SHHB_train数据集txt格式标记头部文本转换为coco格式的txt文本
import glob
import os
import numpy as np
import shutil
from PIL import Image
START_BOUNDING_BOX_ID = 0
def val_train(SHHB_img, SHHB_txt,pro_train,pro_valid):
train_img = ‘D:\project\PyTorch-YOLOv3-master\data\custom\images\SHHB_train\’ #分离的训练图片路径
val_img = ‘D:\project\PyTorch-YOLOv3-master\data\custom\images\SHHB_valid\’ #分离的验证集图片路径
# train_img = 'D:\\project\\zzzzzzz\\ag_image\\image\\SHHB_train\\' # 分离的训练图片路径
# val_img = 'D:\\project\\zzzzzzz\\ag_image\\image\\SHHB_valid\\' #分离的验证集图片路径
if os.path.exists(train_img): #创建文件夹
shutil.rmtree(train_img)
os.makedirs(train_img)
if os.path.exists(val_img):
shutil.rmtree(val_img)
os.makedirs(val_img)
train_ratio = 0.9 #保留90%的图片为训练图片
train_list = glob.glob(SHHB_img + '/*.jpg')
train_list = np.sort(train_list)
np.random.seed(100)
np.random.shuffle(train_list)
train_num = int(len(train_list) * train_ratio)
img_list_train = train_list[:train_num]
img_list_val = train_list[train_num:]
f1 = open('SHHB_train.txt','w') #把数据的路径写入到文件夹
# f1 = open('D:\\project\\zzzzzzz\\ag_image\\SHHB_train.txt','w') #把数据的路径写入到文件夹
f = open(pro_train, 'w')
for img in img_list_train:
f1.write(img + '\n')
shutil.copy(img, train_img) #图片训练和验证数据分离
n_img = img.split('G_')
n_img = n_img[1].split('.') #n_img[0]为图片的序列号
for line in open(SHHB_txt,'r'):
n_txt = line.split('G_')
n_txt = n_txt[1].split('.') #n_txt[0]为图片的序列号
if n_txt[0] == n_img[0]:
f.write(line)
f1.close()
f.close()
f2 = open('SHHB_valid.txt','w')
f = open(pro_valid, 'w')
for img in img_list_val:
f2.write(img + '\n')
shutil.copy(img,val_img)
n_img = img.split('G_')
n_img = n_img[1].split('.') # n_img[0]为图片的序列号
for line in open(SHHB_txt, 'r'):
n_txt = line.split('G_')
n_txt = n_txt[1].split('.') # n_txt[0]为图片的序列号
if n_txt[0] == n_img[0]:
f.write(line)
f2.close()
f.close()
def convert(pro_train,pro_valid):
train_txt = ‘D:\project\PyTorch-YOLOv3-master\data\custom\labels\SHHB_train\’ #分离的训练txt文件
val_txt = ‘D:\project\PyTorch-YOLOv3-master\data\custom\labels\SHHB_valid\’ #分离的验证txt文件
# train_txt = ‘D:\project\zzzzzzz\ag_image\lable\SHHB_train\’ #分离的训练txt文件
# val_txt = ‘D:\project\zzzzzzz\ag_image\lable\SHHB_valid\’ #分离的验证txt文件
if os.path.exists(train_txt):
shutil.rmtree(train_txt)
os.makedirs(train_txt)
if os.path.exists(val_txt):
shutil.rmtree(val_txt)
os.makedirs(val_txt)
# categories = pre_define_categories.copy()
# bnd_id = START_BOUNDING_BOX_ID
# all_categories = {}
flag_continue = [1,2,3,4,5] #设置跳出循环
flag = 0
#训练集
for line in open(pro_train,'r'):
linefeed = False #消除第一行换行
line_feed = False #如果上一个人头信息舍去，则不用换行
line = line.strip('\n')
n_txt = line.split('G_')
n_txt = n_txt[1].split('.jpg')
# n_txt = n_txt[1].split(' ')
txt_id = n_txt[0] #txt_id获取txt文件序号
# txt_id = n_txt[0].split('.') #txt_id获取txt文件序号
# txt_id = txt_id[0]
txt_content = n_txt[1].split(' ')
# n_txt.remove(n_txt[0])
# txt_content = n_txt
txt_content.remove(txt_content[0])
# txt_content.strip('\n')
print(txt_id)
txt_content = list(map(int, txt_content)) #把字符转换成int类型
txt_total = txt_content[0] * 5 #文本每一行数字的总数
txt_content.remove(txt_content[0]) # 移除第一列的人头数
f = open(train_txt + '\\' + 'IMG_' + txt_id + '.txt', 'w') #写入与图片对应的txt文件
for i in range(txt_total):
if i % 5 == 0 and linefeed and line_feed:
f.write('\n')
if i % 5 == 0:
if txt_content[i + 3] <= 5 or txt_content[i + 4] <= 5:
flag = 5
else:
txt_content[i] == 0
# elif (i % 5 == 1) and (txt_content[i + 2] <=5 or txt_content[i + 3] <= 5):
# flag = 4
if flag in flag_continue:
flag = flag - 1
line_feed = False
continue
else: #把人头信息写入对应的txt文件中
