os.remove(fullName)
        tall = tall + 1
    print(tall, small, small/tall * 100)

python脚本文件操作删除满足一定条件的文件内容（darknet标签文件修改）

#删除类型为0（火）的标签
with open(label_file,"r",encoding="utf-8") as f:
    read_lines = f.readlines()
    #print(lines)
with open(label_file,"w",encoding="utf-8") as f_w:
    for read_line in read_lines:
        if int(read_line.split()[0]) == 0:
            continue   
        else:
            #类型为0的标签之外，所有标签类别减1
            modefy_read_line_list = list(read_line)
            if int(modefy_read_line_list[0]) > 0 :
                modefy_read_line_list[0] = str(int(modefy_read_line_list[0]) -1)
                read_line = ''.join(modefy_read_line_list)
        f_w.write(read_line)   

opencv-python图片基本操作

scr_img = cv2.imread(image_path）
#scr_img.shape[0] #图像的高度
#scr_img.shape[1] #图像的宽度
xmins = xmins * scr_img.shape[1]
ymins = ymins * scr_img.shape[0]
xmaxs = xmaxs * scr_img.shape[1]
ymaxs = ymaxs * scr_img.shape[0]
out_iou_img = scr_img[(int)(ymins):(int)(ymaxs),(int)(xmins):(int)(xmaxs)]     #截取ROI区域的图片
cv2.imwrite(output_file + 'cat_{}_{}.jpg'.format(i,objects_num),out_iou_img)   #保存图片

图片的复制粘贴（用于数据增强）

如果想将两张图像进行融合，只需再读取一张图像即可，方法原理类似。 实现代码如下：

# -*- coding:utf-8 -*-
import cv2
import numpy as np

#读取图片
img = cv2.imread("test.jpg", cv2.IMREAD_UNCHANGED)
test = cv2.imread("test3.jpg", cv2.IMREAD_UNCHANGED)

#定义300*100矩阵 3对应BGR
face = np.ones((200, 200, 3))

#显示原始图像
cv2.imshow("Demo", img)

#显示ROI区域
face = img[100:300, 150:350]
test[400:600,400:600] = face
cv2.imshow("Pic", test)

#等待显示
cv2.waitKey(0)
cv2.destroyAllWindows()

修改文件名

#-*- coding: utf-8 -*-
#!/usr/bin/env python
'''
USE Method:python ./modify_file_name.py                
'''
#-*- coding: utf-8 -*-
#!/usr/bin/env python
'''
USE Method:python ./modify_file_name.py                
'''
import os
import glob
import cv2
global count
def get_image_list(image_dir, suffix=['jpg','JPG','png']):
    '''get all image path ends with suffix'''
    if not os.path.exists(image_dir):
        print("PATH:%s not exists" % image_dir)
        return []
    imglist = []
    count = 0
    for root, sdirs, files in os.walk(image_dir):
        if not files:
            continue
        for filename in files:
            filepath = os.path.join(root, filename)
            print('*************')
	        #filename = filename + '.jpg'
            if filename.split('.')[-1] in suffix:
                newpath = os.path.join(root, 'crawler0_20210106_') + str(count) + '.jpg'
                os.rename(filepath,newpath)
                count += 1
    return imglist

if __name__ == '__main__':
    image_src = r'C:\Users\Administrator\Desktop\crawler0'
    image_src_path = get_image_list(image_src, suffix=['jpg','JPG','png'])

随机删除图片

如果图片有点多余，可以随机删除一定比例的图片，也可以删除对应的标签文件。

import os  
import io  
import math
import sys
import cv2
import shutil
import random
import numpy as np
from collections import namedtuple, OrderedDict  

label_names = ['person','car','bus','truck','motorcycle','chemical']

def get_files(dir, suffix): 
    res = []
    for root, directory, files in os.walk(dir):
        for filename in files:
            name, suf = os.path.splitext(filename) 

            if suf in suffix:
                #res.append(filename)
                res.append(os.path.join(root, filename))
    return res

