yolov3训练测试自己的数据集

1.测试Demo

 1.1.下载源码

git clone https://github.com/pjreddie/darknet  
cd darknet 

 1.2.修改Makefile文件

  更改参数：

GPU=1
CUDNN=1
OPENCV=1
NVCC=/usr/local/cuda/bin/nvcc

 1.3运行测试命令

./darknet detect cfg/yolov3.cfg yolov3.weights data/dog.jpg

  如果报错：
  ./darknet: error while loading shared libraries: libcudart.so.9.0: cannot open shared object file: No such file or directory
  可以运行以下命令解决：
  sudo ldconfig /usr/local/cuda/lib64

2.训练自己的voc格式数据集

 2.1准备voc训练数据集

  按下列文件夹结构，将训练数据集放到各个文件夹下面，生成4个训练、测试和验证txt文件列表
  VOCdevkit
  +VOC2007
  ++Annotations
  ++ImageSets
  +++Main
  ++JPEGImages
 Annotations中是所有的xml文件
 JPEGImages中是所有的训练图片
 Main中是2个txt文件:train.txt和val.txt(里面的图片名没有后缀），用“generate-train-txt.py”生成：

#coding=utf-8
import os  
import random 

root_path="/media/wyq/719ffd70-9553-244f-ac13-2d8a0a86e395/数据集/训练数据集+模型/for-rubbish/yolov3/VOCdevkit/VOC2007/JPEGImages"
folderlist = os.listdir(root_path)
folderlist.sort(reverse=False)
file_num=len(folderlist)
file_list=range(file_num)

trainval_percent = 1
train_percent = 0.8
txtsavepath = 'ImageSets/Main' 
ftrainval = open(txtsavepath+'/trainval.txt', 'w')
ftest = open(txtsavepath+'/test.txt', 'w')
ftrain = open(txtsavepath+'/train.txt', 'w')
fval = open(txtsavepath+'/val.txt', 'w')

for i in file_list:
    xmlfilepath = 'Annotations/' + folderlist[i]
    total_xml = os.listdir(xmlfilepath)
    total_xml.sort(reverse=False)

    num=len(total_xml)  
    list=range(num)  
    tv=int(num*trainval_percent)  
    tr=int(tv*train_percent)  
    trainval= random.sample(list,tv)
    train=random.sample(trainval,tr)

    for j  in list:
        name= folderlist[i] + '/' + total_xml[j][:-4]+'\n'  
        if j in trainval:  
            ftrainval.write(name)  
            if j in train:
                ftrain.write(name)  
            else:  
                fval.write(name)  
        else:  
            ftest.write(name)

ftrainval.close()  
ftrain.close()  
fval.close()  
ftest .close()

 

2.2修改并执行 voc_label.py

# voc_label.py
# 生成2007_train.txt 和 2007_test.txt，分别存放了训练集和测试集图片的路径。

import xml.etree.ElementTree as ET
import pickle
import os
from os import listdir, getcwd
from os.path import join

sets=[('2007', 'train'),('2007', 'val')] #[('2012', 'train'), ('2012', 'val'), ('2007', 'train'), ('2007', 'val'), ('2007', 'test')]

classes = ["pbag","butt","box","bottle","cup"] #["aeroplane", "bicycle", "bird", "boat", "bottle", "bus", "car", "cat", "chair", "cow", "diningtable", "dog", "horse", "motorbike", "person", "pottedplant", "sheep", "sofa", "train", "tvmonitor"]


def convert(size, box):
    dw = 1./(size[0])
    dh = 1./(size[1])
    x = (box[0] + box[1])/2.0 - 1
    y = (box[2] + box[3])/2.0 - 1
    w = box[1] - box[0]
    h = box[3] - box[2]
    x = x*dw
    w = w*dw
    y = y*dh
    h = h*dh
    return (x,y,w,h)

def convert_annotation(year, image_id):
    in_file = open('VOCdevkit/VOC%s/Annotations/%s.xml'%(year, image_id))
    out_file = open('VOCdevkit/VOC%s/labels/%s.txt'%(year, image_id), 'w')
    tree=ET.parse(in_file)
    root = tree.getroot()
    size = root.find('size')
    w = int(size.find('width').text)
    h = int(size.find('height').text)

