1. Descarga del conjunto de datos
Dirección del sitio web oficial: https://thudair.baai.ac.cn/index
2. Introducción al formato de archivo de anotación del conjunto de datos Rope3D
标注数据:
car -1 -1 -10 166.00 111.00 181.00 127.00 -1 -1 -1 -1000 -1000 -1000 -10
参数定义:
Type X X X left_top_x left_top_y right_bottom_x right_bottom_y X X X X X X X
Solo es necesario utilizar los archivos de anotación de YOLO,izquierda_arriba_x, izquierda_arriba_y, derecha_abajo_x, derecha_abajo_y, otros parámetros se pueden ignorar directamente.
3. Código de procesamiento
from PIL import Image
import os
# 文件格式为txt
# 1 -1 -1 -10 166.00 111.00 181.00 127.00 -1 -1 -1 -1000 -1000 -1000 -10
# Label:
# Type X X X left_top_x left_top_y right_bottom_x right_bottom_y X X X X X X X
# 需要读取每行数据,并坐标信息进行转化
def convert(size, box):
x_center = (box[0] + box[1]) / 2.0
y_center = (box[2] + box[3]) / 2.0
x = x_center / size[0]
y = y_center / size[1]
w = (box[1] - box[0]) / size[0]
h = (box[3] - box[2]) / size[1]
# print(x, y, w, h)
return (x, y, w, h)
# 获取文件夹下所有图片的高和宽信息,并通过字典返回
def getImageWHInfo():
# 保存数据的字典
image_info_dict = {
}
# 获取文件夹中所有文件的名称
filenames = os.listdir("E:\project\Rope3D dataset\original/train_images")
for filename in filenames:
# 打开图片文件
image = Image.open("E:\project\Rope3D dataset\original/train_images/" + filename)
# 获取图片的宽和高
width, height = image.size
# 保存图片信息
image_info_dict[filename] = [width, height]
# 关闭图片
image.close()
return image_info_dict
if __name__ == '__main__':
# 获取文件夹下所有的图片的width,high信息
image_info_dict = getImageWHInfo()
save_txt_files_path = 'E:\project\Rope3D dataset\original/train_labels_yolo'
success_num = 0
total = len(image_info_dict)
# Rope3D数据集实际分类不只有8类,我进行了筛选,通过下面的分类过滤代码。如果不需要,可以修改为官方类别,并注释掉下面的过滤代码。
category_label = {
'car': 0, 'truck': 1, 'van': 2, 'bus': 3, 'pedestrian': 4, 'cyclist': 5, 'tricyclist': 6,
'motorcyclist': 7}
for fileName, WHList in image_info_dict.items():
print(fileName, WHList)
image_width = WHList[0]
image_height = WHList[1]
fileName = fileName.split('.')[0] + '.txt'
out_txt_path = os.path.join(save_txt_files_path, fileName)
out_txt_f = open(out_txt_path, 'w')
# 打开文件
with open('E:\project\Rope3D dataset\original/training\label_2/' + fileName, 'r') as f:
# 逐行读取文件内容
for line in f:
# 去除行尾的空白字符(包括空格和换行符)
line = line.strip()
# 如果行不为空,则输出该行
if line:
line_list = line.split()
category = line_list[0]
# 分类过滤代码,可以过滤掉不想要分类。
if category in ['barrow', 'unknowns_movable', 'trafficcone', 'unknown_unmovable']:
continue
categoryLabel = category_label[category]
xmin = left_top_x = line_list[4]
ymin = left_top_y = line_list[5]
xmax = right_bottom_x = line_list[6]
ymax = right_bottom_y = line_list[7]
b = (float(xmin), float(xmax), float(ymin), float(ymax))
bb = convert((image_width, image_height), b)
out_txt_f.write(str(categoryLabel) + " " + " ".join([str(a) for a in bb]) + '\n')
success_num = success_num + 1
print("写入成功:" + str(success_num) + '/' + str(total))