def random_filter_image(list_path,random_rate):
    image_list = get_files(list_path, ['.jpg'])
    total_len = len(image_list)
    print('total_label_len', total_len)
    for i in range(0, total_len):
        gen_rate = random.random() 
        #'''
        if (gen_rate < random_rate):
            pass
        else:
            os.remove(image_list[i])
            continue
        #'''
        image_file = image_list[i]
        file_name, type_name = os.path.splitext(image_file)
        file_txt_name = file_name + '.txt'
        with open(file_txt_name,"w",encoding="utf-8") as f_w:
            pass

    random.shuffle(image_list)

def main():  
    global random_rate
    #list_path = r'C:\Users\Administrator\Desktop\negative_image\101\outsourcing_data\201210'
    list_path = r'E:\projection\fair\data\genarate_4to1\negative_image\101\data\20729'

    save_base_dir= r'E:\projection\cat'
    random_rate = 0.25
    random_filter_image(list_path,random_rate)

if __name__ == '__main__':  

    main()

文件重命名

#-*- coding: utf-8 -*-
#!/usr/bin/env python
'''
USE Method:python ./modify_file_name.py                
'''
import os
import glob
import cv2
global count
def get_image_list(image_dir, suffix=['jpg','JPG','png']):
    '''get all image path ends with suffix'''
    if not os.path.exists(image_dir):
        print("PATH:%s not exists" % image_dir)
        return []
    imglist = []
    count = 0
    for root, sdirs, files in os.walk(image_dir):
        if not files:
            continue
        for filename in files:
            filepath = os.path.join(root, filename)
            print('filename*************',filename)
            #print('root*************',root)
	        #filename = filename + '.jpg'
            if filename.split('.')[-1] in suffix:
                newpath = os.path.join(root, 'output') + str(count) + '.mp4'
                print('newpath*************',newpath)
                os.rename(filepath,newpath)
                count += 1
    return imglist

if __name__ == '__main__':
    image_src = r'E:\projection\fair\data\磁盘\output1'
    image_src_path = get_image_list(image_src, suffix=['mp4'])

标签过滤

darknet删除一些不满足要求的标签框，如标签坐标为负值。

import os  
import io  
import math
import sys
import random
import argparse

from collections import namedtuple, OrderedDict  

label_names = ['person','car','bus','truck']

def get_files(dir, suffix): 
    res = []
    for root, directory, files in os.walk(dir): 
        for filename in files:
            name, suf = os.path.splitext(filename) 
            if suf == suffix:
                #res.append(filename)
                res.append(os.path.join(root, filename))
    return res
def gbbox_iou(box1, box2):
    b1_x1, b1_y1, b1_x2, b1_y2 = box1
    b2_x1, b2_y1, b2_x2, b2_y2 = box2

    inter_rect_x1 = max(b1_x1, b2_x1)
    inter_rect_y1 = max(b1_y1, b2_y1)
    inter_rect_x2 = min(b1_x2, b2_x2)
    inter_rect_y2 = min(b1_y2, b2_y2)
   
    inter_width = inter_rect_x2 - inter_rect_x1 + 1
    inter_height = inter_rect_y2 - inter_rect_y1 + 1
    if inter_width > 0 and inter_height > 0:  
        inter_area = inter_width * inter_height
        #iou
        b1_area = (b1_x2 - b1_x1 + 1) * (b1_y2 - b1_y1 + 1)
        b2_area = (b2_x2 - b2_x1 + 1) * (b2_y2 - b2_y1 + 1)
        #iou = inter_area / (b1_area + b2_area - inter_area)
        iou = inter_area / b1_area

    else:
        iou = 0
    return iou

def convert_dataset(list_path, output_file):

    # 读取目录里面所有的 txt标记文件 列表
    label_list = get_files(list_path, '.txt')
    total_label_len = len(label_list)
    random.shuffle(label_list)
    print('total_label_len', total_label_len)
    error_count = 0
    fp=open(output_file,'w')

    for i in range(0, total_label_len):
        sys.stdout.write('\r>> Calculating {}/{} error{}'.format(
            i + 1, total_label_len, error_count))
        sys.stdout.flush()
        
