BDD100K数据集格式转YOLO格式（目标检测）

    conda create -n bdd100k python=3.8
    

    pip install -r requirements.txt
    

    python -m bdd100k.label.to_coco -m det -i "C:\......\det_train.json" -o "C:\......\det_train_coco.json" --nproc 4
    

 # 在使用本格式转换工具前,先使用 bdd100k 自带工具包将 bdd100k 格式转换为 coco 格式,例如：

    
 # python -m bdd100k.label.to_coco -m det -i "C:\......\det_train.json" -o "C:\......\det_train_coco.json" --nproc 4
    
 # 然后修改下方路径,运行此程序,将 coco 格式转换为 yolo 格式
    
 import json
    
 import os
    
  
    
 # 输入 COCO JSON 文件路径
    
 coco_json_path = "C:/Users/BJDYJ/Desktop/BDD100K_dataset/labels/det_val_coco.json"
    
 # 输出标签文件夹路径
    
 output_dir = "C:/Users/BJDYJ/Desktop/BDD100K_dataset/labels/val_labels"
    
  
    
 # 创建输出文件夹
    
 os.makedirs(output_dir, exist_ok=True)
    
  
    
 # 读取 COCO JSON 文件
    
 with open(coco_json_path, 'r', encoding='utf-8') as f:
    
     coco_data = json.load(f)
    
  
    
 # 获取图像信息和标注信息
    
 images = {image['id']: image for image in coco_data['images']}
    
 annotations = coco_data['annotations']
    
 categories = {category['id']: category['name'] for category in coco_data['categories']}
    
  
    
 # 遍历所有标注信息
    
 for annotation in annotations:
    
     image_id = annotation['image_id']
    
     category_id = annotation['category_id']
    
     bbox = annotation['bbox']  # COCO 格式中的 bbox: [x_min, y_min, width, height]
    
  
    
     # 获取图像的宽高
    
     image_info = images[image_id]
    
     img_width = image_info['width']
    
     img_height = image_info['height']
    
  
    
     # 计算 YOLO 格式的 bbox: [center_x, center_y, width, height]
    
     x_min, y_min, width, height = bbox
    
     center_x = (x_min + width / 2) / img_width
    
     center_y = (y_min + height / 2) / img_height
    
     norm_width = width / img_width
    
     norm_height = height / img_height
    
  
    
     # 生成 YOLO 格式的标签行
    
     label_line = f"{category_id - 1} {center_x:.6f} {center_y:.6f} {norm_width:.6f} {norm_height:.6f}"
    
  
    
     # 构建输出文件路径
    
     txt_filename = f"{os.path.splitext(image_info['file_name'])[0]}.txt"
    
     txt_filepath = os.path.join(output_dir, txt_filename)
    
  
    
     # 写入标签文件
    
     with open(txt_filepath, 'a') as txt_file:
    
     txt_file.write(label_line + '\n')
    
  
    
 print("标签文件已成功转换并保存到:", output_dir)
    
    
    
    

 import os

    
 import json
    
  
    
 # COCO格式的标签文件路径
    
 coco_file_path = r'C:\Users\BJDYJ\Desktop\BDD100K_dataset\labels\det_train_coco.json'
    
  
    
 # YOLO格式标签保存路径
    
 output_dir = r'C:\Users\BJDYJ\Desktop\BDD100K_dataset\labels\labels'
    
  
    
 # 如果输出目录不存在，则创建它
    
 if not os.path.exists(output_dir):
    
     os.makedirs(output_dir)
    
  
    
 # 加载COCO格式的JSON文件
    
 with open(coco_file_path, 'r') as f:
    
     coco_data = json.load(f)
    
  
    
 # 提取图像信息和注释信息
    
 images = coco_data['images']
    
 annotations = coco_data['annotations']
    
 categories = coco_data['categories']
    
  
    
 # 创建从类别ID映射到类别索引ID的字典
    
 category_id_to_name = {cat['id']: cat['name'] for cat in categories}
    
 category_id_to_index = {cat['id']: idx for idx, cat in enumerate(categories)}
    
  
    
 # 创建一个字典，将图像ID映射到文件名
    
 image_id_to_filename = {img['id']: img['file_name'] for img in images}
    
  
    
 # 处理每一个注释，转换为YOLO格式并保存到对应的文件中
    
 for annotation in annotations:
    
     image_id = annotation['image_id']
    
     category_id = annotation['category_id']
    
  
    
     # 获取图像的文件名
    
     image_filename = image_id_to_filename[image_id]
    
  
    
     # 获取图片的宽度和高度
    
     image_info = next(img for img in images if img['id'] == image_id)
    
     img_width = image_info['width']
    
     img_height = image_info['height']
    
  
    
     # 获取边界框信息（COCO格式为[x_min, y_min, width, height]）
    
     bbox = annotation['bbox']
    
     x_min, y_min, box_width, box_height = bbox
    
  
    
     # 将COCO格式的bbox转换为YOLO格式
    
     x_center = (x_min + box_width / 2) / img_width
    
     y_center = (y_min + box_height / 2) / img_height
    
     norm_width = box_width / img_width
    
     norm_height = box_height / img_height
    
  
    
     # 获取类别索引ID
    
     class_id = category_id_to_index[category_id]
    
  
    
     # YOLO格式：类ID x_center y_center width height
    
     yolo_annotation = f"{class_id} {x_center} {y_center} {norm_width} {norm_height}\n"
    
  
    
     # 创建输出文件的路径（.txt文件名应与图像文件名对应）
    
     label_filename = os.path.splitext(image_filename)[0] + '.txt'
    
     label_filepath = os.path.join(output_dir, label_filename)
    
  
    
     # 将注释追加写入到对应的txt文件
    
     with open(label_filepath, 'a') as label_file:
    
     label_file.write(yolo_annotation)
    
  
    
 print("COCO格式的标签已经成功转换为YOLO格式！")