faster rcnn 代码解析10

lib/datasets/voc_eval.py
该函数返回准确率、召回率和AP（average precision）

# --------------------------------------------------------
# Fast/er R-CNN
# Licensed under The MIT License [see LICENSE for details]
# Written by Bharath Hariharan
# --------------------------------------------------------

import xml.etree.ElementTree as ET
import os
import cPickle#序列化存储模块
import numpy as np

def parse_rec(filename):#解析读取XML函数
    """ Parse a PASCAL VOC xml file """
    tree = ET.parse(filename)
    objects = []
    for obj in tree.findall('object'):
        obj_struct = {}
        obj_struct['name'] = obj.find('name').text#。text返回name标签中内容例如ant
        obj_struct['pose'] = obj.find('pose').text#unspecified
        obj_struct['truncated'] = int(obj.find('truncated').text)# int(0)
        obj_struct['difficult'] = int(obj.find('difficult').text)#int(0)
        bbox = obj.find('bndbox')#bndbox中包括{ xmin ymin xmax ymax}
        obj_struct['bbox'] = [int(bbox.find('xmin').text),
                              int(bbox.find('ymin').text),
                              int(bbox.find('xmax').text),
                              int(bbox.find('ymax').text)]
        objects.append(obj_struct)#append() 方法用于在列表末尾添加新的对象。将{ xmin ymin xmax ymax}逐个添加到列表中

    return objects

def voc_ap(rec, prec, use_07_metric=False):
    """ ap = voc_ap(rec, prec, [use_07_metric度量的，十进制])
    Compute VOC AP given precision准确率（prec） and recall召回率(rec).
    If use_07_metric is true, uses the
    VOC 07 11 point method (default:False).
    """
'''
AP：average precision 一个评价指标（evaluation measure）.在VOC07计算ap时,取rec上的11个位置[0:0.1:0.1:1]，然后得到近似的ap；12之后取所有rec上的不同坐标值，计算ap，由于两个precision对于recall是分段线性的，故可以得到精确的ap值
'''
    if use_07_metric:#默认为false
        # 11 point metric
        ap = 0.
        for t in np.arange(0., 1.1, 0.1):#采样计算，取11个点
    #t=0.,0.1,0.2，...，1.0隔0.1取一值，最后的1.1取不到
            if np.sum(rec >= t) == 0:#若为空，准确率则设置为0
                p = 0
            else:
                p = np.max(prec[rec >= t])#计算准确率的最大值
            ap = ap + p / 11.
    else:
        # correct AP calculation估算、计算
        # first append sentinel values（标记值，标志值） at the end
        mrec = np.concatenate(([0.], rec, [1.]))#拼接，默认axis=0，按行拼接[[0.],[rec],[1.0]]
        mpre = np.concatenate(([0.], prec, [0.]))

        # compute the precision envelope
        for i in range(mpre.size - 1, 0, -1):#i =mpre.size-1,mpre.size -2,...,2,1
            mpre[i - 1] = np.maximum(mpre[i - 1], mpre[i])#找到prec和0中的最大值

        # to calculate area under PR curve, look for points
        # where X axis (recall) changes value
        i = np.where(mrec[1:] != mrec[:-1])[0]#找到rec的突变点,确定这些点的坐标

        # and sum (\Delta recall) * prec
        ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1])#area= sum((x2-x1)*y)
    return ap

def voc_eval(detpath, #主函数
             annopath,
             imagesetfile,
             classname,
             cachedir,
             ovthresh=0.5,
             use_07_metric=False):
    """rec, prec, ap = voc_eval(detpath,
                                annopath,
                                imagesetfile,
                                classname,
                                [ovthresh],
                                [use_07_metric])

    Top level function that does the PASCAL VOC evaluation.

    detpath: Path to detections产生的txt文件，里面是一张图片的各个detection。
        detpath.format(classname) should produce the detection results file.
    annopath: Path to annotationsxml 文件与对应的图像相呼应。
        annopath.format(imagename) should be the xml annotations file.
    imagesetfile: Text file containing the list of images, one image per line.
    一个txt文件，里面是每个图片的地址，每行一个地址。
    classname: Category name (duh)   种类的名字，即类别。
    cachedir: Directory for caching the annotations 缓存标注的目录。
    [ovthresh]: Overlap threshold (default = 0.5) 重叠的多少大小。
    [use_07_metric]: Whether to use VOC07's 11 point AP computation
        (default False) 是否使用VOC07的11点AP计算。
    """
    # assumes detections are in detpath.format(classname)
    # assumes annotations are in annopath.format(imagename)
    # assumes imagesetfile is a text file with each line an image name
    # cachedir caches the annotations in a pickle file

