数据集

http://host.robots.ox.ac.uk:8080/pascal/VOC/voc2011/index.html

说明：本文使用2012

工程目录

datasets
|_____commodity______Annotations
|
|__________JPEGImages

数据预处理

from xml.etree import ElementTree as ET 
from collections import OrderedDict as Dict
import numpy as np
import json
import pickle
import os

class xmlProcess(object):
	def __init__(self,file_path):
		self.xml_path = file_path
		self.num_class = 11
		self.data = Dict()

	def process_xml(self):
		for f in os.listdir(self.xml_path):
			et = ET.parse(os.path.join(self.xml_path,f))
			root = et.getroot()
			size = root.find('size')

			width = eval(size.find("width").text)
			height =  eval(size.find("height").text)
			depth = eval(size.find("depth").text)

			bounding_boxes = []
			one_hots = []
			for obj in root.findall('object'):
				for res in obj.iter('bndbox'):
					xmin = float(res.find('xmin').text)/width
					ymin = float(res.find('ymin').text)/height
					xmax = float(res.find('xmax').text)/width
					ymax = float(res.find('ymax').text)/height

				bounding_boxes.append([xmin,ymin,xmax,ymax])
				obj_name = obj.find('name').text
				one_hot_name = self.one_hot(obj_name)
				one_hots.append(one_hot_name)

			bounding_boxes = np.asarray(bounding_boxes)
			one_hots  = np.asarray(one_hots)
			image_data = np.hstack((bounding_boxes,one_hots))
			self.data[f]=image_data 
		return None

	def one_hot(self,name):
		one_hot_vector = [0]*self.num_class
		if name == "sheep":
			one_hot_vector[0] = 1
		elif name == "aeroplane":
			one_hot_vector[1] = 1
		elif name == "boat":
			one_hot_vector[2] = 1
		elif name == "tvmonitor":
			one_hot_vector[3] = 1
		elif name == "train":
			one_hot_vector[4] = 1
		elif name == "bird":
			one_hot_vector[5] = 1
		elif name == "dog":
			one_hot_vector[6] = 1
		elif name == "chair":
			one_hot_vector[7] = 1
		elif name == "bicycle":
			one_hot_vector[8] = 1
		elif name == "bottle":
			one_hot_vector[9] = 1
		elif name == "person":
			one_hot_vector[10] = 1
		else:
			pass
		
		return one_hot_vector


if __name__ == "__main__":
	pro=xmlProcess("./commodity/Annotations")
	pro.process_xml()
	pickle.dump(pro.data,open("./image_data.pkl","wb"))
	

说明

以上代码的功能：从所有xml文件里读取标记物体的类别，物体位置（进行了归一化处理），
物体名称（进行one-hot编码），然后进行存储，为模型训练做准备。