在处理HSRC2016数据集中遇到了一些困难，比如将标签转换成yolo格式的，划分数据集，查询目标类别是否修改成功，将33个类别重新修改变成指定个数的类别信息等
VOC转YOLO_obb：
import os
import xml.etree.ElementTree as ET
import numpy as np
def parse_voc_xml(xml_path):
"""解析PASCAL VOC XML文件，提取图像尺寸和旋转框顶点坐标"""
tree = ET.parse(xml_path)
root = tree.getroot()
# 获取图像尺寸
size = root.find('size')
width = int(size.find('width').text)
height = int(size.find('height').text)
# 提取所有旋转框信息
objects = []
for obj in root.iter('object'):
cls_name = obj.find('name').text
polygon = obj.find('polygon')
points = []
for i in range(1, 5):
x = float(polygon.find(f'x{i}').text)
y = float(polygon.find(f'y{i}').text)
points.append([x, y])
objects.append({'class': cls_name, 'points': np.array(points)})
return width, height, objects
def normalize_points(points, img_w, img_h):
"""将顶点坐标归一化"""
points_norm = points.copy()
points_norm[:, 0] = points_norm[:, 0] / img_w # 归一化 x 坐标
points_norm[:, 1] = points_norm[:, 1] / img_h # 归一化 y 坐标
return points_norm
def convert_voc_to_yolo_obb(xml_path, txt_dir, class_dict):
"""转换单个XML文件到YOLO-OBB TXT格式（顶点坐标格式），并保存到指定文件夹"""
img_w, img_h, objects = parse_voc_xml(xml_path)
# 确保输出目录存在
if not os.path.exists(txt_dir):
os.makedirs(txt_dir)
# 生成输出TXT文件路径
xml_filename = os.path.basename(xml_path) # 获取文件名，例如 '000001.xml'
txt_filename = xml_filename.replace('.xml', '.txt') # 替换扩展名，例如 '000001.txt'
txt_path = os.path.join(txt_dir, txt_filename) # 完整输出路径
with open(txt_path, 'w') as f:
for obj in objects:
# 获取类别索引
cls_idx = class_dict.get(obj['class'], -1)
if cls_idx == -1:
continue # 跳过未定义类别
# 归一化顶点坐标
points_norm = normalize_points(obj['points'], img_w, img_h)
# 确保顶点顺序（YOLO-OBB 通常要求顺时针或逆时针）
# 这里假设 XML 中的顶点已经是正确的顺序（x1, y1, x2, y2, x3, y3, x4, y4）
points_flat = points_norm.flatten() # 展平为 [x1, y1, x2, y2, x3, y3, x4, y4]
# 写入TXT文件：class x1 y1 x2 y2 x3 y3 x4 y4
line = f"{cls_idx} {' '.join(map(str, points_flat))}\n"
f.write(line)
return txt_path
# ---------------------- 使用示例 ----------------------
if __name__ == "__main__":
# 定义类别映射字典（根据实际数据集修改）
CLASS_DICT = {'ship': 0, 'airplane': 1} # {'类别名': 类别索引}
# 指定XML文件夹路径和TXT输出文件夹路径
XML_DIR = 'F:\dataset\SSDD(+)\Annotations(SSDD+)' # 输入XML文件目录
TXT_DIR = 'F:\dataset\SSDD(+)\labels' # 输出TXT文件目录
# 遍历转换所有XML文件
for xml_file in os.listdir(XML_DIR):
if xml_file.endswith('.xml'):
xml_path = os.path.join(XML_DIR, xml_file)
txt_path = convert_voc_to_yolo_obb(xml_path, TXT_DIR, CLASS_DICT)
print(f"Converted: {xml_path} -> {txt_path}")
划分数据集:
import os
import shutil
import random
from pathlib import Path
def split_dataset(
data_dir="dataset",
output_dir="splitted_dataset",
ratios=(0.7, 0.2, 0.1), # train, val, test
img_exts=('.jpg', '.jpeg', '.png'),
anno_exts=('.txt', '.xml'),
seed=42
):
"""
数据集划分脚本
参数：
data_dir: 原始数据集路径（需包含images和labels文件夹）
output_dir: 输出目录
ratios: 划分比例 (train, val, test)
img_exts: 支持的图像文件扩展名
anno_exts: 支持的标注文件扩展名
seed: 随机种子
"""
print(f"验证比例参数 | 输入值: {ratios} | 类型: {type(ratios)} | 总和: {sum(ratios):.16f}")
# 验证参数
assert abs(sum(ratios) - 1.0) < 1e-6, \
f"比例总和必须为1.0，当前总和为{sum(ratios):.16f}"
#assert sum(ratios) == 1.0, "比例总和必须为1"
assert len(ratios) == 3, "需要三个比例值（train, val, test）"
# 创建输出目录结构
output_path = Path(output_dir)
(output_path / 'train/images').mkdir(parents=True, exist_ok=True)
(output_path / 'train/labels').mkdir(parents=True, exist_ok=True)
(output_path / 'val/images').mkdir(parents=True, exist_ok=True)
(output_path / 'val/labels').mkdir(parents=True, exist_ok=True)
(output_path / 'test/images').mkdir(parents=True, exist_ok=True)
(output_path / 'test/labels').mkdir(parents=True, exist_ok=True)
# 获取所有图像文件路径
img_dir = Path(data_dir) / 'images'
label_dir = Path(data_dir) / 'labels'
# 验证目录结构
if not img_dir.exists():
raise FileNotFoundError(f"图像目录不存在: {img_dir}")
if not label_dir.exists():
raise FileNotFoundError(f"标注目录不存在: {label_dir}")
# 收集有效文件对
file_pairs = []
for img_file in img_dir.iterdir():
if img_file.suffix.lower() in img_exts:
stem = img_file.stem
# 查找对应的标注文件
for ext in anno_exts:
anno_file = label_dir / f"{stem}{ext}"
if anno_file.exists():
file_pairs.append((img_file, anno_file))
break
else:
print(f"警告：未找到 {stem} 的标注文件，已跳过")
# 随机打乱数据
random.seed(seed)
random.shuffle(file_pairs)