    for obj in root.iter('object'):
        difficult = obj.find('difficult').text
        cls = obj.find('name').text
        if cls not in classes or int(difficult)==1:
            continue
        cls_id = classes.index(cls)
        xmlbox = obj.find('bndbox')
        b = (float(xmlbox.find('xmin').text), float(xmlbox.find('xmax').text), float(xmlbox.find('ymin').text), float(xmlbox.find('ymax').text))
        bb = convert((w,h), b)
        out_file.write(str(cls_id) + " " + " ".join([str(a) for a in bb]) + '\n')

wd = getcwd()

for year, image_set in sets:
    if not os.path.exists('VOCdevkit/VOC%s/labels/'%(year)):
        os.makedirs('VOCdevkit/VOC%s/labels/'%(year))
    image_ids = open('VOCdevkit/VOC%s/ImageSets/Main/%s.txt'%(year, image_set)).read().strip().split()
    list_file = open('%s_%s.txt'%(year, image_set), 'w')
    for image_id in image_ids:
        list_file.write('%s/VOCdevkit/VOC%s/JPEGImages/%s.jpg\n'%(wd, year, image_id))
        convert_annotation(year, image_id)
    list_file.close()

#os.system("cat 2007_train.txt 2007_val.txt > train.txt")
#os.system("cat 2007_train.txt 2007_val.txt 2007_test.txt > train.all.txt")

2.3修改配置文件

①下载预训练模型：

wget https://pjreddie.com/media/files/darknet53.conv.74

②修改 cfg/voc.data：

classes= 5
train  = /media/wyq/719ffd70-9553-244f-ac13-2d8a0a86e395/数据集/训练数据集+模型/for-rubbish/yolov3/2007_train.txt
valid  = /media/wyq/719ffd70-9553-244f-ac13-2d8a0a86e395/数据集/训练数据集+模型/for-rubbish/yolov3/2007_val.txt
names = data/voc.names
backup = backup

③修改data/voc.names

pbag
butt
box
bottle
cup

④修改cfg/yolov3-voc.cfg

[net]
# Testing
# batch=1
# subdivisions=1
# Training          # 训练模式，每次前向的图片数目 = batch/subdivisions
batch=64            # 一批训练样本的样本数量，每batch个样本更新一次参数
subdivisions=16     # batch/subdivisions作为一次性送入训练器的样本数量
                    # 如果内存不够大，将batch分割为subdivisions个子batch
                    # 上面这两个参数如果电脑内存小，则把batch改小一点，batch越大，训练效果越好
                    # subdivisions越大，可以减轻显卡压力
width=608 #640 #416
height=608 #480 #416
channels=3          # 以上三个参数为输入图像的参数信息 width和height影响网络对输入图像的分辨率，
                    # 从而影响precision，只可以设置成32的倍数
momentum=0.9        # DeepLearning1中最优化方法中的动量参数，这个值影响着梯度下降到最优值得速度
decay=0.0005        # 权重衰减正则项，防止过拟合
angle=0             # 通过旋转角度来生成更多训练样本
saturation = 1.5    # 通过调整饱和度来生成更多训练样本
exposure = 1.5      # 通过调整曝光量来生成更多训练样本
hue=.1              # 通过调整色调来生成更多训练样本

learning_rate=0.001 # 学习率决定着权值更新的速度，设置得太大会使结果超过最优值，太小会使下降速度过慢。
                    # 如果仅靠人为干预调整参数，需要不断修改学习率。刚开始训练时可以将学习率设置的高一点，
                    # 而一定轮数之后，将其减小
                    # 在训练过程中，一般根据训练轮数设置动态变化的学习率。
                    # 刚开始训练时：学习率以 0.01 ~ 0.001 为宜。
                    # 一定轮数过后：逐渐减缓。
                    # 接近训练结束：学习速率的衰减应该在100倍以上。
                    # 学习率的调整参考https://blog.csdn.net/qq_33485434/article/details/80452941
burn_in=1000        # 学习率控制的参数,在迭代次数小于burn_in时，其学习率的更新有一种方式，大于burn_in时，才采用policy的更新方式
max_batches = 50200 # 迭代次数,训练达到max_batches后停止学习
policy=steps        # 学习率策略，有policy：constant, steps, exp, poly, step, sig, RANDOM，constant等方式
steps=40000,45000   # 学习率变动步长,下面这两个参数steps和scale是设置学习率的变化，比如迭代到40000次时，学习率衰减十倍。
scales=.1,.1        # 学习率变动因子,45000次迭代时，学习率又会在前一个学习率的基础上衰减十倍



[convolutional]
batch_normalize=1   # BN
filters=32          # 输出特征图的数量
size=3              # 卷积核尺寸
stride=1            # 卷积核步长
pad=1               # 如果pad为0,padding由 padding参数指定
                    # 如果pad为1，padding大小为size/2，padding应该是对输入图像左边缘拓展的像素数量
activation=leaky    # 激活函数的类型

# Downsample

... ...

