[置顶] ImageNet和PASCAL VOC图像描述的xml文件的解析、修改和生成

一、按照我的上篇博客（点击打开链接）配置好环境；

二、根据PASCAL VOC的xml文件的格式构建基类“Annotation”。

1、PASCAL VOC的xml文件的格式如下：

<annotation>
	<folder>VOC2007</folder>
	<filename>000001.jpg</filename>
	<source>
		<database>The VOC2007 Database</database>
		<annotation>PASCAL VOC2007</annotation>
		<image>flickr</image>
		<flickrid>341012865</flickrid>
	</source>
	<owner>
		<flickrid>Fried Camels</flickrid>
		<name>Jinky the Fruit Bat</name>
	</owner>
	<size>
		<width>353</width>
		<height>500</height>
		<depth>3</depth>
	</size>
	<segmented>0</segmented>
	<object>
		<name>dog</name>
		<pose>Left</pose>
		<truncated>1</truncated>
		<difficult>0</difficult>
		<bndbox>
			<xmin>48</xmin>
			<ymin>240</ymin>
			<xmax>195</xmax>
			<ymax>371</ymax>
		</bndbox>
	</object>
	<object>
		<name>person</name>
		<pose>Left</pose>
		<truncated>1</truncated>
		<difficult>0</difficult>
		<bndbox>
			<xmin>8</xmin>
			<ymin>12</ymin>
			<xmax>352</xmax>
			<ymax>498</ymax>
		</bndbox>
	</object>
</annotation>

2、构建的Annotation类如下：

#include <stdio.h>
#include <string>
#include <iostream>
#include <vector>

using namespace std;
typedef struct _Source {
	string database;
	string annotation;
	string image;
	string flickrid;
} Source, *PSource;

typedef struct _Owner {
	string flickrid;
	string name;
} Owner, *POwner;

typedef struct _Size {
	int width;
	int height;
	int depth;
} Size, *PSize;

typedef struct _RECT {
	int xmin;
	int ymin;
	int xmax;
	int ymax;
} RECT, *PRECT;

typedef struct _Object {
	string name;
	string pose;
	string truncated;
	string difficult;
	RECT  bndbox;
} Object, *PObject;

class Annotation
{
public:
	string folder;
	string filename;
	Source source;
	Owner owner;
	Size size;
	string segmented;
	vector<Object> objectList;
	Object object;
	Annotation();
	virtual ~Annotation();
};

三、xml文件读取：

void XMLRead(const char* xmlPath, Annotation & annotation){
	xmlDocPtr doc;
	xmlNodePtr curNode;
	xmlChar *szKey;
	string filename = "";
	doc = xmlReadFile(xmlPath, "UTF-8", XML_PARSE_RECOVER); //解析文件"GB2312"
	if (doc == NULL){
		fprintf(stderr, "Document not parsed successfully.\n");
	}
	curNode = xmlDocGetRootElement(doc); //确定文档根元素
	if (NULL == curNode){
		fprintf(stderr, "empty document\n");
		xmlFreeDoc(doc);
	}
	if (xmlStrcmp(curNode->name, BAD_CAST "annotation")) {
		fprintf(stderr, "document of the wrong type, root node != Message");
		xmlFreeDoc(doc);
	}

	curNode = curNode->xmlChildrenNode;
	xmlNodePtr propNodePtr = curNode;
	xmlNodePtr sourceNodePtr = NULL;
	xmlNodePtr ownerNodePtr = NULL;
	xmlNodePtr sizeNodePtr = NULL;
	xmlNodePtr objectNodePtr = NULL;
	vector<xmlNodePtr> objectListP;
	xmlNodePtr objectNodePtr2 = NULL;
	xmlNodePtr bndboxNodePtr = NULL;

