百度ApolloScape Dataset 用于目标检测任务

百度近期开源了自动驾驶大型数据集，官方网址：http://apolloscape.auto/index.html

数据集的具体介绍就看官方网址就行，如何利用这个数据集做目标检测呢？就是要把像素的实例级分割转为包围框xmin,xmax,ymin,ymax即可。以处理road_04为例，代码展示如何使用。

import json
import os  
from os import listdir, getcwd
from os.path import join
import os.path
rootdir='/home/wang/下载/数据集/apolloscape/Image/road04_ins'　＃图像文件夹绝对地址
def position(pos):
    x=[]
    y=[]
    nums=len(pos)
    for i in range(nums):
        x.append(pos[i][0])
        y.append(pos[i][1])
    x_max=max(x)
    x_min=min(x)
    y_max=max(y)
    y_min=min(y)
    b=(float(x_min),float(x_max),float(y_min),float(y_max))
    #b=(x_min,x_max,y_min,y_max)
    return b
 
def convert(size, box):
    dw = 1./(size[0])
    dh = 1./(size[1])
    x = (box[0] + box[1])/2.0 - 1
    y = (box[2] + box[3])/2.0 - 1
    w = box[1] - box[0]
    h = box[3] - box[2]
    x = x*dw
    w = w*dw
    y = y*dh
    h = h*dh
    return (x,y,w,h)
 
def convert_annotation(image_id):
    load_f=open("./label/road04_ins/%s.json"%(image_id),'r')　＃标签文件夹相对地址
    load_dict = json.load(load_f)
    out_file = open('./voc_type/road04_ins/%s.txt'%(image_id), 'w')　＃输出标签存放地址
    #keys=tuple(load_dict.keys())  
    w=load_dict['imgWidth']
    h=load_dict['imgHeight']
    #print(h)
    objects=load_dict['objects']
    nums=len(objects)
    #print(nums)
    #object_key=tuple(objects.keys()
    
     
     
    for i in range(0,nums):
        labels=objects[i]['label']
        #print(i)
        if (str(labels) in ['36','37']):
            #print(labels)
            pos=objects[i]['polygons'][0]    
            b=position(pos)
            bb = convert((w,h), b)
            cls_id=2
            out_file.write(str(cls_id) + " " + " ".join([str(a) for a in bb]) + '\n')
            #print(type(pos))
        elif (str(labels) in['33','38','39','40']):
            #print(labels)
            pos=objects[i]['polygons'][0]
            b=position(pos)
            bb = convert((w,h), b)
            cls_id=1
            out_file.write(str(cls_id) + " " + " ".join([str(a) for a in bb]) + '\n')
 
def image_id(rootdir):
    a=[]
    for parent,dirnames,filenames in os.walk(rootdir):
        for filename in filenames:
            filename=filename.strip('.jpg')
            #print(filename)
            a.append(filename)
    return a
names=image_id(rootdir)
for image_id in names:
    convert_annotation(image_id)

发现图片比标签多时，就用做差删除无标签图片：

#encoding:utf-8
import os
import os.path
from os import listdir, getcwd
from os.path import join
txt_dir='./label/road04_ins'  
pic_dir='./Image/road04_ins'


def txt(rootdir):
    a=[]
    for parent,dirnames,filenames in os.walk(rootdir):
        for filenames in filenames:
            filenames=filenames.strip('.json')
            a.append(filenames)
    return a
def pic(rootdir):
    b=[]
    for parent,dirnames,filenames in os.walk(rootdir):
        for filenames in filenames:
            filenames=filenames.strip('.jpg')
            b.append(filenames)
    return b

txt_set=txt(txt_dir)
txt_set=set(txt_set)
pic_set=pic(pic_dir)
pic_set=set(pic_set)
#comp=txt_set-pic_set
comp=pic_set-txt_set
print("ok")
print(len(comp))

for item in comp:
    file=pic_dir+'/'+item+'.jpg'
    if os.path.exists(file):
        os.remove(file)
        print(file)
        
#for item in comp:
 #   file=txt_dir+'/'+item+'.json'
  #  if os.path.exists(file):
   #     os.remove(file)
    #    print(file)