        # 单个Label txt文件读取
        label_file = label_list[i]
        file_name, type_name = os.path.splitext(label_file)
        image_path = file_name + '.jpg'
        if type_name != '.txt' or not os.path.exists(image_path):
            error_count += 1
            print("error_file: ",label_file.encode('UTF-8', 'ignore').decode('UTF-8'))
            continue
        fd = open(label_file, 'r')
        lines = [line.split() for line in fd]
        fd.close()
        error_id = 0
        for line in lines:  
            class_index = int(line[0])
            xmins = float(line[1]) - float(line[3]) / 2
            ymins = float(line[2]) - float(line[4]) / 2
            xmaxs = float(line[1]) + float(line[3]) / 2                             
            ymaxs = float(line[2]) + float(line[4]) / 2 
            if  float(line[3])<=0 or float(line[4]) <= 0 :
                error_id = 1
                print('\n error index: ', class_index, 'label_file', label_file)
                continue
            if class_index >= 3:
                error_id = 1
                print('\n error index: ', class_index, 'label_file', label_file)
                continue
            # if xmins < 0 or ymins < 0 :
            #     error_id = 1
            #     print('\n error index: ', class_index, 'label_file', label_file)
            # if  ymaxs > 1  or xmaxs > 1 :
            #     print('\n error index: ', class_index, 'label_file', label_file)
            #     error_id = 1

        if error_id:
            continue
        # is_person_car = False            
        # bbox_num = len(lines)
        # for  i in range(0, bbox_num):

        #     if int(lines[i][0]) != 0:
        #         continue
        #     for j in range(0, bbox_num):

        #         if i==j or int(lines[j][0])==0:
        #             continue
        #         xmins = float(lines[i][1]) - float(lines[i][3]) / 2
        #         ymins = float(lines[i][2]) - float(lines[i][4]) / 2
        #         xmaxs = float(lines[i][1]) + float(lines[i][3]) / 2                             
        #         ymaxs = float(lines[i][2]) + float(lines[i][4]) / 2 

        #         xmins1 = float(lines[j][1]) - float(lines[j][3]) / 2
        #         ymins1 = float(lines[j][2]) - float(lines[j][4]) / 2
        #         xmaxs1 = float(lines[j][1]) + float(lines[j][3]) / 2                             
        #         ymaxs1 = float(lines[j][2]) + float(lines[j][4]) / 2 

        #         box1 = (xmins, ymins, xmaxs, ymaxs)
        #         box2 = (xmins1, ymins1, xmaxs1, ymaxs1)
        #         #过滤行人在车中
        #         iou = gbbox_iou(box1, box2)
        #         if iou > 0.99:
        #            is_person_car = True
        # if  is_person_car:
        #     continue
        print("image_path: ", image_path)
        fp.write(image_path)
        fp.write('\n')
    print('total_label_len', total_label_len)
    fp.close()


def main(): 

    parser = argparse.ArgumentParser(prog='gen_label_list.py')
    parser.add_argument('--img-path', type=str, default='/root/zhangsong/fairworks/github/darknet-master/fireworks/data/smoke', help='test path')
    parser.add_argument('--valid', type=str, default='fireworks/data/test_train.txt', help='*.txt path')
    opt = parser.parse_args()

    print(opt.img_path, opt.valid)
    convert_dataset(opt.img_path, opt.valid)
  
  
if __name__ == '__main__':  
    main()

darknet感兴趣区域截取

import os  
import io  
import math
import sys
import random
import argparse
import cv2

from collections import namedtuple, OrderedDict  

label_names = ['person','car','bus','truck']

def get_files(dir, suffix): 
    res = []
    for root, directory, files in os.walk(dir): 
        for filename in files:
            name, suf = os.path.splitext(filename) 
            if suf == suffix:
                #res.append(filename)
                res.append(os.path.join(root, filename))
    return res
    
def gbbox_iou(box1, box2):
    b1_x1, b1_y1, b1_x2, b1_y2 = box1
    b2_x1, b2_y1, b2_x2, b2_y2 = box2

    inter_rect_x1 = max(b1_x1, b2_x1)
    inter_rect_y1 = max(b1_y1, b2_y1)
    inter_rect_x2 = min(b1_x2, b2_x2)
    inter_rect_y2 = min(b1_y2, b2_y2)
   