[convolutional]
size=1
stride=1
pad=1
filters=30           # 修改为3*（classes+5）即3*（5+5）=30
                     # 每一个[region/yolo]层前的最后一个卷积层中的 filters=num(yolo层个数)*(classes+5)
                     # 5的意义是5个坐标，论文中的tx,ty,tw,th,to
activation=linear

# 在yoloV2中yolo层叫region层
[yolo]
mask = 6,7,8
anchors =  20, 23,  32, 50,  57, 41,  54, 80,  96, 94,  73,147, 140,162, 178,250, 298,332
                     # 22, 19,  36, 39,  70, 37,  57, 65,  75,113, 111, 81, 141,145, 224,186, 307,284
                     # 10,13,  16,30,  33,23,  30,61,  62,45,  59,119,  116,90,  156,198,  373,326
                     # anchors是可以事先通过cmd指令计算出来的，是和图片数量，width,height以及cluster(应该就是下面的num的值，
                     # 即想要使用的anchors的数量)相关的预选框，可以手工挑选，也可以通过k means 从训练样本中学出
classes=5            # 类别
num=9
jitter=.3
ignore_thresh = .5
truth_thresh = 1
random=1             # 1，如果显存很小，将random设置为0，关闭多尺度训练
                     # random设置成1，可以增加检测精度precision

[route]
layers = -4
... ...

[convolutional]
size=1
stride=1
pad=1
filters=30                # 75,修改为3*（classes+5）即3*（5+5）=30
activation=linear

[yolo]
mask = 3,4,5
anchors =  20, 23,  32, 50,  57, 41,  54, 80,  96, 94,  73,147, 140,162, 178,250, 298,332
                          # 22, 19,  36, 39,  70, 37,  57, 65,  75,113, 111, 81, 141,145, 224,186, 307,284
                          # 10,13,  16,30,  33,23,  30,61,  62,45,  59,119,  116,90,  156,198,  373,326
classes=5                 # 类别
num=9                     # 每个grid cell预测几个box,和anchors的数量一致。当想要使用更多anchors时需要调大num，
                          # 且如果调大num后训练时Obj趋近0的话可以尝试调大object_scale
jitter=.3                 # 利用数据抖动产生更多数据，YOLOv2中使用的是crop，filp，以及net层的angle，flip是随机的，
                          # jitter就是crop的参数，tiny-yolo-voc.cfg中jitter=.3，就是在0~0.3中进行crop
ignore_thresh = .5        # 决定是否需要计算IOU误差的参数，大于thresh，IOU误差不会夹在cost function中
truth_thresh = 1
random=1                  # 1，如果显存很小，将random设置为0，关闭多尺度训练；
                          # 如果为1，每次迭代图片大小随机从320到608，步长为32，如果为0，每次训练大小与输入大小一致

[route]
layers = -4

... ...

[convolutional]
size=1
stride=1
pad=1
filters=30                # 75,修改为3*（classes+5）即3*（5+5）=30
activation=linear

[yolo]
mask = 0,1,2
anchors =  20, 23,  32, 50,  57, 41,  54, 80,  96, 94,  73,147, 140,162, 178,250, 298,332
                          # 22, 19,  36, 39,  70, 37,  57, 65,  75,113, 111, 81, 141,145, 224,186, 307,284
                          # 10,13,  16,30,  33,23,  30,61,  62,45,  59,119,  116,90,  156,198,  373,326
classes=5                 # 类别
num=9
jitter=.3                 # 数据扩充的抖动操作
ignore_thresh = .5        # 文章中的阈值1
truth_thresh = 1          # 文章中的阈值2
random=1                  #1，如果显存很小，将random设置为0，关闭多尺度训练；


