VOC数据集格式转化成COCO数据集格式

VOC数据集格式转化成COCO数据集格式
一、唠叨
之前写过一篇关于coco数据集转化成VOC格式的博客COCO2VOC，最近读到CenterNet的官方代码，实现上则是将voc转化成coco数据格式，这样的操作我个人感觉很不习惯，也觉得有些奇葩，可能是每个人习惯不一样吧，我们知道有时候我们会采用labelImg标注数据，标注出来的格式就是voc，如果直接训练就可以用来训练是不是更加友好。

为了不大规模修改原始训练代码（虽然自己修改了一个版本的voc数据集就能直接训练centernet 😊）,同时也看到网上很多大佬也做了将VOC数据格式转化成COCO用于其训练，这里我自己做一个精细一点的，作为笔记。

二、转化过程
数据格式的转换实际是annotation标注文件的转化，voc的数据标注文件为以.xml结尾的文件，而且每张图片均有一个对应的同名标注文件；COCO则是将所有的标注信息写在一个json文件中。VOC数据集目录如下：

在之前的coco2voc博客中做了详细的介绍，现在直接开始转化，目标就是将Annotations中的所有标注文件中的bbox标注信息转化为json文件，根据训练集和测试集，则主要转化为四个json文件，分别是test.json、train.json、val.json和trainval.json .这里我根据ImageSets中的train.txt val.txt trainval.txt生成后三个json文件，当然也可以直接从Annotations文件夹生成。

VOC2COCO.py

import xml.etree.ElementTree as ET
import os
import json
 
coco = dict()
coco['images'] = []
coco['type'] = 'instances'
coco['annotations'] = []
coco['categories'] = []
 
category_set = dict()
image_set = set()
 
category_item_id = -1
image_id = 20180000000
annotation_id = 0
 
def addCatItem(name):
    global category_item_id
    category_item = dict()
    category_item['supercategory'] = 'none'
    category_item_id += 1
    category_item['id'] = category_item_id
    category_item['name'] = name
    coco['categories'].append(category_item)
    category_set[name] = category_item_id
    return category_item_id
 
def addImgItem(file_name, size):
    global image_id
    if file_name is None:
        raise Exception('Could not find filename tag in xml file.')
    if size['width'] is None:
        raise Exception('Could not find width tag in xml file.')
    if size['height'] is None:
        raise Exception('Could not find height tag in xml file.')
    image_id += 1
    image_item = dict()
    image_item['id'] = image_id
    image_item['file_name'] = file_name
    image_item['width'] = size['width']
    image_item['height'] = size['height']
    coco['images'].append(image_item)
    image_set.add(file_name)
    return image_id
 
def addAnnoItem(object_name, image_id, category_id, bbox):
    global annotation_id
    annotation_item = dict()
    annotation_item['segmentation'] = []
    seg = []
    # bbox[] is x,y,w,h
    # left_top
    seg.append(bbox[0])
    seg.append(bbox[1])
    # left_bottom
    seg.append(bbox[0])
    seg.append(bbox[1] + bbox[3])
    # right_bottom
    seg.append(bbox[0] + bbox[2])
    seg.append(bbox[1] + bbox[3])
    # right_top
    seg.append(bbox[0] + bbox[2])
    seg.append(bbox[1])
 
    annotation_item['segmentation'].append(seg)
 
    annotation_item['area'] = bbox[2] * bbox[3]
    annotation_item['iscrowd'] = 0
    annotation_item['ignore'] = 0
    annotation_item['image_id'] = image_id
    annotation_item['bbox'] = bbox
    annotation_item['category_id'] = category_id
    annotation_id += 1
    annotation_item['id'] = annotation_id
    coco['annotations'].append(annotation_item)
 
def _read_image_ids(image_sets_file):
    ids = []
    with open(image_sets_file) as f:
        for line in f:
            ids.append(line.rstrip())
    return ids
 
"""通过txt文件生成"""
#split ='train' 'va' 'trainval' 'test'
def parseXmlFiles_by_txt(data_dir,json_save_path,split='train'):
    print("hello")
    labelfile=split+".txt"
    image_sets_file = data_dir + "/ImageSets/Main/"+labelfile
    ids=_read_image_ids(image_sets_file)
 
    for _id in ids:
        xml_file=data_dir + f"/Annotations/{_id}.xml"
 
        bndbox = dict()
        size = dict()
        current_image_id = None
        current_category_id = None
        file_name = None
        size['width'] = None
        size['height'] = None
        size['depth'] = None
 
        tree = ET.parse(xml_file)
        root = tree.getroot()
        if root.tag != 'annotation':
            raise Exception('pascal voc xml root element should be annotation, rather than {}'.format(root.tag))
 
        # elem is <folder>, <filename>, <size>, <object>
        for elem in root:
            current_parent = elem.tag
            current_sub = None
            object_name = None
 
            if elem.tag == 'folder':
                continue
 
            if elem.tag == 'filename':
                file_name = elem.text
                if file_name in category_set:
                    raise Exception('file_name duplicated')
 