    inter_width = inter_rect_x2 - inter_rect_x1 + 1
    inter_height = inter_rect_y2 - inter_rect_y1 + 1
    if inter_width > 0 and inter_height > 0:  
        inter_area = inter_width * inter_height
        #iou
        b1_area = (b1_x2 - b1_x1 + 1) * (b1_y2 - b1_y1 + 1)
        b2_area = (b2_x2 - b2_x1 + 1) * (b2_y2 - b2_y1 + 1)
        #iou = inter_area / (b1_area + b2_area - inter_area)
        iou = inter_area / b1_area
    else:
        iou = 0
    return iou

def convert_dataset(list_path, output_file):
    # 读取目录里面所有的 txt标记文件 列表
    label_list = get_files(list_path, '.txt')
    total_label_len = len(label_list)
    random.shuffle(label_list)
    print('total_label_len', total_label_len)
    error_count = 0

    for i in range(0, total_label_len):
        sys.stdout.write('\r>> Calculating {}/{} error{}\n'.format(
            i + 1, total_label_len, error_count))
        sys.stdout.flush()
        
        # 单个Label txt文件读取
        label_file = label_list[i]
        file_name, type_name = os.path.splitext(label_file)
        #print(file_name)
        #print(file_name.split('\\')[-1] + 'aaaaaa')
        cut_image_name_list = file_name.split('\\')[-2:] #cut_image_name_list is list
        cut_image_name = ''.join(cut_image_name_list) #list to str
        image_path = file_name + '.jpg'
        if type_name != '.txt' or not os.path.exists(image_path):
            error_count += 1
            print("error_file: ",label_file.encode('UTF-8', 'ignore').decode('UTF-8'))
            continue

        scr_img = cv2.imread(image_path) 
        #'''
        #删除类型为0（火）的标签
        with open(label_file,"r",encoding="utf-8") as f:
            read_lines = f.readlines()
            #print(lines)
        with open(label_file,"w",encoding="utf-8") as f_w:
            for read_line in read_lines:
                if int(read_line.split()[0]) == 0:
                    continue   
                else:
                    #所有标签类别减1
                    modefy_read_line_list = list(read_line)
                    if int(modefy_read_line_list[0]) > 0 :
                        modefy_read_line_list[0] = str(int(modefy_read_line_list[0]) -1)
                        read_line = ''.join(modefy_read_line_list)
                #删除标签矩形长宽及面积比较小的图片
                labels_coordi = read_line.split()
                if float(labels_coordi[3]) * scr_img.shape[1] < 30.0 :
                    continue
                if float(labels_coordi[4]) * scr_img.shape[0] < 30.0 :
                    continue
                
                f_w.write(read_line)   
        #'''    

        #截取图片
        fd = open(label_file, 'r')
         
        lines = [line.split()  for line in fd]
        fd.close()

        error_id = 0
        
        objects_num = 0
        #newlabels = []
        for line in lines: 
            '''
            if int(line[0]) == 0:
                continue
            else:
                newlabels
                '''
            class_index = int(line[0])
            xmins = float(line[1]) - float(line[3]) / 2
            ymins = float(line[2]) - float(line[4]) / 2
            xmaxs = float(line[1]) + float(line[3]) / 2                             
            ymaxs = float(line[2]) + float(line[4]) / 2 
            if  float(line[3])<=0 or float(line[4]) <= 0 :
                error_id = 1
                print('\n error index: ', class_index, 'label_file', label_file)
                continue
            if class_index >= 3: 
                error_id = 1
                print('\n error index: ', class_index, 'label_file', label_file)
                continue
            if xmins < 0 :
                xmins = 0 
            if ymins < 0 :
                ymins = 0
            if ymaxs > 1 :
                ymaxs = 1
            if xmaxs > 1 :
                xmaxs = 1
            xmins = xmins * scr_img.shape[1]
            ymins = ymins * scr_img.shape[0]
            xmaxs = xmaxs * scr_img.shape[1]
            ymaxs = ymaxs * scr_img.shape[0]
            out_iou_img = scr_img[(int)(ymins):(int)(ymaxs),(int)(xmins):(int)(xmaxs)]
            '''
            #实现感兴趣区域ROI的复制粘贴
            temp_iou_img = scr_img[0:80,0:200]
            scr_img[100:180,100:300] = temp_iou_img
            cv2.imwrite(output_file + 'zhantie_{}_{}.jpg'.format(i,objects_num),scr_img)
            '''
            