#https://github.com/AlexeyAB/darknet (AlexeyAB版本yolov3）介绍的关键点：
#    图片样本的获取可以自己找，也可以下载相关数据集，coco,voc数据集中可能也含有相关能用的样本
#
#    Note: If during training you see nan values for avg (loss) field - then training goes wrong,
#    but if nan is in some other lines - then training goes well.
#
#    When should I stop training:
#    When you see that average loss 0.xxxxxx avg no longer decreases at many iterations then you should stop training.
#    Once training is stopped, you should take some of last .weights-files from darknet\build\darknet\x64\backup and choose the best of them.
#
#    Overfitting - is case when you can detect objects on images from training-dataset,
#    but can't detect objects on any others images. You should get weights from Early Stopping Point.
#
#    IoU (intersect of union) - average instersect of union of objects and detections for a certain threshold = 0.24
#
#    How to improve object detection:
#    Before training:
#    set flag random=1 in your .cfg-file - it will increase precision by training Yolo for different resolutions.
#    increase network resolution in your .cfg-file (height=608, width=608 or any value multiple of 32) - it will increase precision.
#    recalculate anchors for your dataset for width and height from cfg-file:
#    darknet.exe detector calc_anchors data/obj.data -num_of_clusters 9 -width 416 -height 416 then set the same 9 anchors in each of 3 [yolo]-layers in your cfg-file
#    设置锚点
#
#    desirable that your training dataset include images with objects at diffrent:
#    scales, rotations, lightings, from different sides, on different backgrounds
#    样本特点尽量多样化，亮度，旋转，背景，目标位置，尺寸
#
#    desirable that your training dataset include images with non-labeled objects that you do not want to detect - negative samples without bounded box (empty .txt files)
#    可以添加没有标注框的图片和其空的txt文件，作为negative数据
#
#    for training with a large number of objects in each image, add the parameter max=200 or higher value in the last layer [region] in your cfg-file
#
#    to speedup training (with decreasing detection accuracy) do Fine-Tuning instead of Transfer-Learning,
#    set param stopbackward=1 in one of the penultimate convolutional layers before the 1-st [yolo]-layer,
#    for example here: https://github.com/AlexeyAB/darknet/blob/0039fd26786ab5f71d5af725fc18b3f521e7acfd/cfg/yolov3.cfg#L598
#    可以在第一个[yolo]层之前的倒数第二个[convolutional]层末尾添加 stopbackward=1,以此提升训练速度
#
#    After training - for detection:
#    Increase network-resolution by set in your .cfg-file (height=608 and width=608) or (height=832 and width=832)
#    or (any value multiple of 32) - this increases the precision and makes it possible to detect small objects,
#    you do not need to train the network again, just use .weights-file already trained for 416x416 resolution.
#    即使在用416*416训练完之后，也可以在cfg文件中设置较大的width和height，增加网络对图像的分辨率，从而更可能检测出图像中的小目标，而不需要重新训练
#
#    if error Out of memory occurs then in .cfg-file you should increase subdivisions=16, 32 or 64
#    Out of memory的错误需要通过增大subdivisions来解决
#官网介绍中的关键点：
#    图片样本的获取可以自己找，也可以下载相关数据集，coco,voc数据集中可能也含有相关能用的样本
#
#    Note: If during training you see nan values for avg (loss) field - then training goes wrong,
#    but if nan is in some other lines - then training goes well.
#
#    When should I stop training:
#    When you see that average loss 0.xxxxxx avg no longer decreases at many iterations then you should stop training.
#    Once training is stopped, you should take some of last .weights-files from darknet\build\darknet\x64\backup and choose the best of them.
#
#    Overfitting - is case when you can detect objects on images from training-dataset,
#    but can't detect objects on any others images. You should get weights from Early Stopping Point.
#
#    IoU (intersect of union) - average instersect of union of objects and detections for a certain threshold = 0.24
#
#    How to improve object detection:
#    Before training:
#    set flag random=1 in your .cfg-file - it will increase precision by training Yolo for different resolutions.
#    increase network resolution in your .cfg-file (height=608, width=608 or any value multiple of 32) - it will increase precision.
#    recalculate anchors for your dataset for width and height from cfg-file:
#    darknet.exe detector calc_anchors data/obj.data -num_of_clusters 9 -width 416 -height 416 then set the same 9 anchors in each of 3 [yolo]-layers in your cfg-file
#    设置锚点
#
#    desirable that your training dataset include images with objects at diffrent:
#    scales, rotations, lightings, from different sides, on different backgrounds
#    样本特点尽量多样化，亮度，旋转，背景，目标位置，尺寸
#
#    desirable that your training dataset include images with non-labeled objects that you do not want to detect - negative samples without bounded box (empty .txt files)
#    可以添加没有标注框的图片和其空的txt文件，作为negative数据
#
#    for training with a large number of objects in each image, add the parameter max=200 or higher value in the last layer [region] in your cfg-file
#
#    to speedup training (with decreasing detection accuracy) do Fine-Tuning instead of Transfer-Learning,
#    set param stopbackward=1 in one of the penultimate convolutional layers before the 1-st [yolo]-layer,
#    for example here: https://github.com/AlexeyAB/darknet/blob/0039fd26786ab5f71d5af725fc18b3f521e7acfd/cfg/yolov3.cfg#L598
#    可以在第一个[yolo]层之前的倒数第二个[convolutional]层末尾添加 stopbackward=1,以此提升训练速度
#
#    After training - for detection:
#    Increase network-resolution by set in your .cfg-file (height=608 and width=608) or (height=832 and width=832)
#    or (any value multiple of 32) - this increases the precision and makes it possible to detect small objects,
#    you do not need to train the network again, just use .weights-file already trained for 416x416 resolution.
#    即使在用416*416训练完之后，也可以在cfg文件中设置较大的width和height，增加网络对图像的分辨率，从而更可能检测出图像中的小目标，而不需要重新训练
#
#    if error Out of memory occurs then in .cfg-file you should increase subdivisions=16, 32 or 64
#    Out of memory的错误需要通过增大subdivisions来解决