	// annotation
	while (curNode != NULL){
		//取出文件名称
		if (!xmlStrcmp(curNode->name, BAD_CAST "folder")) {
			annotation.folder = (char*)xmlNodeGetContent(curNode);
		}
		if (!xmlStrcmp(curNode->name, BAD_CAST "filename")) {
			annotation.filename = (char*)xmlNodeGetContent(curNode);
		}
		if (!xmlStrcmp(curNode->name, BAD_CAST "source")) {
			sourceNodePtr = curNode->xmlChildrenNode;
		}
		if (!xmlStrcmp(curNode->name, BAD_CAST "owner")) {
			ownerNodePtr = curNode->xmlChildrenNode;
		}
		if (!xmlStrcmp(curNode->name, BAD_CAST "size")) {
			sizeNodePtr = curNode->xmlChildrenNode;
		}
		if (!xmlStrcmp(curNode->name, BAD_CAST "segmented")) {
			annotation.segmented = (char*)xmlNodeGetContent(curNode);
		}
		if (!xmlStrcmp(curNode->name, BAD_CAST "object")) {
			objectNodePtr = curNode->xmlChildrenNode;
			objectListP.push_back(objectNodePtr);
		}
		curNode = curNode->next;
	}
	// source
	while (sourceNodePtr != NULL){
		if (!xmlStrcmp(sourceNodePtr->name, BAD_CAST "database")) {
			annotation.source.database = (char*)xmlNodeGetContent(sourceNodePtr);
		}
		if (!xmlStrcmp(sourceNodePtr->name, BAD_CAST "annotation")) {
			annotation.source.annotation = (char*)xmlNodeGetContent(sourceNodePtr);
		}
		if (!xmlStrcmp(sourceNodePtr->name, BAD_CAST "image")) {
			annotation.source.image = (char*)xmlNodeGetContent(sourceNodePtr);
		}
		if (!xmlStrcmp(sourceNodePtr->name, BAD_CAST "flickrid")) {
			annotation.source.flickrid = (char*)xmlNodeGetContent(sourceNodePtr);
		}
		sourceNodePtr = sourceNodePtr->next;
	}
	// owner
	while (ownerNodePtr != NULL){
		if (!xmlStrcmp(ownerNodePtr->name, BAD_CAST "flickrid")) {
			annotation.owner.flickrid = (char*)xmlNodeGetContent(ownerNodePtr);
		}
		if (!xmlStrcmp(ownerNodePtr->name, BAD_CAST "name")) {
			annotation.owner.name = (char*)xmlNodeGetContent(ownerNodePtr);
		}
		ownerNodePtr = ownerNodePtr->next;
	}
	// size
	while (sizeNodePtr != NULL){
		if (!xmlStrcmp(sizeNodePtr->name, BAD_CAST "width")) {
			annotation.size.width = atoi((char*)xmlNodeGetContent(sizeNodePtr));
		}
		if (!xmlStrcmp(sizeNodePtr->name, BAD_CAST "height")) {
			annotation.size.height = atoi((char*)xmlNodeGetContent(sizeNodePtr));
		}
		if (!xmlStrcmp(sizeNodePtr->name, BAD_CAST "depth")) {
			annotation.size.depth = atoi((char*)xmlNodeGetContent(sizeNodePtr));
		}
		sizeNodePtr = sizeNodePtr->next;
	}
	// object
	for (int i = 0; i < objectListP.size(); i++){
		objectNodePtr = objectListP[i];
		while (objectNodePtr != NULL)
		{
			if (!xmlStrcmp(objectNodePtr->name, BAD_CAST "name")) {
				annotation.object.name = (char*)xmlNodeGetContent(objectNodePtr);
			}
			if (!xmlStrcmp(objectNodePtr->name, BAD_CAST "pose")) {
				annotation.object.pose = (char*)xmlNodeGetContent(objectNodePtr);
			}
			if (!xmlStrcmp(objectNodePtr->name, BAD_CAST "truncated")) {
				annotation.object.truncated = (char*)xmlNodeGetContent(objectNodePtr);
			}
			if (!xmlStrcmp(objectNodePtr->name, BAD_CAST "difficult")) {
				annotation.object.difficult = (char*)xmlNodeGetContent(objectNodePtr);
			}
			if (!xmlStrcmp(objectNodePtr->name, BAD_CAST "bndbox")) {
				bndboxNodePtr = objectNodePtr->xmlChildrenNode;
			}
			objectNodePtr = objectNodePtr->next;
		}
		// bndbox
		while (bndboxNodePtr != NULL){
			if (!xmlStrcmp(bndboxNodePtr->name, BAD_CAST "xmin")) {
				annotation.object.bndbox.xmin = atoi((char*)xmlNodeGetContent(bndboxNodePtr));
			}
			if (!xmlStrcmp(bndboxNodePtr->name, BAD_CAST "ymin")) {
				annotation.object.bndbox.ymin = atoi((char*)xmlNodeGetContent(bndboxNodePtr));
			}
			if (!xmlStrcmp(bndboxNodePtr->name, BAD_CAST "xmax")) {
				annotation.object.bndbox.xmax = atoi((char*)xmlNodeGetContent(bndboxNodePtr));
			}
			if (!xmlStrcmp(bndboxNodePtr->name, BAD_CAST "ymax")) {
				annotation.object.bndbox.ymax = atoi((char*)xmlNodeGetContent(bndboxNodePtr));
			}
			bndboxNodePtr = bndboxNodePtr->next;
		}
		annotation.objectList.push_back(annotation.object);
	}
	xmlFreeDoc(doc);
}

