TFFRCNN修改自己的样本

最新推荐文章于 2018-09-02 16:58:30 发布

yuexin2

最新推荐文章于 2018-09-02 16:58:30 发布

阅读量2.1k

点赞数 1

分类专栏：机器学习

本文链接：https://blog.csdn.net/yuexin2/article/details/75025720

版权

机器学习专栏收录该内容

2 篇文章 0 订阅

订阅专栏

下面修改自己的样本

1. 这里使用行人的样本，路径为 /home/china409/Myprogram/PythonProgram/yx/TFFRCNN-master/human_data

2. 建立软链接

ln -s/home/china409/Myprogram/PythonProgram/yx/TFFRCNN-master/human_data human2017

并将该链接copy到data目录下

3. train_net.py文件中设置自己数据库的名称，这里我自己设置的叫做‘my_2017_human’

parser.add_argument('--imdb',dest='imdb_name',

help='dataset to train on',

default='my_2017_human', type=str)

4. 修改文件：/home/china409/Myprogram/PythonProgram/yx/TFFRCNN-master/lib/datasets/factory.py

添加如下代码：

#设置自己的数据

for mydbname in['my_2017_human']:

name = 'my_2017_human'

__sets[name] = (lambdasplit=name,year='2017':

pascal_voc(split, year))

5. 修改配置文件config.py

__C.NCLASSES = 7 (此处为训练样本特征的个数加一，本文样本为6个特征)

6. 修改文件：/home/china409/Myprogram/PythonProgram/yx/TFFRCNN-master/lib/datasets/pascal_voc.py。该文件中修改的较多

(1) 首先修改初始化数据库语句：

(2) 设置默认路径：

def _get_default_path(self):

"""

Return thedefault path where PASCAL VOC is expected to be installed.

"""

# return os.path.join(cfg.DATA_DIR,'VOCdevkit' + self._year)

returnos.path.join(cfg.DATA_DIR,

'human_data') # 设置软链接的路径-----/home/china409/Myprogram/PythonProgram/yx/TFFRCNN-master-human/data/human_data

(3) 第二设置训练集路径：

(4) 设置图片集合文件：

def _load_image_set_index(self):

"""

Load theindexes listed in this dataset's image set file.

"""

# Examplepath to image set file:

#self._devkit_path + /VOCdevkit2007/VOC2007/ImageSets/Main/val.txt

#image_set_file = os.path.join(self._data_path, 'ImageSets', 'Main',

# self._image_set+ '.txt')

image_set_file = os.path.join(self._data_path,'IMAGES_SET',

'walker.txt')#设置图片集合文件

assert os.path.exists(image_set_file), \

'Path doesnot exist: {}'.format(image_set_file)

with open(image_set_file)as f:

image_index = [x.strip() for xin f.readlines()]

return image_index

(5) 设置训练图片路径：

def image_path_from_index(self, index):

"""

Constructan image path from the image's "index" identifier.

"""

image_path = os.path.join(self._data_path,'IMAGES_TRAIN',

index + self._image_ext)

assert os.path.exists(image_path), \

'Path doesnot exist: {}'.format(image_path)

return image_path

(6) 设置样本中检测特征的内容

# self._classes = ('__background__',# always index 0

# 'aeroplane', 'bicycle','bird', 'boat',

# 'bottle', 'bus', 'car','cat', 'chair',

# 'cow', 'diningtable', 'dog', 'horse',

# 'motorbike', 'person','pottedplant',

# 'sheep', 'sofa', 'train','tvmonitor')

self._classes = ('__background__', # alwaysindex 0

'head','top', 'down','shoes','hat','bag')

