目标检测训练数据旋转python代码——数据增广(一)

转载请在首行附上原文链接!有帮助的话记得点赞大笑

针对目标检测任务,对训练数据做旋转进而达到数据增广的python2代码,网上没找到,自己写了一份,花了大半天时间。。。发帖不易,转发还请在首行附上原文链接。

另外附上一份,检查旋转后效果的Python代码(读取annotations中xml,画图的代码),在最下面

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

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

#获取路径下所有文件的完整路径,用于读取文件用
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
#图像旋转用,里面的angle是角度制的
def im_rotate(im,angle,center = None,scale = 1.0):
    h,w = im.shape[:2]
    if center is None:
        center = (w/2,h/2)
    M = cv2.getRotationMatrix2D(center,angle,scale)
    im_rot = cv2.warpAffine(im,M,(w,h))
    return im_rot

#读取xml文件,xmlfile参数表示xml的路径 
def readXml(xmlfile):
    DomTree = xml.dom.minidom.parse(xmlfile)  
    annotation = DomTree.documentElement  
    sizelist = annotation.getElementsByTagName('size') #[<DOM Element: filename at 0x381f788>]  
    heights = sizelist[0].getElementsByTagName('height')
    height = int(heights[0].childNodes[0].data)
    widths =sizelist[0].getElementsByTagName('width')
    width = int(widths[0].childNodes[0].data)
    depths = sizelist[0].getElementsByTagName('depth')
    depth = int(depths[0].childNodes[0].data)
    objectlist = annotation.getElementsByTagName('object')        
    bboxes = []
    for objects in objectlist:  
        namelist = objects.getElementsByTagName('name')  
        class_label = namelist[0].childNodes[0].data  
        bndbox = objects.getElementsByTagName('bndbox')[0]     
        x1_list = bndbox.getElementsByTagName('xmin')  
        x1 = int(float(x1_list[0].childNodes[0].data))  
        y1_list = bndbox.getElementsByTagName('ymin') 
        y1 = int(float(y1_list[0].childNodes[0].data)) 
        x2_list = bndbox.getElementsByTagName('xmax')  
        x2 = int(float(x2_list[0].childNodes[0].data))  
        y2_list = bndbox.getElementsByTagName('ymax')  
        y2 = int(float(y2_list[0].childNodes[0].data))
        #这里我box的格式【xmin,ymin,xmax,ymax,classname】
        bbox = [x1,y1,x2,y2,class_label]
        bboxes.append(bbox)
    return bboxes,width,height,depth
     
#写xml文件,参数中tmp表示路径,imgname是文件名(没有尾缀)ps有尾缀也无所谓
def writeXml(tmp, imgname, w, h, d, bboxes):  
    doc = Document()  
    #owner  
    annotation = doc.createElement('annotation')  
    doc.appendChild(annotation)  
    #owner  
    folder = doc.createElement('folder')  
    annotation.appendChild(folder)  
    folder_txt = doc.createTextNode("VOC2007")  
    folder.appendChild(folder_txt)  
  
    filename = doc.createElement('filename')  
    annotation.appendChild(filename)  
    filename_txt = doc.createTextNode(imgname)  
    filename.appendChild(filename_txt)  
    #ones#  
    source = doc.createElement('source')  
    annotation.appendChild(source)  
  
    database = doc.createElement('database')  
    source.appendChild(database)  
    database_txt = doc.createTextNode("My Database")  
    database.appendChild(database_txt)  
  
    annotation_new = doc.createElement('annotation')  
    source.appendChild(annotation_new)  
    annotation_new_txt = doc.createTextNode("VOC2007")  
    annotation_new.appendChild(annotation_new_txt)  
  
    image = doc.createElement('image')  
    source.appendChild(image)  
    image_txt = doc.createTextNode("flickr")  
    image.appendChild(image_txt) 
    #owner
    owner = doc.createElement('owner')  
    annotation.appendChild(owner)  
  
    flickrid = doc.createElement('flickrid')  
    owner.appendChild(flickrid)  
    flickrid_txt = doc.createTextNode("NULL")  
    flickrid.appendChild(flickrid_txt) 
    
    ow_name = doc.createElement('name')  
    owner.appendChild(ow_name)  
    ow_name_txt = doc.createTextNode("idannel")  
    ow_name.appendChild(ow_name_txt)
    #onee#  
    #twos#  
    size = doc.createElement('size')  
    annotation.appendChild(size)  
  
    width = doc.createElement('width')  
    size.appendChild(width)  
    width_txt = doc.createTextNode(str(w))  
    width.appendChild(width_txt)  
  
    height = doc.createElement('height')  
    size.appendChild(height)  
    height_txt = doc.createTextNode(str(h))  
    height.appendChild(height_txt)  
  
    depth = doc.createElement('depth')  
    size.appendChild(depth)  
    depth_txt = doc.createTextNode(str(d))  
    depth.appendChild(depth_txt)  
    #twoe#  
    segmented = doc.createElement('segmented')  
    annotation.appendChild(segmented)  
    segmented_txt = doc.createTextNode("0")  
    segmented.appendChild(segmented_txt)  
  