四、XML文件生成和保存

void XMLWrite(const char* xmlPath, Annotation & annotation){
	xmlDocPtr doc = NULL;
	xmlNodePtr root_node = NULL, node = NULL, objectNode = NULL, node1 = NULL;
	//Creates a new document, a node and set it as a root node
	doc = xmlNewDoc(BAD_CAST "1.0");
	root_node = xmlNewNode(NULL, BAD_CAST "annotation");
	//xmlNewProp(root_node, BAD_CAST "version", BAD_CAST "1.0");
	xmlDocSetRootElement(doc, root_node);

	//creates a new node, which is "attached" as child node of root_node node.
	xmlNewChild(root_node, NULL, BAD_CAST "folder", BAD_CAST annotation.folder.c_str());
	xmlNewChild(root_node, NULL, BAD_CAST "filename", BAD_CAST annotation.filename.c_str());

	node = xmlNewChild(root_node, NULL, BAD_CAST "source", NULL);
	xmlNewChild(node, NULL, BAD_CAST "database", BAD_CAST annotation.source.database.c_str());
	xmlNewChild(node, NULL, BAD_CAST "annotation", BAD_CAST annotation.source.annotation.c_str());
	xmlNewChild(node, NULL, BAD_CAST "image", BAD_CAST annotation.source.image.c_str());
	xmlNewChild(node, NULL, BAD_CAST "flickrid", BAD_CAST annotation.source.flickrid.c_str());

	node = xmlNewChild(root_node, NULL, BAD_CAST "owner", NULL);
	xmlNewChild(node, NULL, BAD_CAST "flickrid", BAD_CAST annotation.owner.flickrid.c_str());
	xmlNewChild(node, NULL, BAD_CAST "name", BAD_CAST annotation.owner.name.c_str());

	node = xmlNewChild(root_node, NULL, BAD_CAST "size", NULL);
	char temp[32];
	_itoa(annotation.size.width, temp, 10);
	xmlNewChild(node, NULL, BAD_CAST "width", BAD_CAST _itoa(annotation.size.width, temp, 10));
	xmlNewChild(node, NULL, BAD_CAST "height", BAD_CAST _itoa(annotation.size.height, temp, 10));
	xmlNewChild(node, NULL, BAD_CAST "depth", BAD_CAST _itoa(annotation.size.depth, temp, 10));

	xmlNewChild(root_node, NULL, BAD_CAST "segmented", BAD_CAST annotation.segmented.c_str());