linefeed = True
line_feed = True
filename = 'D:\\project\\PyTorch-YOLOv3-master\\data\\custom\\images\\SHHB_train\\' + 'IMG_' + txt_id + '.jpg'
# filename = 'D:\\project\\zzzzzzz\\ag_image\\image\\SHHB_train\\' + 'IMG_' + txt_id + '.jpg'
img = Image.open(filename) #获得图片的size
if i % 5 == 1:
f.write(str(((txt_content[i] * 2 + txt_content[i + 2]) / 2) / img.size[0]) + ' ')
elif i % 5 == 2:
f.write(str(((txt_content[i] * 2 + txt_content[i + 2]) / 2) / img.size[1]) + ' ')
# elif i % 5 == 3:
# f.write(str(txt_content[i] / img.size[0]) + ' ')
# elif i % 5 == 4:
# f.write(str(txt_content[i] / img.size[1]) + ' ')
elif i % 5 == 3:
f.write(str(txt_content[i] / ((img.size[0] + img.size[1]) / 2)) + ' ')
elif i % 5 == 4:
f.write(str(txt_content[i] / ((img.size[0] + img.size[1]) / 2)) + ' ')
# if i % 5 == 1 or i % 5 == 3:
# f.write(str(txt_content[i] / img.size[0]) + ' ')
# elif i % 5 == 2 or i % 5 == 4:
# f.write(str(txt_content[i] / img.size[1]) + ' ')
else:
f.write(str(txt_content[i]*0) + ' ')
f.close()
#验证集
for line in open(pro_valid, 'r'):
linefeed = False # 消除第一行换行
line_feed = False # 如果上一个人头信息舍去，则不用换行
line = line.strip('\n')
n_txt = line.split('G_')
n_txt = n_txt[1].split('.jpg')
txt_id = n_txt[0] # txt_id获取txt文件序号
txt_content = n_txt[1].split(' ')
txt_content.remove(txt_content[0])
# txt_content.strip('\n')
print(txt_id)
txt_content = list(map(int, txt_content))
txt_total = txt_content[0] * 5 # 文本每一行数字的总数
txt_content.remove(txt_content[0]) # 移除第一列的人头数
f = open(val_txt + '\\' + 'IMG_' + txt_id + '.txt', 'w') # 写入与图片对应的txt文件
for i in range(txt_total):
if i % 5 == 0 and linefeed and line_feed:
f.write('\n')
if i % 5 == 0:
if txt_content[i + 3] <= 5 or txt_content[i + 4] <= 5:
flag = 5
else:
txt_content[i] == 0
if flag in flag_continue:
flag = flag - 1
line_feed = False
continue
else: # 把人头信息写入对应的txt文件中
linefeed = True
line_feed = True
filename = 'D:\\project\\PyTorch-YOLOv3-master\\data\\custom\\images\\SHHB_valid\\' + 'IMG_' + txt_id + '.jpg'
img = Image.open(filename) # 获得图片的size
if i % 5 == 1:
f.write(str(((txt_content[i] * 2 + txt_content[i + 2]) / 2) / img.size[0]) + ' ')
elif i % 5 == 2:
f.write(str(((txt_content[i] * 2 + txt_content[i + 2]) / 2) / img.size[1]) + ' ')
# elif i % 5 == 3:
# f.write(str(txt_content[i] / img.size[0]) + ' ')
# elif i % 5 == 4:
# f.write(str(txt_content[i] / img.size[1]) + ' ')
elif i % 5 == 3:
f.write(str(txt_content[i] / ((img.size[0] + img.size[1]) / 2)) + ' ')
elif i % 5 == 4:
f.write(str(txt_content[i] / ((img.size[0] + img.size[1]) / 2)) + ' ')
# if i % 5 == 1 or i % 5 == 3:
# f.write(str(txt_content[i] / img.size[0]) + ' ')
# elif i % 5 == 2 or i % 5 == 4:
# f.write(str(txt_content[i] / img.size[1]) + ' ')
else:
f.write(str(txt_content[i]*0) + ' ')
f.close()
if name == ‘main’:
SHHB_img = ‘D:\project\crowd couting\part_B_final\train_data\images’ # 上海图片数据集
SHHB_txt = ‘D:\project\crowd couting\part_B_final\train_data\Part_B_train.txt’ # txt数据文件
pro_train = ‘D:\project\PyTorch-YOLOv3-master\pro_train.txt’ # 把txt中的文件分为训练和验证集写到此文件中
pro_valid = ‘D:\project\PyTorch-YOLOv3-master\pro_valid.txt’
# pro_train = ‘D:\project\zzzzzzz\ag_image\pro_train.txt’ # 把txt中的文件分为训练和验证集写到此文件中
# pro_valid = ‘D:\project\zzzzzzz\ag_image\pro_valid.txt’
# classes = ['head']
# pre_define_categories = {}
# for i, cls in enumerate(classes):
# pre_define_categories[cls] = i
# only_care_pre_define_categories = True
# only_care_pre_define_categories = False
val_train(SHHB_img, SHHB_txt, pro_train, pro_valid)
convert(pro_train, pro_valid)