【更改输入图像大小，重新设置anchors】

 得下载AlexeyAB版本的yolov3（自带计算工具）：

https://github.com/AlexeyAB/darknet

通过相同的方式编译，再运行命令：

./darknet detector calc_anchors cfg/voc.data -num_of_clusters 9 -width 640 -height 480

 将生成的结果替换.cfg文件中的anchors

2.4开始训练

./darknet detector train cfg/voc.data cfg/yolov3-voc.cfg darknet53.conv.74 -gpus 0

3.训练过程中出现的参数解析

  终端输出：

Region 106 Avg IOU: 0.061216, Class: 0.626917, Obj: 0.302514, No Obj: 0.458097, .5R: 0.000000, .75R: 0.000000,  count: 4
Region 82 Avg IOU: -nan, Class: -nan, Obj: -nan, No Obj: 0.492722, .5R: -nan, .75R: -nan,  count: 0
Region 94 Avg IOU: 0.143258, Class: 0.711037, Obj: 0.514405, No Obj: 0.513470, .5R: 0.000000, .75R: 0.000000,  count: 2
Region 106 Avg IOU: 0.263153, Class: 0.434489, Obj: 0.432080, No Obj: 0.457346, .5R: 0.000000, .75R: 0.000000,  count: 2
4: 557.440735, 558.787842 avg, 0.000000 rate, 1.717108 seconds, 256 images
Loaded: 1.211514 seconds
...
Region 106 Avg IOU: 0.104959, Class: 0.519788, Obj: 0.193895, No Obj: 0.326036, .5R: 0.000000, .75R: 0.000000,  count: 3
Region 82 Avg IOU: 0.143481, Class: 0.373649, Obj: 0.561790, No Obj: 0.452239, .5R: 0.000000, .75R: 0.000000,  count: 2
Region 94 Avg IOU: 0.359399, Class: 0.912672, Obj: 0.276435, No Obj: 0.434352, .5R: 0.000000, .75R: 0.000000,  count: 1
Region 106 Avg IOU: 0.116004, Class: 0.368650, Obj: 0.245646, No Obj: 0.324594, .5R: 0.000000, .75R: 0.000000,  count: 2
Region 82 Avg IOU: 0.397991, Class: 0.361905, Obj: 0.791880, No Obj: 0.451690, .5R: 0.000000, .75R: 0.000000,  count: 1
Region 94 Avg IOU: 0.388091, Class: 0.398659, Obj: 0.498814, No Obj: 0.434391, .5R: 0.333333, .75R: 0.000000,  count: 3
Region 106 Avg IOU: 0.084905, Class: 0.563230, Obj: 0.281137, No Obj: 0.325844, .5R: 0.000000, .75R: 0.000000,  count: 3
Region 82 Avg IOU: -nan, Class: -nan, Obj: -nan, No Obj: 0.453356, .5R: -nan, .75R: -nan,  count: 0
Region 94 Avg IOU: 0.069556, Class: 0.379951, Obj: 0.705520, No Obj: 0.433259, .5R: 0.000000, .75R: 0.000000,  count: 2
Region 106 Avg IOU: 0.017326, Class: 0.713215, Obj: 0.136531, No Obj: 0.325978, .5R: 0.000000, .75R: 0.000000,  count: 3
100: 338.473236, 562.332581 avg, 0.000000 rate, 1.687635 seconds, 6400 images
Saving weights to backup/yolov3-voc.backup
Saving weights to backup/yolov3-voc_100.weights
Resizing
416
Loaded: 0.878086 seconds

