coco数据集制作-多文件夹

背景：标准的coco格式数据集需要把所有图片放到一个文件夹里，而很多情况下，会有很多个文件夹，我们并不想把所有图片放到一起。

本文把多个文件夹下的yolov8(5)的txt标签转换至coco标签，转换标签代码如下：

import os
import json
import cv2

# 图片和标签路径
img_root = './soccernet/tracking/images/test/'
label_root = './soccernet/tracking/labels/test/'
# 模板
coco = {
    "info": {
        "year": 2023,
        "version": "1.0",
        "description": "Example COCO Dataset",
        "contributor": "jefft",
        "date_created": "2023-08-22"
    },
    "licenses": [
        {
            "id": 1,
            "name": "License Name",
            "url": "https://license-url.com"
        }
    ],
    "categories": [
        {
            "id": 0,
            "name": "person",
        },
        {
            "id": 1,
            "name": "soccer",
        }
    ],
    "images": [ 
    ],
    "annotations": [
    ]
}

image_tp = {
            "id": 1,
            "width": 640,
            "height": 480,
            "file_name": "cat1.jpg",
            "license": 1
        }

anno_tp = {
            "id": 1,
            "image_id": 1,
            "category_id": 1,
            "bbox": [],
            "area": 0,
            "segmentation": [],
            "iscrowd": 0
        }

idx = 0
img_id_= 0 
for root, dirs, files in os.walk(label_root):
    
    for file in files:
        if file.endswith('txt'): # 遍历所有yolov8格式的txt标注文件
            txt_path = os.path.join(root,file)
            print("Current directory:", txt_path)
            img_path = txt_path.replace('labels','images')
            img_path = img_path.replace('.txt','.jpg') # 找到图片路径
            if 'old' not in img_path: # 不用管,自行修改
                anno = open(txt_path).read().splitlines()
                img = cv2.imread(img_path)
                h,w,_ = img.shape
                
                image_tp["id"] = idx
                image_tp["width"] = w
                image_tp["height"] = h
                image_tp["file_name"] = img_path # 写入完整路径
                
                coco["images"].append(image_tp.copy()) # 添加图片信息
                for a in anno:
                    l = a.split(' ')
                    cat,cx,cy,lw,lh = int(l[0]),float(l[1])*w,float(l[2])*h,float(l[3])*w,float(l[4])*h
                    anno_tp["id"] = img_id_
                    anno_tp["image_id"] = img_path
                    img_id_+=1
                    anno_tp["bbox"] = [cx-lw/2,cy-lh/2,lw,lh] # 转换标注格式
                    anno_tp["category_id"] = cat
                    anno_tp["area"] = lw*lh
                    coco["annotations"].append(anno_tp.copy())  # 添加标注信息
                idx+=1
                assert os.path.exists(img_path)
                
    # if idx>500:
    #     break

with open('./test_soccer_coco.json', 'w') as l:
    l.write(json.dumps(coco))

验证是否转换正确代码如下：

from pycocotools.coco import COCO
import os
import cv2
import numpy as np

# 替换为你的数据集标注文件的路径和图像文件夹路径
annotation_file = 'test_soccer_coco.json'
image_folder = ''

# 初始化COCO对象
coco = COCO(annotation_file)

idx = 0
# 遍历每个图像并绘制标注
for image_id in coco.getImgIds():
    image_info = coco.loadImgs(image_id)[0]
    image_path = os.path.join(image_folder, image_info['file_name'])
    image = cv2.imread(image_path)

    annotations = coco.loadAnns(coco.getAnnIds(imgIds=[image_info['file_name']]))  # 原来是imgIds=image_id,进行了修改

    for ann in annotations:
        bbox = ann['bbox']
        category_info = coco.loadCats(ann['category_id'])[0]
        category_name = category_info['name']

        # 在图像上绘制边界框
        x, y, w, h = map(int, bbox)
        if category_info['id'] == 0:
            cv2.rectangle(image, (x, y), (x + w, y + h), (0, 255, 0), 2)
            cv2.putText(image, category_name, (x, y - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 255, 0), 2)
        else:
            cv2.rectangle(image, (x, y), (x + w, y + h), (0, 0, 255), 2)
            cv2.putText(image, category_name, (x, y - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0, 0, 255), 2)
    # 保存绘制标注后的图像
   
    cv2.imwrite('tmp/{}.jpg'.format(idx), image)
    idx+=1
print("Annotation visualization and saving complete.")

另外，使用别人的代码训练的时候可能需要修改，就比如图中的位置：

有用的话点个赞哦