# 计算划分数量
total = len(file_pairs)
train_end = int(total * ratios[0])
val_end = train_end + int(total * ratios[1])
# 划分数据集
splits = {
'train': file_pairs[:train_end],
'val': file_pairs[train_end:val_end],
'test': file_pairs[val_end:]
}
# 复制文件到对应目录
for split_name, pairs in splits.items():
print(f"正在处理 {split_name} 集 ({len(pairs)} 个样本)")
for img_src, anno_src in pairs:
# 复制图像
img_dst = output_path / split_name / 'images' / img_src.name
shutil.copy(img_src, img_dst)
# 复制标注
anno_dst = output_path / split_name / 'labels' / anno_src.name
shutil.copy(anno_src, anno_dst)
print("\n数据集划分完成！")
print(f"总样本数: {total}")
print(f"训练集: {len(splits['train'])} ({len(splits['train']) / total:.1%})")
print(f"验证集: {len(splits['val'])} ({len(splits['val']) / total:.1%})")
print(f"测试集: {len(splits['test'])} ({len(splits['test']) / total:.1%})")
if __name__ == "__main__":
# 使用示例 - 修改参数以适应你的需求
split_dataset(
data_dir=PATH_TO_YOUR_DATASET, # 原始数据集路径
output_dir=PATH_TO_SPILIT_OUTPUT, # 输出路径
ratios=(0.7, 0.2, 0.1), # 划分比例
img_exts=('.jpg', '.png'), # 支持的图像格式
anno_exts=('.txt', '.xml'), # 支持的标注格式
seed=42 # 随机种子
)
查询yolo格式下的类别标签是否存在：
import os
def find_files_with_category_in_multiple_folders(folder_paths, target_category):
for folder_path in folder_paths:
files_with_target_category = []
# 检查文件夹是否存在
if not os.path.exists(folder_path):
print(f"Error: The folder {folder_path} does not exist.")
continue
# 遍历文件夹中的所有文件
for file_name in os.listdir(folder_path):
if file_name.endswith('.txt'): # 假设标签文件是txt格式
file_path = os.path.join(folder_path, file_name)
try:
with open(file_path, 'r') as file:
for line in file:
# 提取类别编号 (行的第一个元素)
category_id = int(line.split()[0])
if category_id == target_category:
files_with_target_category.append(file_name)
break # 找到目标类别后跳出循环，继续下一个文件
except Exception as e:
print(f"Error reading file {file_name} in folder {folder_path}: {e}")
# 根据查询结果输出
file_count = len(files_with_target_category)
if file_count > 0:
print(f"\nFolder: {folder_path}")
print(f"Files containing category {target_category}: {files_with_target_category}")
print(f"Total number of files containing category {target_category} in this folder: {file_count}")
else:
print(f"\nFolder: {folder_path}")
print(f"No files containing category {target_category} were found in this folder.")
# 使用示例
folder_paths = [
r'F:\yolov8_test\ultralytics-main\dataset\HRSC_train\labels_obb_originalname33',
]
# r'F:\yolov8\ultralytics-main\dataset\HRSC_val\labels',
# r'F:\yolov8\ultralytics-main\dataset\HRSC_test\labels'
# ] # 替换为你要查询的多个文件夹路径
target_category = 6 # 替换为要查询的类别编号
find_files_with_category_in_multiple_folders(folder_paths, target_category)
将标签中的某类别修改成目标类别：
import os
def replace_category_in_files(folder_path, target_categories, new_category):
# 检查文件夹是否存在
if not os.path.exists(folder_path):
print(f"Error: The folder {folder_path} does not exist.")
return
# 遍历文件夹中的所有文件
for file_name in os.listdir(folder_path):
if file_name.endswith('.txt'): # 假设标签文件是txt格式
file_path = os.path.join(folder_path, file_name)
# 读取文件内容
with open(file_path, 'r') as file:
lines = file.readlines()
# 创建一个新的文件内容
new_lines = []
for line in lines:
elements = line.split()
category_id = int(elements[0])
# 如果类别编号在目标类别列表中，替换为新的类别编号
if category_id in target_categories:
elements[0] = str(new_category)
# 将修改后的行加入新的内容
new_lines.append(" ".join(elements) + "\n")
# 写入修改后的内容回到文件
with open(file_path, 'w') as file:
file.writelines(new_lines)
print(f"All occurrences of categories {target_categories} have been replaced with {new_category}.")
# 使用示例
#folder_path = r'F:\yolov8\ultralytics-main\dataset\HRSC_train\labels'
#folder_path = r'F:\yolov8\ultralytics-main\dataset\HRSC_val\labels'
folder_path = r'F:\yolov8_test\ultralytics-main\dataset\HRSC_val\labels_obb_originalname33'# 替换为标签文件夹路径
target_categories = [17,19,21,22,23,24,25,28,29] # 替换为要修改的类别编号列表
new_category = 3 # 替换为新的统一类别编号
replace_category_in_files(folder_path, target_categories, new_category)
#1:4,5,11,12,30,31yiwancheng
#2:6,7,8,9,10,13,14,16,18,20,27yiwancheng
#3:17,19,21,22,23,24,25,28,29yiwancheng
#0:26,25,23wancheng
扫一扫
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
