147 lines
5.8 KiB
Python
Executable File
147 lines
5.8 KiB
Python
Executable File
|
|
from scipy.io import loadmat
|
|
import glob,os
|
|
import xml.dom.minidom
|
|
from tqdm import tqdm
|
|
from creat_xml_fordetection import make_xml_iccv
|
|
den_path = "../data_MPKFNet/data/VOC/VOCdevkit/VOCiccvtrain/ground_truth/"
|
|
voc_path = "../data_MPKFNet/data/VOC/VOCdevkit/VOCiccvtrain/Annotations/"
|
|
save_path = "../data_MPKFNet/data/VOC/VOCdevkit/VOCiccvtrain/Bird_detection/"
|
|
def get_mat_point (path):
|
|
mat_data = loadmat(path)
|
|
positions = mat_data['image_info'][0, 0][0, 0][0]
|
|
count = mat_data['image_info'][0][0][0][0][1][0][0]
|
|
return positions
|
|
def get_xml_box (path):
|
|
dom = xml.dom.minidom.parse(path)
|
|
objects = dom.getElementsByTagName("object")
|
|
all_boxes = []
|
|
for object in objects:
|
|
bndbox = object.getElementsByTagName('bndbox')[0]
|
|
xmin = bndbox.getElementsByTagName('xmin')[0]
|
|
ymin = bndbox.getElementsByTagName('ymin')[0]
|
|
xmax = bndbox.getElementsByTagName('xmax')[0]
|
|
ymax = bndbox.getElementsByTagName('ymax')[0]
|
|
|
|
xmin_data = int(xmin.childNodes[0].data)
|
|
ymin_data = int(ymin.childNodes[0].data)
|
|
xmax_data = int(xmax.childNodes[0].data)
|
|
ymax_data = int(ymax.childNodes[0].data)
|
|
all_boxes.append((xmin_data,ymin_data,xmax_data,ymax_data))
|
|
sizes = dom.getElementsByTagName("size")
|
|
for size in sizes:
|
|
w = size.getElementsByTagName('width')[0]
|
|
w_data = int(w.childNodes[0].data)
|
|
h = size.getElementsByTagName('height')[0]
|
|
h_data = int(h.childNodes[0].data)
|
|
|
|
return all_boxes ,w_data,h_data
|
|
|
|
def point_in_box (site,box):
|
|
flag = False
|
|
if (site[0]>=box[0]) and (site[0]<=box[2]) and (site[1]>=box[1]) and (site[1]<=box[3]):
|
|
flag = True
|
|
return flag
|
|
|
|
def ifsame_box(site1,site2,box):
|
|
centerx = (box[0]+box[2])/2
|
|
centery = (box[1]+box[3])/2
|
|
dis1 = ((site1[0]-centerx)*(site1[0]-centerx) +(site1[1]-centery)*(site1[1]-centery))**0.5
|
|
dis2 = ((site2[0] - centerx) * (site2[0] - centerx) + (site2[1] - centery) * (site2[1] - centery))**0.5
|
|
if dis1>dis2:
|
|
flag = True
|
|
else:
|
|
flag = False
|
|
return flag
|
|
def get_new(site,boxes,img_w,img_h):
|
|
min_dis = 10e10
|
|
min_index = 0
|
|
for i in range(len(boxes)):
|
|
centerx = (boxes[i][0] + boxes[i][2]) / 2
|
|
centery = (boxes[i][1] + boxes[i][3]) / 2
|
|
dis = ((site[0]-centerx)*(site[0]-centerx) +(site[1]-centery)*(site[1]-centery))**0.5
|
|
if dis < min_dis:
|
|
min_dis = dis
|
|
min_index = i
|
|
w = (boxes[min_index][2]-boxes[min_index][0])/2
|
|
h = (boxes[min_index][3]-boxes[min_index][1])/2
|
|
xmin = site[0]-w
|
|
if xmin <0:
|
|
xmin =0
|
|
xmax = site[0]+w
|
|
if xmax >img_w:
|
|
xmax = img_w-1
|
|
ymin = site[1]-h
|
|
if ymin<0:
|
|
ymin = 0
|
|
ymax = site[1]+h
|
|
if ymax >img_h:
|
|
ymax = img_h-1
|
|
return ((xmin,ymin,xmax,ymax))
|
|
|
|
|
|
|
|
|
|
|
|
if __name__ == '__main__':
|
|
xml_lists = os.listdir(voc_path)
|
|
for xml_path in tqdm(xml_lists) :
|
|
print(xml_path+"is doing")
|
|
mat_path = xml_path.replace(".xml",".mat").replace("HatIMG","GT_HatIMG")
|
|
ALL_points = get_mat_point(den_path+mat_path)
|
|
ALl_boxes, image_W,image_H = get_xml_box (voc_path+xml_path)
|
|
ALL_point_box = [] #点对应的边框的编号
|
|
No_Label_point = []
|
|
Yes_Label_point = []
|
|
Yes_Box_index = []
|
|
###############步1 完成对ALL_point的所有点的遍历，寻找每个点对应的box在ALL_boxes中的索引，最终得到一个list ALL_point_box 代表每个点对应的box的索引，没有的是100000
|
|
for i in range(len(ALL_points)):
|
|
ALL_point_box.append(100000)
|
|
for j in range(len(ALl_boxes)):
|
|
if point_in_box(ALL_points[i],ALl_boxes[j]): #判断是否在边框里面，是的话就将边框的编号放到点对应边框的数组里面，否则该点对应的边框为100000
|
|
ALL_point_box[i]=j
|
|
break
|
|
#########################################步2.1 筛选出来准标注的和未标注的
|
|
for i in range(len(ALL_point_box)):
|
|
if ALL_point_box[i]==100000:
|
|
No_Label_point.append(ALL_points[i])
|
|
|
|
else:
|
|
Yes_Label_point.append(ALL_points[i])
|
|
Yes_Box_index.append(ALL_point_box[i])
|
|
###################步2.2对已标记的点进行遍历，去除掉具有重复边框的点
|
|
for i in range(5):
|
|
for i in range(len(Yes_Label_point)):
|
|
if i >= len(Yes_Label_point):
|
|
break
|
|
else:
|
|
for j in range(i+1,len(Yes_Label_point)):
|
|
#print(len(Yes_Label_point))
|
|
if j >= len(Yes_Label_point):
|
|
break
|
|
else:
|
|
if Yes_Box_index[i] == Yes_Box_index[j]: #当前点和遍历点是一样的
|
|
|
|
if ifsame_box(Yes_Label_point[i],Yes_Label_point[j],ALl_boxes[Yes_Box_index[i]]):
|
|
No_Label_point.append(Yes_Label_point[j])
|
|
del Yes_Label_point[j]
|
|
Yes_Box_index.remove(Yes_Box_index[j])
|
|
else:
|
|
No_Label_point.append(Yes_Label_point[i])
|
|
del Yes_Label_point[i]
|
|
Yes_Box_index.remove(Yes_Box_index[i])
|
|
break
|
|
##########步3推理未标记的点的边框信息 对No_Laber_point进行遍历
|
|
final_boxes = ALl_boxes
|
|
for i in range(len(No_Label_point)):
|
|
final_boxes.append(get_new(No_Label_point[i],ALl_boxes,image_W,image_H))
|
|
#########步4存储为xml
|
|
make_xml_iccv(xml_path.replace(".xml",".jpg"),image_H,image_W,final_boxes,save_path+xml_path)
|
|
|
|
print(final_boxes)
|
|
|
|
|
|
|
|
|
|
|