• Region Avg IOU： 表示在当前subdivision内的图片的平均IOU，代表预测的矩形框和真实目标的交集与并集之比
• Class： 标注物体分类的正确率，期望该值趋近于1
• Obj： 越接近1越好
• No Obj： 期望该值越来越小，但不为零
• Avg Recall： 是在recall/count中定义的，是当前模型在所有subdivision图片中检测出的正样本与实际的正样本的比值。在本例中，只有八分之一的正样本被正确的检测到。（和最开始初定的阈值有关系）
• count ： count后的值是所有的当前subdivision图片（本例中一共8张）中包含正样本的图片的数量。在输出log中的其他行中，可以看到其他subdivision也有的只含有6或7个正样本，说明在subdivision中含有不含检测对象的图片

  ---

• 100             ： 迭代次数
• 338.473236      ： 总体的Loss（损失）
• 562.332581 avg  ： 是平均Loss，这个数值应该越低越好，一般来说，一旦这个数值低于0.060730 avg就可以终止训练了
• 0.000000 rate   ： 代表当前的学习率，是在.cfg文件中定义的
• 1.687635 seconds： 表示当前批次训练花费的总时间
• 6400 images     ： 表示到目前为止，参与训练的图片的总量

  ---

   

4.训练后的测试

  (yolov3-voc.cfg)文件中batch和subdivisions两项必须为1

4.1测试图片的命令

./darknet detector test cfg/voc.data cfg/yolov3-voc.cfg backup/yolov3-voc_10000.weights /home/wyq/Documents/myCode/build-ssd-Desktop-Debug/635.png

4.2测试并保存视频的命令

./darknet detector demo cfg/my-voc.data cfg/my-yolov3-voc.cfg backup/my-yolov3-voc_10000.weights -thresh 0.4 -ext_output /media/wyq/719ffd70-9553-244f-ac13-2d8a0a86e395/数据集/原始数据集/rubbish/rubbish-dataset-20190304-nianhui/video0/006.mp4 -out_filename /...输出视频.mp4/.avi

  其中，-ext_output 是为了在终端输出bbox，thresh可以根据检测的score控制画面上的bbox(是否)显示（此处只显示score>0.4的检测结果），增加-out_filename可以将检测结果保存成视频。

4.3其他命令

  1) 批量测试：
    ./darknet detector test cfg/voc.data cfg/yolo-voc.cfg ./svt/backup/yolo-voc_final.weights
    /*不现实评价指标，输入图片路径，只显示框好后的图片和类别、置信率*/
    Enter Image Path：后面输入你的txt文件路径（你准备好的所有测试图片的路径全部存放在一个txt文件里），你可以复制voc.data文件里的valid后面的路径

  2) 生成预测结果 :
    ./darknet detector valid cfg/voc.data cfg/yolo-voc.cfg backup/yolo-voc_final.weights
    /*在终端只返回用时，在./results/comp4_det_test_[类名].txt里保存测试结果*/

  3) 生成预测结果 :
    ./darknet detector recall cfg/voc.data cfg/yolo-voc.cfg backup/yolo-voc_final.weights（这个命令需修改dectector.c文件）
    /*依次ID:图片序号从0开始，correct:累计检测正确的总框数，total:累计的总ground truth数，RPs/Img: 累计的总proposals/已检测图片数，IOU，Recall: correct / total，proposals:累计的总框数，Precision: correct / proposals*/

    ***detector.c修改（example文件夹下）
        validate_detector_recall函数定义和调用改为：
            void validate_detector_recall(char *datacfg, char *cfgfile, char *weightfile)
            validate_detector_recall(datacfg, cfg, weights);
        validate_detector_recall内的plist和paths的如下初始化代码：
            list *plist = get_paths("data/voc.2007.test");
            char **paths = (char **)list_to_array(plist);
        修改为：
            list *options = read_data_cfg(datacfg);
            char *valid_images = option_find_str(options, "valid", "data/train.list");
            list *plist = get_paths(valid_images);
            char **paths = (char **)list_to_array(plist);
    上述修改完之后务必记住要在darknet下重新make一下就可以进行recall命令了。