将 RSOD 遥感图像数据集转为 PASCAL VOC 格式

PASCAL VOC 格式的 RSOD 数据集：下载地址（所需积分/C币: 0）（原始数据集 playground.zip 中的图片有189张，而标注只有149张，PASCAL VOC 格式的数据集中，只选用了有标注的149张图片）

原

1、下载原数据集 https://github.com/RSIA-LIESMARS-WHU/RSOD-Dataset-
2、将四个类别中的Annotation\xml\和JPEGImages分别放在Annotations和JPEGImages里（类似 VOC 的文件顺序）
3、重命名文件

import os
print(os.getcwd())

ann_filepath = 'F:/VOC2007/VOC2007/Annotations/'
img_filepath = 'F:/VOC2007/VOC2007/JPEGImages/'

ann_num = 0
img_num = 0

for file in os.listdir(ann_filepath):
    new_name = '%06d' % int(ann_num)+'.xml'
    ann_num += 1
    print(file, new_name)
    os.rename(os.path.join(ann_filepath, file), 
              os.path.join(ann_filepath, new_name))
    
for file in os.listdir(img_filepath):
    new_name = '%06d' % int(img_num)+'.jpg'
    img_num += 1
    print(file, new_name)
    os.rename(os.path.join(img_filepath, file), 
              os.path.join(img_filepath, new_name))

4、最后，自己制作 trainval.txt、test.txt 等文件，里面放的是图片名（可能会报“没有这个文件”的错误，需要手动新建……能力有限）。

from sklearn.model_selection import train_test_split
import os

name_path = r'F:/VOC2007/VOC2007/JPEGImages'
name_list = os.listdir(name_path)
names = []

for i in name_list:
    names.append(i.split('.')[0])

trainval, test = train_test_split(names, test_size=0.5, shuffle=10)
val, train = train_test_split(trainval, test_size=0.5, shuffle=10)

with open('F:/VOC2007/VOC2007/ImageSets/Main/trainval.txt', 'w') as fw:
    for i in trainval:
        fw.write(i + '\n')

with open('F:/VOC2007/VOC2007/ImageSets/Main/test.txt', 'w') as fw:
    for i in test:
        fw.write(i + '\n')

with open('F:/VOC2007/VOC2007/ImageSets/Main/val.txt', 'w') as fw:
    for i in val:
        fw.write(i + '\n')

with open('F:/VOC2007/VOC2007/ImageSets/Main/train.txt', 'w') as fw:
    for i in train:
        fw.write(i + '\n')

print('done!')

2021.07.10 改

多谢评论大佬提醒，按上述3、重命名文件时，私以为 Annotations 和 JPEGImages中的文件名都是同一排序方式，其实并不是……

将重命名后的 Annotations 标注在 JPEGImages上时，发现只有aircraft能够一一对应，oiltank 、playground 和 overpass 则对应不上，如下：

aircraft	oiltank	overpass	playground

修改后的3、重命名文件如下：

import os

print(os.getcwd())

ann_filepath = 'F:/VOC2007/VOC2007/Annotations/'
img_filepath = 'F:/VOC2007/VOC2007/JPEGImages/'

ann_num = 0
img_num = 0

for old_file in os.listdir(ann_filepath):
    new_file_name = '%06d' % int(ann_num) + '.xml'
    new_img_name = '%06d' % int(ann_num) + '.jpg'
    ann_num += 1

    os.rename(os.path.join(ann_filepath, old_file),
              os.path.join(ann_filepath, new_file_name))

    old_img = old_file.split('.')[0] + '.jpg'
    os.rename(os.path.join(img_filepath, old_img),
              os.path.join(img_filepath, new_img_name))

