已知COCO格式丢失了datatrucks源文件将新datatrucks加入COCO

该博客内容涉及将不同数据格式转换为COCO标准JSON格式的过程。首先，它演示了如何从一个JSON文件读取数据并重新组织成COCO格式，包括图像信息、类别定义和注解。接着，它展示了如何处理另一个数据集，将边界框和区域转换为COCO所需的格式。最后，将所有这些信息整合到一个统一的COCO JSON文件中，用于目标检测或分割任务。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

将COCO转json:

import json

with open('/home/lixuan/fromLX/train.json') as f:
    jsdict = json.load(f)
    images = jsdict['images']
    categories = jsdict['categories']
    annotations = jsdict['annotations']
    print(annotations[0])
    exit()
    imagedict = {}
    for image in images:
        id = image['id']
        del image['id']
        imagedict[id] = image
        imagedict[id]['annotation'] = []
    for annotation in annotations:
        image_id = annotation['image_id']
        del annotation['id']
        del annotation['image_id']
        imagedict[image_id]['annotation'].append(annotation)

将datatrucks转json：

import json
import cv2
import numpy as np

def getbox(cnt):
    cnt = np.array(cnt)
    rect = cv2.minAreaRect(cnt)  # 得到最小外接矩形的（中心(x,y), (宽,高), 旋转角度）
    box = cv2.boxPoints(rect)  # 获取最小外接矩形的4个顶点坐标(ps: cv2.boxPoints(rect) for OpenCV 3.x)
    return [max(int(np.min(box[:, 0])),0), max(int(np.min(box[:, 1])),0), int(np.max(box[:, 0])), int(np.max(box[:, 1]))]


def getarea(points):
    return cv2.contourArea(np.array(points))

with open('IcecreamSeg_第一组.json') as f:
    for data in f.readlines():
        data = data.strip()
        dic = json.loads(data)
        content = dic['content']
        annotation = dic['annotation']
        file_name = content.split('/')[-1]
        height = annotation[0]['imageHeight']
        width = annotation[0]['imageWidth']
        image = {}
        image['height'] = height
        image['width'] = width
        image['file_name'] = file_name
        image['annotation'] = []
        for anno in annotation:
            annodict = {}
            annodict['iscrowd'] = 0
            points = anno['points']
            counts = []
            # if [None, None] not in points:
            for point in points:
                counts.append(point[0] * width)
                counts.append(point[1] * height)
                point[0] = int(point[0] * width)
                point[1] = int(point[1] * height)
            area = getarea(points)
            bbox = getbox(points)
            annodict['bbox'] = bbox
            annodict['segmentation'] = [counts]
            if anno['label'][0] == 'interference':
                annodict['category_id'] = 1
            elif anno['label'][0] == 'icecream_clear':
                annodict['category_id'] = 2
            elif anno['label'][0] == 'icecream_notso_clear':
                annodict['category_id'] = 3
            else:
                annodict['category_id'] = 4
            annodict['area'] = area
            image['annotation'].append(annodict)
        with open(file_name.split('.')[0] + '.json','w') as fw:
            json.dump(image,fw)

将json合成为COCO：

import json
import os

COCO_dict = {}
COCO_dict['images'] = []#{'height': 1920, 'width': 2560, 'id': 1, 'file_name': 'val_95.jpg'}
COCO_dict['categories'] = [{'supercategory': 'interference', 'id': 1, 'name': 'interference'}, {'supercategory': 'icecream_clear', 'id': 2, 'name': 'icecream_clear'}, {'supercategory': 'icecream_not_so_clear', 'id': 3, 'name': 'icecream_not_so_clear'}, {'supercategory': 'adv', 'id': 4, 'name': 'adv'}]
COCO_dict['annotations'] = []

"""
{'iscrowd': 0, 'image_id': 1, 'bbox': [1936.0, 1497.0, 218.0, 268.0], 'segmentation': [[]], 'category_id': 1, 'id': 1, 'area': 4915200, 'extreme_points': [2129.0, 1497.0, 1936.0, 1540.0, 2061.0, 1765.0, 2154.0, 1527.0, 2045.0, 1631.0]}
"""

image_id = 1
id = 1
for file in os.listdir('/home/lixuan/fromLX/icecreamtxt'):
    with open('/home/lixuan/fromLX/icecreamtxt/{}'.format(file)) as f:
        dic = json.load(f)
        image = {}
        image['height'] = dic['height']
        image['width'] = dic['width']
        image['id'] = image_id
        image['file_name'] = dic['file_name']
        COCO_dict['images'].append(image)
        for anno in dic['annotation']:
            anno['image_id'] = image_id
            anno['id'] = id
            COCO_dict['annotations'].append(anno)
            id += 1
        image_id += 1
with open('train.json','w') as f:
    json.dump(COCO_dict,f)