    # first load gt
    if not os.path.isdir(cachedir):#判断cachedir是否存在，如不存在，创建cachedir
        os.mkdir(cachedir)
    cachefile = os.path.join(cachedir, 'annots.pkl')#添加文件annots.pkl
    # read list of images
    with open(imagesetfile, 'r') as f:
        lines = f.readlines()
    imagenames = [x.strip() for x in lines]#获取每一个图片的地址 ima = [...,'3152','3329',...]

    if not os.path.isfile(cachefile):
        # load annots
        recs = {}#字典
        for i, imagename in enumerate(imagenames):
            recs[imagename] = parse_rec(annopath.format(imagename))#赋值 例如{‘0085’：[xmin,ymin,xmax,ymax]}
            if i % 100 == 0:#每个一百，输出一次
                print 'Reading annotation for {:d}/{:d}'.format(
                    i + 1, len(imagenames))
        # save
        print 'Saving cached annotations to {:s}'.format(cachefile)
        with open(cachefile, 'w') as f:
            cPickle.dump(recs, f)#将recs字典存储到annots.pkl文件中
    else:
        # load
        with open(cachefile, 'r') as f:
            recs = cPickle.load(f)#从annots.pkl中读取recs字典

    # extract gt objects for this class 对每张图片的xml获取函数指定类的bbox等。
    class_recs = {}
    npos = 0
    for imagename in imagenames:
        R = [obj for obj in recs[imagename] if obj['name'] == classname]#获取每个文件中某种类别的物体
        bbox = np.array([x['bbox'] for x in R])#抽取bbox
        difficult = np.array([x['difficult'] for x in R]).astype(np.bool)#difficult基本都是0，并将其转为bool型
        det = [False] * len(R)#list中形参len(R)个false
        npos = npos + sum(~difficult)#自增，sum求得的值基本都是0
        class_recs[imagename] = {'bbox': bbox,
                                 'difficult': difficult,
                                 'det': det}#{[imagename] ：{'bbox': bbox, 'difficult': difficult,'det': det}}

    # read dets
    detfile = detpath.format(classname)
    with open(detfile, 'r') as f:
        lines = f.readlines()
    if any(lines) == 1:#判断lines是否全为空，如不都为空，返回true
        splitlines = [x.strip().split(' ') for x in lines]#删除空格，并分割 spl=[...,['3152'],['1600'],...]
        image_ids = [x[0] for x in splitlines]#图片的index
        confidence = np.array([float(x[1]) for x in splitlines])#类别置信度
        BB = np.array([[float(z) for z in x[2:]] for x in splitlines])#变为浮点型的bbox

        # sort by confidence
        sorted_ind = np.argsort(-confidence)#返回confidence从大到小的索引值
        sorted_scores = np.sort(-confidence)#将confidence降序排列
        BB = BB[sorted_ind, :]#重排bbox,由大概率到小概率
        image_ids = [image_ids[x] for x in sorted_ind]#对图片进行重排

        # go down dets and mark TPs（true positive） and FPs（false positive）
        nd = len(image_ids)#获取图像数量
        tp = np.zeros(nd)#ndx1的零矩阵
        fp = np.zeros(nd)
        for d in range(nd):
            R = class_recs[image_ids[d]]
            bb = BB[d, :].astype(float)
            ovmax = -np.inf
            BBGT = R['bbox'].astype(float)

            if BBGT.size > 0:
                # compute overlaps
                # intersection
                ixmin = np.maximum(BBGT[:, 0], bb[0])
                iymin = np.maximum(BBGT[:, 1], bb[1])
                ixmax = np.minimum(BBGT[:, 2], bb[2])
                iymax = np.minimum(BBGT[:, 3], bb[3])
                iw = np.maximum(ixmax - ixmin + 1., 0.)
                ih = np.maximum(iymax - iymin + 1., 0.)
                inters = iw * ih

                # union
                uni = ((bb[2] - bb[0] + 1.) * (bb[3] - bb[1] + 1.) +
                       (BBGT[:, 2] - BBGT[:, 0] + 1.) *
                       (BBGT[:, 3] - BBGT[:, 1] + 1.) - inters)

                overlaps = inters / uni
                ovmax = np.max(overlaps)#返回最大重叠率
                jmax = np.argmax(overlaps)#返回最大重合率的索引值

            if ovmax > ovthresh:
                if not R['difficult'][jmax]:
                    if not R['det'][jmax]:
                        tp[d] = 1. #正样本
                        R['det'][jmax] = 1
                    else:
                        fp[d] = 1. #负样本
            else:
                fp[d] = 1.

        # compute precision recall
        fp = np.cumsum(fp)#返回沿给定列的累积和
        tp = np.cumsum(tp)
        rec = tp / float(npos)#Recall=TP/(TP+FN)
        # avoid divide by zero in case the first detection matches a difficult
        # ground truth
        prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps)#Precision=TP/(TP+FP),
        ap = voc_ap(rec, prec, use_07_metric)
    else:
         rec = -1
         prec = -1
         ap = -1


    return rec, prec, ap