@[toc]

DanceTrack 运动跟踪数据集

简介

DanceTrack 是一个大规模的多对象跟踪数据集。用于在遮挡、频繁交叉、同样服装和多样化身体姿态条件下对人进行跟踪。强调运动分析在多对象跟踪中的重要性。

GitHub地址：https://github.com/DanceTrack/DanceTrack
数据集下载地址：https://pan.baidu.com/s/19O3IvYNzzrcLqlODHKYUwA
提取码：awew

转为Labelme标注的物体检测数据集格式

import sys
import base64
import os
import cv2
import shutil
import glob
module_path = os.path.abspath(os.path.join('..'))
if module_path not in sys.path:
    sys.path.append(module_path)
import json
from PIL import Image

Image.MAX_IMAGE_PIXELS = None

xmlpathNames_path='../train1/*/gt/gt.txt'
xmlpathNames=glob.glob(xmlpathNames_path)
print(xmlpathNames)
version = '3.16.7'
flags = {}
lineColor = [0, 255, 0, 128]
fillColor = [255, 0, 0, 128]
image_t='../images/'
os.makedirs(image_t,exist_ok=True)
for xmlpathName in xmlpathNames:
    xmlpathName=xmlpathName.replace("\\","/")
    dancetrack_name=xmlpathName.split("/")[-3]
    dic_info = {}
    with open(xmlpathName) as fs:
        lines = fs.readlines()
        lines = sorted(lines)
        for line in lines:
            line = line.replace("\n", '')
            line_info = line.split(',')
            frame = line_info[0]
            frame_image_name = '{:0>8d}'.format(int(frame)) + ".jpg"
            box = [int(line_info[2]), int(line_info[3]), int(line_info[2]) + int(line_info[4]),
                   int(line_info[3]) + int(line_info[5])]
            if frame_image_name in dic_info:
                box_list = dic_info[frame_image_name]
                box_list.append(box)
                dic_info[frame_image_name] = box_list
            else:
                box_list = [box]
                dic_info[frame_image_name] = box_list
        for image_name in dic_info.keys():
            dic = {}
            dic['version'] = version
            dic['flags'] = flags
            dic['shapes'] = []
            img_path = "../train1/"+dancetrack_name+"/img1/" + image_name
            img_new_name = dancetrack_name + "_" + image_name
            img_new_path = image_t + img_new_name
            try:
                shutil.copy(img_path, image_t + img_new_name)
            except :
                continue
            img = cv2.imread(img_new_path)
            imageHeight, imageWidth, _ = img.shape
            for data in dic_info[image_name]:
                shape = {}
                shape['label'] = 'person'
                shape['line_color'] = None
                shape['fill_color'] = None
                x1 = int(data[0])
                y1 = int(data[1])
                x2 = int(data[2])
                y2 = int(data[3])
                shape['points'] = [[x1, y1], [x2, y2]]
                shape['shape_type'] = 'rectangle'
                shape['flags'] = {}
                dic['shapes'].append(shape)
            dic['lineColor'] = lineColor
            dic['fillColor'] = fillColor
            dic['imagePath'] = img_new_name
            dic['imageData'] = base64.b64encode(
                open('{}'.format(img_new_path), "rb").read()).decode('utf-8')
            dic['imageHeight'] = imageHeight
            dic['imageWidth'] = imageWidth

            fw = open('{}json'.format(img_new_path.replace(img_new_path.split('.')[-1], "")), 'w')
            json.dump(dic, fw)
            fw.close()

WiderPerson行人检测数据集

简介

WiderPerson 是关于户外行人检测的基准数据集。该数据集图像场景多样，不再局限于交通场景。该数据集包含 13,382 张图像，40 万个遮挡物的标注，其中 8,000 张图像用于训练，1,000 张图像用于验证，4,382 张图像用于测试。与 CityPersons 和 WIDER FACE 数据集类似，该数据集不公布测试图像的 bounding box ground truth。该数据集包含密集的行人和各种遮挡，适合进行户外环境的行人检测评估。

官网地址：http://www.cbsr.ia.ac.cn/users/sfzhang/WiderPerson/

百度网盘：https://pan.baidu.com/s/1ulMlbw_zhNUYwdyXONLrwg
提取码：uq3u

转为Labelme标注的物体检测数据集格式

import os
import numpy as np
import scipy.io as sio
import shutil
from lxml.etree import Element, SubElement, tostring
from xml.dom.minidom import parseString
import cv2
import base64
import json




if __name__ == '__main__':
    # < class_label =1: pedestrians > 行人
    # < class_label =2: riders >      骑车的
    # < class_label =3: partially-visible persons > 遮挡的部分行人
    # < class_label =4: ignore regions > 一些假人，比如图画上的人
    # < class_label =5: crowd > 拥挤人群，直接大框覆盖了

    version = '3.16.7'
    flags = {}
    lineColor = [0, 255, 0, 128]
    fillColor = [255, 0, 0, 128]
    classes = {'1': 'person',
               '2': 'person',
               '3': 'person',
  			   '4': 'person',
               '5': 'person',
                # 不需要哪个类的话直接删去
              }  # 这里如果自己只要人，可以把1-5全标记为people，也可以根据自己场景需要筛选
    VOCRoot = './images/'  # 生成的voc2007的位置
    os.makedirs(VOCRoot,exist_ok=True)
    widerDir = './WiderPerson'  # widerperson文件夹所在的路径
    wider_path = './WiderPerson/train.txt'  # widerperson文件夹所中训练集+验证集txt标签所在位置
    with open(wider_path, 'r') as f:
        imgIds = [x for x in f.read().splitlines()]

    for imgId in imgIds:
        objCount = 0  # 一个标志位，用来判断该img是否包含我们需要的标注
        filename = imgId + '.jpg'
        img_path = './WiderPerson/images/' + filename
        file_new_name='wider_'+filename
        print('Img :%s' % img_path)
        img = cv2.imread(img_path)
        width = img.shape[1]  # 获取图片尺寸
        height = img.shape[0]  # 获取图片尺寸 360
        dic = {}
        dic['version'] = version
        dic['flags'] = flags
        dic['shapes'] = []
        label_path = img_path.replace('images', 'Annotations') + '.txt'
        with open(label_path) as file:
            line = file.readline()
            count = int(line.split('\n')[0])  # 里面行人个数
            lines = file.readlines()
            for line in lines:
                cls_id = line.split(' ')[0]
                if cls_id not in classes:
                    print(cls_id)
                    continue
                shape = {}
                shape['label'] = cls_name = classes[cls_id]
                shape['line_color'] = None
                shape['fill_color'] = None
                x1 =  int(line.split(' ')[1]) + 1
                y1 = int(line.split(' ')[2]) + 1
                x2 = int(line.split(' ')[3]) + 1
                y2 = int(line.split(' ')[4].split('\n')[0]) + 1
                shape['points'] = [[x1, y1], [x2, y2]]
                shape['shape_type'] = 'rectangle'
                shape['flags'] = {}
                dic['shapes'].append(shape)
        dic['lineColor'] = lineColor
        dic['fillColor'] = fillColor
        dic['imagePath'] = filename
        dic['imageData'] = base64.b64encode(
            open('{}'.format(img_path), "rb").read()).decode('utf-8')
        dic['imageHeight'] = height
        dic['imageWidth'] = width
        suffix=(VOCRoot + file_new_name).split('.')[-1]
        fw = open('{}'.format((VOCRoot + file_new_name).replace(suffix,"json")), 'w')
        json.dump(dic, fw)
        fw.close()
        shutil.copy(img_path, VOCRoot + file_new_name)