前言

旋转框目标检测（Rotated bounding box object detection）是计算机视觉领域的一项技术，它用于检测图像中具有任意方向的目标。与传统的水平矩形框目标检测相比，旋转框目标检测能够更准确地描述物体的形状和位置，尤其是对于那些长宽比差异较大或者方向各异的物体，如遥感图像中的建筑物、文本行、车辆等,本文将详细介绍YOLOV11-OBB自定义数据集训练测试流程，帮您实现旋转框目标检测。

一、数据集制作

本次标注软件采用的是X-AnyLabeling，github地址：windows直接下载。

点击X-AnyLabeling-CPU.exe下载，下载好之后打开，界面如下所示：

使用方法如下所示：

标注好数据集之后标签是json格式，运行下面的代码将数据集转化为yolov11所需要的txt格式：

def order_points(points):
    # 1. 计算中心点
    center_x = sum([p[0] for p in points]) / 4
    center_y = sum([p[1] for p in points]) / 4

    # 2. 计算每个点相对于中心点的角度，并排序
    def angle_from_center(point):
        return math.atan2(point[1] - center_y, point[0] - center_x)

    # 按角度逆时针排序
    points = sorted(points, key=angle_from_center, reverse=True)

    # 3. 按"右上、右下、左下、左上"的顺序排列
    ordered_points = [points[0], points[1], points[2], points[3]]

    return ordered_points


import math, os, json

# 定义类别映射字典，键为类别名称，值为类别索引
category_mapping = {"box_top": 0}


def order_points(points):
    # 计算四个顶点的中心点坐标
    center_x = sum([p[0] for p in points]) / 4
    center_y = sum([p[1] for p in points]) / 4
    # 计算每个点相对于中心点的角度，按逆时针方向排序，确保点的顺序一致
    points = sorted(points, key=lambda p: math.atan2(p[1] - center_y, p[0] - center_x), reverse=True)
    # 返回按顺序排列的四个点，顺序为“右上、右下、左下、左上”
    return [points[0], points[1], points[2], points[3]]


def convert_json_to_yolo11(json_folder, output_folder, category_mapping):
    os.makedirs(output_folder, exist_ok=True)  # 确保输出文件夹存在，如果不存在则创建
    for filename in os.listdir(json_folder):  # 遍历JSON文件夹中的每个文件
        if filename.endswith('.json'):  # 只处理.json结尾的文件
            json_path = os.path.join(json_folder, filename)
            try:
                with open(json_path, 'r') as f:
                    data = json.load(f)  # 读取JSON文件内容
            except (FileNotFoundError, json.JSONDecodeError) as e:
                print(f"文件读取错误或格式无效：{filename}，错误信息：{e}")  # 如果读取出错，输出错误信息并跳过
                continue
            # 获取图像的宽度和高度
            image_width = data.get("imageWidth")
            image_height = data.get("imageHeight")
            if not image_width or not image_height:  # 如果图像尺寸信息缺失，输出警告并跳过该文件
                print(f"图像尺寸信息缺失：{filename}")
                continue
            yolo_lines = []  # 初始化YOLO格式的标注行
            for shape in data.get("shapes", []):  # 遍历JSON中的每个标注对象
                label = shape.get("label")  # 获取标注的类别名称
                if label in category_mapping:  # 检查类别名称是否在类别映射字典中
                    class_index = category_mapping[label]  # 获取对应的类别索引
                else:
                    print(f"未识别的类别标签：{label}，跳过该标注")  # 如果类别标签未定义，输出警告并跳过该标注
                    continue
                points = shape.get("points")  # 获取标注的四个顶点坐标
                if len(points) == 4:  # 确保标注包含四个顶点，符合OBB要求
                    ordered_points = order_points(points)  # 使用order_points函数对顶点进行顺序排列
                    # 将顶点坐标归一化到0-1之间，并保留六位有效数字
                    normalized_points = [[round(x / image_width, 6), round(y / image_height, 6)] for x, y in
                                         ordered_points]
                    # 构造YOLO格式的标注行，包含类别索引和四个归一化顶点坐标
                    yolo_line = [class_index] + [coord for point in normalized_points for coord in point]
                    yolo_lines.append(" ".join(map(str, yolo_line)))  # 将标注行添加到YOLO行列表中
            if yolo_lines:  # 如果存在标注数据，则写入到对应的TXT文件中
                txt_filename = os.path.splitext(filename)[0] + ".txt"  # 生成输出TXT文件名
                output_path = os.path.join(output_folder, txt_filename)
                with open(output_path, 'w') as out_file:
                    out_file.write("\n".join(yolo_lines))  # 将所有标注行写入TXT文件
                print(f"转换完成: {output_path}")  # 输出转换完成信息

# 使用示例
json_folder = "/home/build/yhgt/json"  # JSON文件夹路径，需要修改
output_folder = "/home/build/yhgt/txt"  # 输出TXT文件夹路径，需要修改
convert_json_to_yolo11(json_folder, output_folder, category_mapping)

二、模型训练

2.1 划分训练集验证集:

2.2 配置yaml文件:

# YOLOv5 🚀 by Ultralytics, GPL-3.0 license
# PASCAL VOC dataset http://host.robots.ox.ac.uk/pascal/VOC by University of Oxford
# Example usage: python train.py --data VOC.yaml
# parent
# ├── yolov5
# └── datasets
#     └── VOC  ← downloads here


# Train/val/test sets as 1) dir: path/to/imgs, 2) file: path/to/imgs.txt, or 3) list: [path/to/imgs1, path/to/imgs2, ..]
#path: ../VOCdevkit_wpeson_Tanker_01-20/VOC2007/ImageSets/Main
train: /home/build/yhgt/images/train/ # train images (relative to 'path')  16551 images

val: /home/build/yhgt/images/val/
 
 # train images (relative to 'path')  16551 images
#val: # val images (relative to 'path')  4952 images
#  - val.txt
#test: # test images (optional)
#  - test.txt

# Classes
nc: 1 # number of classes
names: ['box_top']  # class names


# Download script/URL (optional) ---------------------------------------------------------------------------------------

2.3 训练

from ultralytics import YOLO

# Load a model
model = YOLO("/home/build/下载/ultralytics-main (1)/yolo11n-obb.pt")  # load a pretrained model (recommended for training)

# Train the model with 2 GPUs
results = model.train(data="/home/build/yhgt/cc.yaml", epochs=200, imgsz=640, device="0，1")