将caltech数据集转换成VOC格式

目的：将Caltech行人数据集转换为Pascal VOC格式

• 参考来源https://www.cnblogs.com/arkenstone/p/7337077.html 但是这里面的代码有一些问题，我在其中修改了一些
• 操作步骤如下：

1. 将下载好的caltech pedestrian dataset解压，数据集下载地址，并按如下格式存放：（最好是按照下图的格式存放，不然容易报错）                                                                                                                                                                                                        
2.                             

运行程序如下：

import os, glob, argparse
import cv2
from scipy.io import loadmat
from collections import defaultdict
import numpy as np
from lxml import etree, objectify
"""
将caltech数据集格式转换成VOC格式，即.seq转换成.jpg; .vbb转换成xml
"""

def vbb_anno2dict(vbb_file, cam_id, person_types=None):
    """
    Parse caltech vbb annotation file to dict
    Args:
        vbb_file: input vbb file path
        cam_id: camera id
        person_types: list of person type that will be used (total 4 types: person, person-fa, person?, people).
            If None, all will be used:
    Return:
        Annotation info dict with filename as key and anno info as value
    """
    filename = os.path.splitext(os.path.basename(vbb_file))[0]
    annos = defaultdict(dict)
    vbb = loadmat(vbb_file)
    # object info in each frame: id, pos, occlusion, lock, posv
    objLists = vbb['A'][0][0][1][0]
    objLbl = [str(v[0]) for v in vbb['A'][0][0][4][0]]
    # person index
    if not person_types:
        person_types = ["person", "person-fa", "person?", "people"]
    person_index_list = [x for x in range(len(objLbl)) if objLbl[x] in person_types]
    for frame_id, obj in enumerate(objLists):
        if len(obj) > 0:
            frame_name = str(cam_id) + "_" + str(filename) + "_" + str(frame_id+1) + ".jpg"
            annos[frame_name] = defaultdict(list)
            annos[frame_name]["id"] = frame_name
            for fid, pos, occl in zip(obj['id'][0], obj['pos'][0], obj['occl'][0]):
                fid = int(fid[0][0]) - 1  # for matlab start from 1 not 0
                if not fid in person_index_list:  # only use bbox whose label is given person type
                    continue
                annos[frame_name]["label"] = objLbl[fid]
                pos = pos[0].tolist()
                occl = int(occl[0][0])
                annos[frame_name]["occlusion"].append(occl)
                annos[frame_name]["bbox"].append(pos)
            if not annos[frame_name]["bbox"]:
                del annos[frame_name]
    return annos


def seq2img(annos, seq_file, outdir, cam_id):
    """
    Extract frames in seq files to given output directories
    Args:
         annos: annos dict returned from parsed vbb file
         seq_file: seq file path
         outdir: frame save dir
         cam_id: camera id
    Returns:
        camera captured image size
    """
    cap = cv2.VideoCapture(seq_file)
    index = 1
    # captured frame list
    v_id = os.path.splitext(os.path.basename(seq_file))[0]
    cap_frames_index = np.sort([int(os.path.splitext(id)[0].split("_")[2]) for id in annos.keys()])
    while True:
        ret, frame = cap.read()
        if ret:
            if not index in cap_frames_index:
                index += 1
                continue
            if not os.path.exists(outdir):
                os.makedirs(outdir)
            outname = os.path.join(outdir, str(cam_id)+"_"+v_id+"_"+str(index)+".jpg")
            print("Current frame: ", v_id, str(index))
            cv2.imwrite(outname, frame)
            height, width, _ = frame.shape
        else:
            break
        index += 1
    img_size = (width, height)
    return img_size


def instance2xml_base(anno, img_size, bbox_type='xyxy'):
    """
    Parse annotation data to VOC XML format
    Args:
        anno: annotation info returned by vbb_anno2dict function
        img_size: camera captured image size
        bbox_type: bbox coordinate record format: xyxy (xmin, ymin, xmax, ymax); xywh (xmin, ymin, width, height)
    Returns:
        Annotation xml info tree
    """
    assert bbox_type in ['xyxy', 'xywh']
    E = objectify.ElementMaker(annotate=False)
    anno_tree = E.annotation(
        E.folder('VOC2014_instance/person'),
        E.filename(anno['id']),
        E.source(
            E.database('Caltech pedestrian'),
            E.annotation('Caltech pedestrian'),
            E.image('Caltech pedestrian'),
            E.url('None')
        ),
        E.size(
            E.width(img_size[0]),
            E.height(img_size[1]),
            E.depth(3)
        ),
        E.segmented(0),
    )
    for index, bbox in enumerate(anno['bbox']):
        bbox = [float(x) for x in bbox]
        if bbox_type == 'xyxy':
            xmin, ymin, w, h = bbox
            xmax = xmin+w
            ymax = ymin+h
        else:
            xmin, ymin, xmax, ymax = bbox
        xmin = int(xmin)
        ymin = int(ymin)
        xmax = int(xmax)
        ymax = int(ymax)
        if xmin < 0:
            xmin = 0
        if xmax > img_size[0] - 1:
            xmax = img_size[0] - 1
        if ymin < 0:
            ymin = 0
        if ymax > img_size[1] - 1:
            ymax = img_size[1] - 1
        if ymax <= ymin or xmax <= xmin:
            continue
        E = objectify.ElementMaker(annotate=False)
        anno_tree.append(
            E.object(
            E.name(anno['label']),
            E.bndbox(
                E.xmin(xmin),
                E.ymin(ymin),
                E.xmax(xmax),
                E.ymax(ymax)
            ),
            E.difficult(0),
            E.occlusion(anno["occlusion"][index])
            )
        )
    return anno_tree


