import os import json import cv2 import random import time from PIL import Image # 部分同学都用的autodl, 用antodl举例 # 使用绝对路径 #数据集 txt格式-labels标签文件夹 txt_labels_path='/mnt/e/datasets/cocomini/val/labels' #数据集图片images文件夹 datasets_img_path='/mnt/e/datasets/cocomini/val/images' # 这里 voc 为数据集文件名字，可以改成自己的路径 # xx.json生成之后存放的地址 save_path='/mnt/e/basic model/contrast experiment/yolov12-main/yolov12-main/TXTToCOCO' #自己数据集的类别, 一行一个类 例如: class.txt文件 # car # phone # ... classes_txt='/mnt/e/basic model/contrast experiment/yolov12-main/yolov12-main/TXTToCOCO/class.txt' with open(classes_txt,'r') as fr: lines1=fr.readlines() categories=[] for j,label in enumerate(lines1): label=label.strip() categories.append({'id':j,'name':label,'supercategory':'None'}) print(categories) write_json_context=dict() write_json_context['info']= {'description': 'For object detection', 'url': '', 'version': '', 'year': 2021, 'contributor': '', 'date_created': '2021'} write_json_context['licenses']=[{'i解释

### 将YOLO TXT格式转换为COCO JSON格式 以下是将YOLO格式的TXT标注文件转换为COCO格式JSON文件的完整代码实现。此代码会处理类别、图片以及标注信息。 #### 转换逻辑说明 1. YOLO格式的标注文件通常存储在`.txt`文件中，每行表示一个目标框，格式为： `<class_id> <x_center> <y_center> <width> <height>` (归一化坐标)。 2. COCO格式是一个复杂的JSON结构，包含多个字段，如`images`, `annotations`, 和 `categories`。需要将其映射到对应的字段。 3. 需要提供以下输入： - 图像路径（用于获取图像尺寸） - 标注路径（YOLO `.txt` 文件所在的目录） - 类别列表（保存在`classes.txt`） --- ```python import os import json from PIL import Image def yolo_to_coco(image_dir, label_dir, classes_file, output_json): """ Convert YOLO format annotations to COCO format. :param image_dir: Directory containing images. :param label_dir: Directory containing YOLO .txt labels. :param classes_file: File with class names. :param output_json: Output path for the COCO-format JSON file. """ # Load categories from classes.txt with open(classes_file, 'r') as f: categories = [{'id': idx + 1, 'name': line.strip()} for idx, line in enumerate(f.readlines())] images = [] annotations = [] annotation_id = 1 # Annotation ID starts at 1 # Iterate over all files in the label directory for txt_file_name in sorted(os.listdir(label_dir)): if not txt_file_name.endswith('.txt'): continue img_file_name = txt_file_name.replace('.txt', '.jpg') img_path = os.path.join(image_dir, img_file_name) try: with Image.open(img_path) as img: width, height = img.size except FileNotFoundError: print(f"Image {img_file_name} not found.") continue # Add image entry image_info = { "id": len(images) + 1, "file_name": img_file_name, "width": width, "height": height } images.append(image_info) # Parse and convert bounding boxes with open(os.path.join(label_dir, txt_file_name), 'r') as f: lines = f.readlines() for line in lines: parts = line.strip().split(' ') category_id = int(parts[0]) + 1 # Category IDs start at 1 # Convert normalized coordinates back to absolute values x_center_norm, y_center_norm, w_norm, h_norm = map(float, parts[1:]) x_center_abs = x_center_norm * width y_center_abs = y_center_norm * height box_w = w_norm * width box_h = h_norm * height # Calculate top-left corner of bbox x_min = max(0, round(x_center_abs - box_w / 2)) y_min = max(0, round(y_center_abs - box_h / 2)) area = box_w * box_h # Create annotation object annotation = { "id": annotation_id, "image_id": image_info["id"], "category_id": category_id, "bbox": [x_min, y_min, box_w, box_h], "area": area, "iscrowd": 0 } annotations.append(annotation) annotation_id += 1 # Build final COCO structure coco_data = { "info": {"description": "Converted dataset"}, "licenses": [], "categories": categories, "images": images, "annotations": annotations } # Save to JSON file with open(output_json, 'w') as f: json.dump(coco_data, f) # Example usage if __name__ == "__main__": IMAGE_DIR = "./images" LABEL_DIR = "./labels" CLASSES_FILE = "./classes.txt" OUTPUT_JSON = "./output/coco_annotations.json" yolo_to_coco(IMAGE_DIR, LABEL_DIR, CLASSES_FILE, OUTPUT_JSON) ``` --- ### 参数解释 - **`image_dir`**: 存储图像的目录路径。 - **`label_dir`**: 存储YOLO格式标注文件的目录路径。 - **`classes_file`**: 包含类别名称的文本文件路径，每一类占一行。 - **`output_json`**: 输出的COCO格式JSON文件路径。 --- ### 注意事项 1. 如果某些图像缺少对应标签文件，则这些图像不会被加入最终的JSON文件中[^1]。 2. 假设所有图像均为JPEG格式；如果存在其他格式，请调整扩展名匹配部分。 3. 使用PIL库读取图像大小时，需确保安装依赖项：`pip install pillow`. ---