            # add img item only after parse <size> tag
            elif current_image_id is None and file_name is not None and size['width'] is not None:
                if file_name not in image_set:
                    current_image_id = addImgItem(file_name, size)
                    print('add image with {} and {}'.format(file_name, size))
                else:
                    raise Exception('duplicated image: {}'.format(file_name))
                    # subelem is <width>, <height>, <depth>, <name>, <bndbox>
            for subelem in elem:
                bndbox['xmin'] = None
                bndbox['xmax'] = None
                bndbox['ymin'] = None
                bndbox['ymax'] = None
 
                current_sub = subelem.tag
                if current_parent == 'object' and subelem.tag == 'name':
                    object_name = subelem.text
                    if object_name not in category_set:
                        current_category_id = addCatItem(object_name)
                    else:
                        current_category_id = category_set[object_name]
 
                elif current_parent == 'size':
                    if size[subelem.tag] is not None:
                        raise Exception('xml structure broken at size tag.')
                    size[subelem.tag] = int(subelem.text)
 
                # option is <xmin>, <ymin>, <xmax>, <ymax>, when subelem is <bndbox>
                for option in subelem:
                    if current_sub == 'bndbox':
                        if bndbox[option.tag] is not None:
                            raise Exception('xml structure corrupted at bndbox tag.')
                        bndbox[option.tag] = int(option.text)
 
                # only after parse the <object> tag
                if bndbox['xmin'] is not None:
                    if object_name is None:
                        raise Exception('xml structure broken at bndbox tag')
                    if current_image_id is None:
                        raise Exception('xml structure broken at bndbox tag')
                    if current_category_id is None:
                        raise Exception('xml structure broken at bndbox tag')
                    bbox = []
                    # x
                    bbox.append(bndbox['xmin'])
                    # y
                    bbox.append(bndbox['ymin'])
                    # w
                    bbox.append(bndbox['xmax'] - bndbox['xmin'])
                    # h
                    bbox.append(bndbox['ymax'] - bndbox['ymin'])
                    print('add annotation with {},{},{},{}'.format(object_name, current_image_id, current_category_id,
                                                                   bbox))
                    addAnnoItem(object_name, current_image_id, current_category_id, bbox)
    json.dump(coco, open(json_save_path, 'w'))
 
"""直接从xml文件夹中生成"""
def parseXmlFiles(xml_path,json_save_path):
    for f in os.listdir(xml_path):
        if not f.endswith('.xml'):
            continue
 
        bndbox = dict()
        size = dict()
        current_image_id = None
        current_category_id = None
        file_name = None
        size['width'] = None
        size['height'] = None
        size['depth'] = None
 
        xml_file = os.path.join(xml_path, f)
        print(xml_file)
 
        tree = ET.parse(xml_file)
        root = tree.getroot()
        if root.tag != 'annotation':
            raise Exception('pascal voc xml root element should be annotation, rather than {}'.format(root.tag))
 
        # elem is <folder>, <filename>, <size>, <object>
        for elem in root:
            current_parent = elem.tag
            current_sub = None
            object_name = None
 
            if elem.tag == 'folder':
                continue
 
            if elem.tag == 'filename':
                file_name = elem.text
                if file_name in category_set:
                    raise Exception('file_name duplicated')
 
            # add img item only after parse <size> tag
            elif current_image_id is None and file_name is not None and size['width'] is not None:
                if file_name not in image_set:
                    current_image_id = addImgItem(file_name, size)
                    print('add image with {} and {}'.format(file_name, size))
                else:
                    raise Exception('duplicated image: {}'.format(file_name))
                    # subelem is <width>, <height>, <depth>, <name>, <bndbox>
            for subelem in elem:
                bndbox['xmin'] = None
                bndbox['xmax'] = None
                bndbox['ymin'] = None
                bndbox['ymax'] = None
 
                current_sub = subelem.tag
                if current_parent == 'object' and subelem.tag == 'name':
                    object_name = subelem.text
                    if object_name not in category_set:
                        current_category_id = addCatItem(object_name)
                    else:
                        current_category_id = category_set[object_name]
 
                elif current_parent == 'size':
                    if size[subelem.tag] is not None:
                        raise Exception('xml structure broken at size tag.')
                    size[subelem.tag] = int(subelem.text)
 
                # option is <xmin>, <ymin>, <xmax>, <ymax>, when subelem is <bndbox>
                for option in subelem:
                    if current_sub == 'bndbox':
                        if bndbox[option.tag] is not None:
                            raise Exception('xml structure corrupted at bndbox tag.')
                        bndbox[option.tag] = int(option.text)
 
                # only after parse the <object> tag
                if bndbox['xmin'] is not None:
                    if object_name is None:
                        raise Exception('xml structure broken at bndbox tag')
                    if current_image_id is None:
                        raise Exception('xml structure broken at bndbox tag')
                    if current_category_id is None:
                        raise Exception('xml structure broken at bndbox tag')
                    bbox = []
                    # x
                    bbox.append(bndbox['xmin'])
                    # y
                    bbox.append(bndbox['ymin'])
                    # w
                    bbox.append(bndbox['xmax'] - bndbox['xmin'])
                    # h
                    bbox.append(bndbox['ymax'] - bndbox['ymin'])
                    print('add annotation with {},{},{},{}'.format(object_name, current_image_id, current_category_id,
                                                                   bbox))
                    addAnnoItem(object_name, current_image_id, current_category_id, bbox)
    json.dump(coco, open(json_save_path, 'w'))
 
 
 
if __name__ == '__main__':
    #通过txt文件生成
    # voc_data_dir="E:/VOCdevkit/VOC2007"
    # json_save_path="E:/VOCdevkit/voc2007trainval.json"
    # parseXmlFiles_by_txt(voc_data_dir,json_save_path,"trainval")
 
    #通过文件夹生成
    ann_path="E:/VOCdevkit/VOC2007/Annotations"
    json_save_path="E:/VOCdevkit/test.json"
    parseXmlFiles(ann_path,json_save_path)

生成之后就是这样：