(7) 修改_load_pascal_annotation方法，以适合自己样本的xml文件：

def _load_pascal_annotation(self, index):
    """
    Load imageand bounding boxes info from XML file in the PASCAL VOC
    format.
    该函数针对xml文件进行解析。
    """
    filename = os.path.join(self._data_path, 'ANNOTATIONS_TRAIN',index + '.xml')
    tree = ET.parse(filename)
    objs = tree.findall('subcomponent')#检测的事物标签
    # if notself.config['use_diff']:
    #     # Exclude the samples labeled as difficult
    #     non_diff_objs = [
    #         obj for obj in objs ifint(obj.find('difficult').text) == 0]
    #     # if len(non_diff_objs) != len(objs):
    #     #    print 'Removed {} difficult objects'.format(
    #     #        len(objs) - len(non_diff_objs))
    #     objs = non_diff_objs
    num_objs = len(objs)
    boxes = np.zeros((num_objs, 4), dtype=np.uint16)
    gt_classes = np.zeros((num_objs), dtype=np.int32)
    overlaps = np.zeros((num_objs, self.num_classes), dtype=np.float32)
    #"Seg" area for pascal is just the box area
    seg_areas= np.zeros((num_objs), dtype=np.float32)
    ishards = np.zeros((num_objs), dtype=np.int32)
    # Loadobject bounding boxes into a data frame.
    offset = 0;
    for ixtemp, obj in enumerate(objs):
        ix = ixtemp + offset;
        bbox = obj.find('bndbox')
        # Makepixel indexes 0-based
        if type(bbox) is type(None) :#没有bndbox节点时为shoes结构或者bag结构，有变化
           # printET.tostring(obj)
            otherStructure = obj.find('name');
            if otherStructure.text == 'shoes':
                #shoes
                if obj.find('xmin_l').text == 'NULL':
                    continue
                x1 = float(obj.find('xmin_l').text) - 1
                y1 = float(obj.find('ymin_l').text) - 1
                x2 = float(obj.find('xmax_l').text) - 1
                y2 = float(obj.find('ymax_l').text) - 1
                # diffc =obj.find('difficult')
                # difficult= 0 if diffc == None else int(diffc.text)
                #ishards[ix] = difficult
                #
                # cls =self._class_to_ind[obj.find('name').text.lower().strip()]
                # boxes[ix,:] = [x1, y1, x2, y2]
                #gt_classes[ix] = cls
                #overlaps[ix, cls] = 1.0
                #seg_areas[ix] = (x2 - x1 + 1) * (y2 - y1 + 1)
                #
                # offset+=1;
                # ix =ixtemp + offset;
                # x1 =float(obj.find('xmin_r').text) - 1
                # y1 =float(obj.find('ymin_r').text) - 1
                # x2 = float(obj.find('xmax_r').text)- 1
                # y2 =float(obj.find('ymax_r').text) - 1
            elif otherStructure.text == 'bag':
                #bag
                bbbox = obj.find('id_1')
                bbox = bbbox.find('bndbox')
                if bbox.find('xmin').text != 'NULL':
                    x1 = float(bbox.find('xmin').text) - 1
                    y1 = float(bbox.find('ymin').text) - 1
                    x2 = float(bbox.find('xmax').text) - 1
                    y2 = float(bbox.find('ymax').text) - 1
            else:
                continue
        else:
            if bbox.find('xmin').text == 'NULL':#在有bndbox节点时排除属性为NULL的节点
                continue
            else:
                x1 = float(bbox.find('xmin').text) - 1
                y1 = float(bbox.find('ymin').text) - 1
                x2 = float(bbox.find('xmax').text) - 1
                y2 = float(bbox.find('ymax').text) - 1
        #排除小于1的情况
        if x1 < 0:
            x1 = 0
        if y1 < 0:
            y1 = 0
        if x2 < 0:
            x2 = 0
        if y2 < 0:
            y2 = 0
        diffc = obj.find('difficult')
        difficult = 0 if diffc == None else int(diffc.text)
        ishards[ix] = difficult
        cls = self._class_to_ind[obj.find('name').text.lower().strip()]
        boxes[ix, :] = [x1, y1, x2, y2]
        gt_classes[ix] = cls
        overlaps[ix, cls] = 1.0
        seg_areas[ix] = (x2 - x1 + 1) * (y2 - y1 + 1)
    overlaps = scipy.sparse.csr_matrix(overlaps)
    return {'boxes' : boxes,
            'gt_classes': gt_classes,
            'gt_ishard': ishards,
            'gt_overlaps': overlaps,
            'flipped' : False,
           'seg_areas' :seg_areas}

7. 修改文件：/home/china409/Myprogram/PythonProgram/yx/TFFRCNN-master/experiments/cfgs/faster_rcnn_end2end.yml

可以拷贝后修改：

/home/china409/Myprogram/PythonProgram/yx/TFFRCNN-master/experiments/cfgs/faster_rcnn_end2end_human.yml

主要修改前两个路径以及特征的个数。

8. 修改train_net.py文件：

parser.add_argument('--cfg',dest='cfg_file',

help='optional config file',

default='/home/china409/Myprogram/PythonProgram/yx/TFFRCNN-master-human/experiments/cfgs/faster_rcnn_end2end_walker.yml',type=str)

9. 在运行前删除缓存文件

10. 训练完成后修改demo.py文件。

修改为自己模型的特征

# CLASSES = ('__background__',

# 'aeroplane', 'bicycle', 'bird','boat',

# 'bottle', 'bus', 'car', 'cat','chair',

# 'cow', 'diningtable', 'dog','horse',

# 'motorbike', 'person', 'pottedplant',

# 'sheep', 'sofa', 'train','tvmonitor')

CLASSES = ('__background__','head','top', 'down','shoes','hat','bag')

修改模型路径:

parser.add_argument('--model',dest='model',help='Model path',

default='/home/china409/Myprogram/PythonProgram/yx/TFFRCNN-master/output/walker/my_2017_human')

11. 替换/home/china409/Myprogram/PythonProgram/yx/TFFRCNN-master/data/demo中的测试集。运行demo.py进行测试

yuexin2

关注

1
点赞
踩
2

收藏

觉得还不错? 一键收藏
2
评论
TFFRCNN修改自己的样本

下面修改自己的样本1. 这里使用行人的样本，路径为 /home/china409/Myprogram/PythonProgram/yx/TFFRCNN-master/human_data 2. 建立软链接ln -s/home/china409/Myprogram/PythonProgram/yx/TFFRCNN-master/human_data human20
复制链接

扫一扫