def parse_anno_file(vbb_inputdir, seq_inputdir, vbb_outputdir, seq_outputdir, person_types=None):
    """
    Parse Caltech data stored in seq and vbb files to VOC xml format
    Args:
        vbb_inputdir: vbb file saved pth
        seq_inputdir: seq file saved path
        vbb_outputdir: vbb data converted xml file saved path
        seq_outputdir: seq data converted frame image file saved path
        person_types: list of person type that will be used (total 4 types: person, person-fa, person?, people).
            If None, all will be used:
    """
    # annotation sub-directories in hda annotation input directory
    assert os.path.exists(vbb_inputdir)
    sub_dirs = os.listdir(vbb_inputdir)
    for sub_dir in sub_dirs:
        print("Parsing annotations of camera: ", sub_dir)
        cam_id = sub_dir
        vbb_files = glob.glob(os.path.join(vbb_inputdir, sub_dir, "*.vbb"))
        for vbb_file in vbb_files:
            annos = vbb_anno2dict(vbb_file, cam_id, person_types=person_types)
            if annos:
                vbb_outdir = os.path.join(vbb_outputdir, "annotations", sub_dir, "bbox")
                # extract frames from seq
                seq_file = os.path.join(seq_inputdir, sub_dir, os.path.splitext(os.path.basename(vbb_file))[0]+".seq")
                seq_outdir = os.path.join(seq_outputdir, sub_dir, "frame")
                if not os.path.exists(vbb_outdir):
                    os.makedirs(vbb_outdir)
                if not os.path.exists(seq_outdir):
                    os.makedirs(seq_outdir)
                img_size = seq2img(annos, seq_file, seq_outdir, cam_id)
                for filename, anno in sorted(annos.items(), key=lambda x: x[0]):
                    if "bbox" in anno:
                        anno_tree = instance2xml_base(anno, img_size)
                        outfile = os.path.join(vbb_outdir, os.path.splitext(filename)[0]+".xml")
                        print("Generating annotation xml file of picture: ", filename)
                        etree.ElementTree(anno_tree).write(outfile, pretty_print=True)


def visualize_bbox(xml_file, img_file):
    import cv2
    tree = etree.parse(xml_file)
    # load image
    image = cv2.imread(img_file)
    # get bbox
    for bbox in tree.xpath('//bndbox'):
        coord = []
        for corner in bbox.getchildren():
            coord.append(int(float(corner.text)))
        # draw rectangle
        # coord = [int(x) for x in coord]
        image = cv2.rectangle(image, (coord[0], coord[1]), (coord[2], coord[3]), (0, 0, 255), 2)
    # visualize image
    cv2.imshow("test", image)
    cv2.waitKey(0)

#注意！以下输入输出目录地址请参照上文示意图，确保seq文件被正确读入，不然可能会产生空文件夹或报错

def main():
    # parser = argparse.ArgumentParser()
    # parser.add_argument("seq_dir", default="C:\\workspace\\data\\images", help="Caltech dataset seq data root directory")
    # parser.add_argument("vbb_dir", default="C:\\workspace\\data\\annotations", help="Caltech dataset vbb data root directory")
    # parser.add_argument("output_dir",default="C:\\workspace\\data\\VOCdevkit2007\\Caltech", help="Root saving path for frame and annotation files")
    # # parser.add_argument("person_type", default="person", type=str, help="Person type extracted within 4 options: "
    # #                                                   "'person', 'person-fa', 'person?', 'people'. If multiple type used,"
    # #                                                   "separated with comma",
    # #                     choices=["person", "person-fa", "person?", "people"])
    # parser.add_argument("person_type", default="person", type=str)
    # args = parser.parse_args()
    outdir = "/VOCdevkit2007"
    #outdir = args.output_dir
    frame_out = os.path.join(outdir, "frame")
    anno_out = os.path.join(outdir, "annotation")
    # person_type = args.person_type.split(",")
    person_type = "person"
    seq_dir = "/Caltech"
    vbb_dir = "/Caltech/annotations"
    parse_anno_file(vbb_dir, seq_dir, anno_out, frame_out, person_type)
    print("Generating done!")
#
# def test():
#     seq_inputdir = "/startdt_data/caltech_pedestrian_dataset"
#     vbb_inputdir = "/startdt_data/caltech_pedestrian_dataset/annotations"
#     seq_outputdir = "/startdt_data/caltech_pedestrian_dataset/test"
#     vbb_outputdir = "/startdt_data/caltech_pedestrian_dataset/test"
#     person_types = ["person"]
#     parse_anno_file(vbb_inputdir, seq_inputdir, vbb_outputdir, seq_outputdir, person_types=person_types)
#
#     # xml_file = "/startdt_data/caltech_pedestrian_dataset/annotations/set00/bbox/set00_V013_1511.xml"
#     # img_file = "/startdt_data/caltech_pedestrian_dataset/set00/frame/set00_V013_1511.jpg"
#     # visualize_bbox(xml_file, img_file)


if __name__ == "__main__":
    main()

 

• 出现的一个错误：UnboundLocalError: local variable 'width' referenced before assignment

• 原因：.seq文件没有读到导致的，要检查一下输入目录https://github.com/CasiaFan/Dataset_to_VOC_converter/issues/1

 

 有一个问题值得注意：caltech中与行人有关的类有四种：person_types = ["person", "person-fa", "person?", "people"]

如果不特别设置，直接用一些现成的代码去转换.seq文件为.jpg是把所有frame都截取出来，而一些代码中转换vbb文件为xml文件时却只把person类的标注出来，就会导致images和annotation最后的总数不一致，如：https://blog.csdn.net/qq_33297776/article/details/79869813

我参考的代码设置了.seq和.vbb均只截取person类，所以数量一致