    for bbox in bboxes:
        #threes#  
        object_new = doc.createElement("object")  
        annotation.appendChild(object_new)  
        
        name = doc.createElement('name')  
        object_new.appendChild(name)  
        name_txt = doc.createTextNode(str(bbox[4]))  
        name.appendChild(name_txt)  
  
        pose = doc.createElement('pose')  
        object_new.appendChild(pose)  
        pose_txt = doc.createTextNode("Unspecified")  
        pose.appendChild(pose_txt)  
  
        truncated = doc.createElement('truncated')  
        object_new.appendChild(truncated)  
        truncated_txt = doc.createTextNode("0")  
        truncated.appendChild(truncated_txt)  
  
        difficult = doc.createElement('difficult')  
        object_new.appendChild(difficult)  
        difficult_txt = doc.createTextNode("0")  
        difficult.appendChild(difficult_txt)  
        #threes-1#  
        bndbox = doc.createElement('bndbox')  
        object_new.appendChild(bndbox)  
  
        xmin = doc.createElement('xmin')  
        bndbox.appendChild(xmin)  
        xmin_txt = doc.createTextNode(str(bbox[0]))
        xmin.appendChild(xmin_txt)  
  
        ymin = doc.createElement('ymin')  
        bndbox.appendChild(ymin)  
        ymin_txt = doc.createTextNode(str(bbox[1]))
        ymin.appendChild(ymin_txt)    
  
        xmax = doc.createElement('xmax')  
        bndbox.appendChild(xmax)  
        xmax_txt = doc.createTextNode(str(bbox[2]))
        xmax.appendChild(xmax_txt)  
        
        ymax = doc.createElement('ymax')  
        bndbox.appendChild(ymax)  
        ymax_txt = doc.createTextNode(str(bbox[3]))
        ymax.appendChild(ymax_txt)  

    tempfile = tmp +"%s.xml"%imgname  
    with open(tempfile, 'w') as f:
        f.write(doc.toprettyxml(indent='\t', encoding='utf-8'))
    return  

#voc路径
root = '/home/ygx/cascade-rcnn/data/VOCdevkit_car/VOC2007/10602+36000/'
img_dir = root + 'JPEGImages/'
anno_path = root + 'Annotations/' 
#存储新的anno位置
anno_new_path = root + 'NewAnnotations/' 
if not os.path.isdir(anno_new_path):
    os.makedirs(anno_new_path)
#读取原图全路径  
imgs_path=GetFileFromThisRootDir(img_dir)
#存储旋转后图片位置
pro_dir = root+'train_translate_scale_rotate/'
if not os.path.isdir(pro_dir):
    os.makedirs(pro_dir)
#旋转角的大小,整数表示逆时针旋转
angles = [1,359]#角度im_rotate用到的是角度制
angle_rad = [angle*math.pi/180.0 for angle in angles] #cos三角函数里要用到弧度制的    
j=0 # 计数用
angle_num = len(angles)
for img_path in imgs_path :
    #读取原图像
    im = cv2.imread(img_path)
    for i in range(angle_num):
        gt_new = []
        im_rot = im_rotate(im,angles[i])
        file_name = img_path.split('/')[-1][:-4]
        #画出旋转后图像
        cv2.imwrite(os.path.join(pro_dir,'P%s_%s.jpg'%(angles[i],file_name)),im_rot)
        anno = os.path.join(anno_path,'%s.xml'%file_name)
        #读取anno标签数据
        [gts,w,h,d] =readXml(anno) 
        #计算旋转后gt框四点的坐标变换
        [xc,yc] = [float(w)/2,float(h)/2]
        for gt in gts:
            #计算左上角点的变换
            x1 = (gt[0]-xc)*math.cos(angle_rad[i]) - (yc-gt[1])*math.sin(angle_rad[i]) + xc
            if x1<0 : x1=0
            if x1>w-1 : x1=w-1                        
            y1 = yc - (gt[0]-xc)*math.sin(angle_rad[i]) - (yc-gt[1])*math.cos(angle_rad[i]) 
            if y1<0 : y1=0         
            if y1>h-1 : y1=h-1 
            #计算右上角点的变换   
            x2 = (gt[2]-xc)*math.cos(angle_rad[i]) - (yc-gt[1])*math.sin(angle_rad[i]) + xc
            if x2<0 : x2=0
            if x2>w-1 : x2=w-1                        
            y2 = yc - (gt[2]-xc)*math.sin(angle_rad[i]) - (yc-gt[1])*math.cos(angle_rad[i])
            if y2<0 : y2=0         
            if y2>h-1 : y2=h-1 
            #计算左下角点的变换
            x3 = (gt[0]-xc)*math.cos(angle_rad[i]) - (yc-gt[3])*math.sin(angle_rad[i]) + xc
            if x3<0 : x3=0
            if x3>w-1 : x3=w-1                        
            y3 = yc - (gt[0]-xc)*math.sin(angle_rad[i]) - (yc-gt[3])*math.cos(angle_rad[i])
            if y3<0 : y3=0         
            if y3>h-1 : y3=h-1
            #计算右下角点的变换
            x4 = (gt[2]-xc)*math.cos(angle_rad[i]) - (yc-gt[3])*math.sin(angle_rad[i]) + xc
            if x4<0 : x4=0          
            if x4>w-1 : x4=w-1
            y4 = yc - (gt[2]-xc)*math.sin(angle_rad[i]) - (yc-gt[3])*math.cos(angle_rad[i])
            if y4<0 :y4=0      
            if y4>h-1 : y4=h-1
            xmin = min(x1,x2,x3,x4)
            xmax = max(x1,x2,x3,x4)
            ymin = min(y1,y2,y3,y4)
            ymax = max(y1,y2,y3,y4)
            #把因为旋转导致的特别小的 长线型的去掉
            w_new = xmax-xmin+1
            h_new = ymax-ymin+1
            ratio1 = float(w_new)/h_new
            ratio2 = float(h_new)/w_new
            if(1.0/6.0<ratio1<6 and 1.0/6.0<ratio2<6 and w_new>9 and h_new>9):
                classname = str(gt[4])
                gt_new.append([xmin,ymin,xmax,ymax,classname])
            #写出新的xml
            writeXml(anno_new_path,'P%s_%s'%(angles[i],file_name) , w, h, d, gt_new)
    j = j+1
    if j%100==0 : print '----%s----'%j
   