	for (int i = 0; i < annotation.objectList.size(); i++){
		objectNode = xmlNewChild(root_node, NULL, BAD_CAST "object", NULL);
		xmlNewChild(objectNode, NULL, BAD_CAST "name", BAD_CAST annotation.objectList[i].name.c_str());
		xmlNewChild(objectNode, NULL, BAD_CAST "pose", BAD_CAST annotation.objectList[i].pose.c_str());
		xmlNewChild(objectNode, NULL, BAD_CAST "truncated", BAD_CAST annotation.objectList[i].truncated.c_str());
		xmlNewChild(objectNode, NULL, BAD_CAST "difficult", BAD_CAST annotation.objectList[i].difficult.c_str());
		node = xmlNewChild(objectNode, NULL, BAD_CAST "bndbox", NULL);
		xmlNewChild(node, NULL, BAD_CAST "xmin", BAD_CAST _itoa(annotation.objectList[i].bndbox.xmin, temp, 10));
		xmlNewChild(node, NULL, BAD_CAST "ymin", BAD_CAST _itoa(annotation.objectList[i].bndbox.ymin, temp, 10));
		xmlNewChild(node, NULL, BAD_CAST "xmax", BAD_CAST _itoa(annotation.objectList[i].bndbox.xmax, temp, 10));
		xmlNewChild(node, NULL, BAD_CAST "ymax", BAD_CAST _itoa(annotation.objectList[i].bndbox.ymax, temp, 10));
	}

	//// xmlNewProp() creates attributes, which is "attached" to an node.
	//node = xmlNewChild(root_node, NULL, BAD_CAST "node3", BAD_CAST"node has attributes");
	//xmlNewProp(node, BAD_CAST "attribute", BAD_CAST "yes");
	////Here goes another way to create nodes.
	//node = xmlNewNode(NULL, BAD_CAST "node4");
	//node1 = xmlNewText(BAD_CAST"other way to create content");
	//xmlAddChild(node, node1);
	//xmlAddChild(root_node, node);

	xmlSaveFormatFileEnc(xmlPath, doc, "UTF-8", 1);//Dumping document to stdio or file
	xmlFreeDoc(doc);
	xmlCleanupParser();
	xmlMemoryDump();//debug memory for regression tests
}

五、对PASCAL VOC图像描述的xml文件进行解析，然后重新生成。

int main(){
	const char* xmlPath = "C:\\Users\\zyy\\Desktop\\XML_test\\PASCAL\\000001.xml";
	//const char* xmlPath = "C:\\Users\\zyy\\Desktop\\XML_test\\ImageNet\\n01440764_18.xml";
	Annotation annotation;
	XMLRead(xmlPath, annotation);
	XMLWrite("./temp.xml", annotation);
	//system("pause");
	return 0;
}

结果如下：

<?xml version="1.0" encoding="UTF-8"?>
<annotation>
  <folder>VOC2007</folder>
  <filename>000001.jpg</filename>
  <source>
    <database>The VOC2007 Database</database>
    <annotation>PASCAL VOC2007</annotation>
    <image>flickr</image>
    <flickrid>341012865</flickrid>
  </source>
  <owner>
    <flickrid>Fried Camels</flickrid>
    <name>Jinky the Fruit Bat</name>
  </owner>
  <size>
    <width>353</width>
    <height>500</height>
    <depth>3</depth>
  </size>
  <segmented>0</segmented>
  <object>
    <name>dog</name>
    <pose>Left</pose>
    <truncated>1</truncated>
    <difficult>0</difficult>
    <bndbox>
      <xmin>48</xmin>
      <ymin>240</ymin>
      <xmax>195</xmax>
      <ymax>371</ymax>
    </bndbox>
  </object>
  <object>
    <name>person</name>
    <pose>Left</pose>
    <truncated>1</truncated>
    <difficult>0</difficult>
    <bndbox>
      <xmin>8</xmin>
      <ymin>12</ymin>
      <xmax>352</xmax>
      <ymax>498</ymax>
    </bndbox>
  </object>
</annotation>