目标检测标签文件json格式转化为txt格式

原创

于 2022-09-26 10:13:09 发布 · 1.4k 阅读

7 ·

CC 4.0 BY-SA版权

文章标签：

#目标检测 #json

本文介绍如何将BDD100K数据集中的JSON格式标注文件转换为YOLOv3等目标检测算法所需的TXT格式。通过解析JSON文件中的图像名称、类别及边界框坐标，并将其标准化到0-1范围内，最终写入TXT文件中。

import json
import os
import cv2


if __name__ == '__main__':
    json_file = "G:/other/bdd100k_det_20_labels_trainval/bdd100k/labels/det_20/det_val.json"  # json file path
    save_path = "G:/other/bdd100k_images_100k/bdd100k/labels/val"  # txt save path
    with open(json_file, 'r') as f:
        diction = json.load(f)
        for i in range(len(diction)):
            names = diction[i].get("name")  # image name
            img = cv2.imread('G:/other/bdd100k_images_100k/bdd100k/images/100k/val/' + names)
            height, width = img.shape[:2]
            labels = diction[i].get("labels")
            label_txt = os.path.join(save_path, names[:-4] + ".txt")  # txt name
            with open(label_txt, 'a') as file:
                for j in range(len(labels)):
                    boxes = labels[j].get("box2d")  # box
                    box_class = labels[j].get("category")  # class
                    if box_class == 'car':