每个类别如下：

aircraft	oiltank	overpass	playground

附：对比代码

# 对比 RSOD 数据集变为 PASCAL VOC 前后是否一一对应

import numpy as np
import cv2
import os
import xml.etree.ElementTree as ET

color_gt = (255, 0, 0)
color_md = (0, 255, 0)
thickness = 1

img_path = 'F:/VOC2007/VOC2007/JPEGImages/'
file_path = 'F:/VOC2007/VOC2007/Annotations/'

for f in os.listdir(file_path):
    file = os.path.join(file_path, f)
    tree = ET.parse(file)
    root = tree.getroot()

    size = tree.find('size')
    width = float(size.find('width').text)
    height = float(size.find('height').text)
    objs = root.findall('object')

    file_name = f.split('.')
    img_source = os.path.join(img_path, file_name[0] + '.jpg')
    img = cv2.imread(img_source, 0)

    for ix, obj in enumerate(objs):
        bbox = obj.find('bndbox')
        obj_name = str(obj.find('name').text)
        y_max = float(bbox.find('ymax').text) - 1
        y_min = float(bbox.find('ymin').text) - 1
        x_max = float(bbox.find('xmax').text) - 1
        x_min = float(bbox.find('xmin').text) - 1

        ptLeftTop = (int(x_min), int(y_min))
        ptRightBottom = (int(x_max), int(y_max))

        cv2.rectangle(img, ptLeftTop, ptRightBottom, color_md)
        cv2.putText(img, obj_name, (int(x_min), int(y_min)), cv2.FONT_HERSHEY_COMPLEX, thickness, color_md)

    image = np.expand_dims(img, axis=2)
    image = np.concatenate((image, image, image), axis=-1)

    if obj_name == 'aircraft':
            # ['aircraft', 'oiltank', 'overpass', 'playground']
        cv2.imshow('image', img)
        k = cv2.waitKey(0)
        if k == 27:  # wait for ESC key to exit
            cv2.destroyAllWindows()
        elif k == ord('s'):  # wait for 's' key to save and exit
            cv2.imwrite(file_name[0] + '.png', img)
            cv2.destroyAllWindows()

# 对比依次取 VOC 数据集与原数据集标注文件中的内容，查看是否一致

img_path = 'F:/VOC2007/VOC2007/JPEGImages/'
file_path = 'F:/VOC2007/VOC2007/Annotations/'

origin_path = "F:/RSOD"
origin_name_list = ['aircraft', 'oiltank', 'overpass', 'playground']

voc_list = []
origin_list = []

for f in os.listdir(file_path):
    file = os.path.join(file_path, f)
    tree = ET.parse(file)
    root = tree.getroot()

    size = tree.find('size')
    width = float(size.find('width').text)
    height = float(size.find('height').text)
    objs = root.findall('object')

    for ix, obj in enumerate(objs):
        obj_name = str(obj.find('name').text)
        bbox = obj.find('bndbox')
        y_max = float(bbox.find('ymax').text)
        y_min = float(bbox.find('ymin').text)
        x_max = float(bbox.find('xmax').text)
        x_min = float(bbox.find('xmin').text)

        voc_list.append([y_max, y_min, x_max, x_min, obj_name])

for name in origin_name_list:
    origin_class_path = origin_path + '\\' + name + '\Annotation\\xml'
    for f in os.listdir(origin_class_path):
        file = os.path.join(origin_class_path, f)
        tree = ET.parse(file)
        root = tree.getroot()

        size = tree.find('size')
        width = float(size.find('width').text)
        height = float(size.find('height').text)
        objs = root.findall('object')

        for ix, obj in enumerate(objs):
            obj_name = str(obj.find('name').text)
            bbox = obj.find('bndbox')
            y_max = float(bbox.find('ymax').text)
            y_min = float(bbox.find('ymin').text)
            x_max = float(bbox.find('xmax').text)
            x_min = float(bbox.find('xmin').text)

            origin_list.append([y_max, y_min, x_max, x_min, obj_name])

print(voc_list)
print(origin_list)
print(voc_list == origin_list)