            cv2.imwrite(output_file + '{}_{}.jpg'.format(cut_image_name,objects_num),out_iou_img)
            objects_num += 1
        if error_id:
            continue
        
    print('total_label_len', total_label_len)


def main(): 

    parser = argparse.ArgumentParser(prog='gen_label_list.py')
    parser.add_argument('--img-path', type=str, default='E:\\projection\\fair\\data\\smoke', help='test path')
    parser.add_argument('--out-path', type=str, default='E:\\projection\\forestout\\', help='*.txt path')
    opt = parser.parse_args()

    print(opt.img_path, opt.out_path)
    convert_dataset(opt.img_path, opt.out_path)
  
if __name__ == '__main__':  
    main()

统一图片大小

import os  
import io  
import math
import sys
import cv2
import shutil
import random
import numpy as np
from collections import namedtuple, OrderedDict  

label_names = ['person','car','bus','truck','motorcycle','chemical']

def get_files(dir, suffix): 

    res = []

    for root, directory, files in os.walk(dir): 

        for filename in files:

            name, suf = os.path.splitext(filename) 

            if suf in suffix:
                #res.append(filename)

                res.append(os.path.join(root, filename))
    return res

def uniform_image_size(list_path,width_size,height_size):
    image_list = get_files(list_path, ['.jpg'])
    total_len = len(image_list)
    print('total_label_len', total_len)
    for i in range(0, total_len):
        image_file = image_list[i]
        img = cv2.imread(image_file)
        if img.shape[0] != height_size or img.shape[1] != width_size:
            img = cv2.resize(img, (width_size,height_size), interpolation=cv2.INTER_LINEAR)
            os.remove(image_file)
            cv2.imwrite(image_file, img)
        print('img.shape[0]',img.shape[0],'--------',image_file)

    random.shuffle(image_list)

def main():  

    list_path = r'E:\projection\fair\data\genarate_4to1\negative_image\101'

    width_size = 1280
    height_size = 720

    uniform_image_size(list_path,width_size,height_size)

if __name__ == '__main__':  

    main()

图片打mask

import os
import io
import math
import sys
import cv2
import shutil
import random
import numpy as np
from collections import namedtuple, OrderedDict

label_names = ['person', 'car', 'bus', 'truck', 'motorcycle', 'chemical']

def get_files(dir, suffix):
    res = []

    for root, directory, files in os.walk(dir):

        for filename in files:

            name, suf = os.path.splitext(filename)

            if suf == suffix:
                # res.append(filename)

                res.append(os.path.join(root, filename))
    return res


def convert_dataset(list_path, save_base_dir):
    while True:
        width_rate = random.random()
        if (width_rate > 0.3 and width_rate < 0.7):
            break
        else:
            continue
    while True:
        height_rate = random.random()
        if (height_rate > 0.3 and height_rate < 0.7):
            break
        else:
            continue
    image_list = get_files(list_path, '.jpg')
    total_len = len(image_list)
    print('total_label_len', total_len)

    if not os.path.exists(save_base_dir):
        os.makedirs(save_base_dir)

    random.shuffle(image_list)

    error_count = 0

    image_count = 0

    for i in range(0, total_len):
        image_file = image_list[i]

        img = cv2.imread(image_file)

        img_h, img_w, img_c = img.shape

        print("img: ", img_w, img_h)

        img_name, img_type = os.path.splitext(image_file)

        _, image_name = os.path.split(img_name)

        image_count += 1

        img4 = np.full((img_h, img_w, img.shape[2]), 0, dtype=np.uint8)

        img4[700:1200, :] = img[700:1200, :]

        save_img = save_base_dir + '/' + image_name + str(image_count) + 'preprocess.jpg'

        cv2.imwrite(save_img, img4)


def main():
    list_path = r'F:\gongye\data\biaozhu\JPEGImages\IMG_3462'

    # base_dir = os.getcwd()

    save_base_dir = r'F:\gongye\data\biaozhu\JPEGImages\IMG_3462cutmask'

    # save_base_dir = os.path.join(save_base_dir, 'r554DergCvk4h00')

    convert_dataset(list_path, save_base_dir)


if __name__ == '__main__':
    main()

结果如下：

只列举了一些基本的python脚本,拿去用吧.