    
    

#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
Created on Tue Jul 18 11:07:38 2017

@author: cetc_ic
"""

from __future__ import division  
import os   
import xml.dom.minidom  
import cv2
  
root='/home/ygx/cascade-rcnn/data/VOCdevkit_car/VOC2007/'
#ImgPath = root+'JPEGImages'   
#AnnoPath = root+'Annotations'  
#ProcessedPath = root+'gt_show'

ImgPath = root+'train_translate_scale_rotate'   
AnnoPath = root+'NewAnnotations'  
ProcessedPath = root+'gt_show_rotate'


if not os.path.exists(ProcessedPath):  
    os.makedirs(ProcessedPath)  
  
label_list = os.listdir(AnnoPath)  

num = 0

for labelfile in label_list:
    xmlfile = os.path.join(AnnoPath,labelfile)
    imgfile = os.path.join(ImgPath,'%s.jpg' % labelfile[:-4])   
    
    DomTree = xml.dom.minidom.parse(xmlfile)  
    annotation = DomTree.documentElement  

#    filenamelist = annotation.getElementsByTagName('filename') #[<DOM Element: filename at 0x381f788>]  
#    filename = filenamelist[0].childNodes[0].data  
    objectlist = annotation.getElementsByTagName('object')        
    bboxes = []
    for objects in objectlist:  
        namelist = objects.getElementsByTagName('name')  
        class_label = namelist[0].childNodes[0].data  

        bndbox = objects.getElementsByTagName('bndbox')[0]
        
        x1_list = bndbox.getElementsByTagName('xmin')  
        x1 = int(float(x1_list[0].childNodes[0].data))
        y1_list = bndbox.getElementsByTagName('ymin')  
        y1 = int(float(y1_list[0].childNodes[0].data)) 
        x2_list = bndbox.getElementsByTagName('xmax')  
        x2 = int(float(x2_list[0].childNodes[0].data))
        y2_list = bndbox.getElementsByTagName('ymax')  
        y2 = int(float(y2_list[0].childNodes[0].data))
        
        bbox = [x1,y1,x2,y2,class_label]
        bboxes.append(bbox)
    
    img = cv2.imread(imgfile)
    for bbox in bboxes:
        x1 = bbox[0]
        y1 = bbox[1]
        x2 = bbox[2]
        y2 = bbox[3]
        
        if x1 < 0 : x1 = 0
        if y1 < 0 : y1 = 0
        if x2 > img.shape[1] - 1 : x2 = img.shape[1] - 1
        if y2 > img.shape[0] - 1 : y2 = img.shape[0] - 1
        
        cv2.rectangle(img,(x1,y1),(x2,y2),(255,255,0),2)
        #cv2.putText(img,bbox[4],(bbox[0],bbox[1]),0.6,cv2.FONT_HERSHEY_SIMPLEX,2)
    cv2.imwrite(os.path.join(ProcessedPath,'%s.jpg' % labelfile[:-4]),img)
    print labelfile

    num += 1
    print num

阅读更多
版权声明:本文为博主原创文章,未经博主允许不得转载。 https://blog.csdn.net/e01528/article/details/80685978
上一篇fater-rcnn绘制loss图
下一篇目标检测utils常用代码系列(不断更新ing)
想对作者说点什么? 我来说一句

没有更多推荐了,返回首页

关闭
关闭
关闭