【Tensorflow】Object Detection API学习

操作系统：Centos7.4

1. 准备工作

1.1 install protobuf

下载protobuf-2.6.1

1.2. install tensorflow

For CPU

pip install tensorflow

For GPU

pip install tensorflow-gpu

1.3. 配置tensorflow/models

参考：https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/installation.md

Tensorflow Object Detection API 依赖以下库:

  • Protobuf 2.6
  • Pillow 1.0
  • lxml
  • tf Slim (which is included in the "tensorflow/models/research/" checkout)
  • Jupyter notebook
  • Matplotlib
  • Tensorflow

具体步骤如下：

  • 下载TensorFlow Models

git clone https://github.com/tensorflow/models.git

  • 编译protobuf（在object_detection/protos/下生成若干py文件）

#From tensorflow/models/research
cd models/research
protoc object_detection/protos/*.proto --python_out=.

  • 添加PYTHONPATH

From tensorflow/models/research
export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim
或者在/root/.bashrc中加入：export PYTHONPATH=$PYTHONPATH:/home/tensorflow_install/models/research:/home/tensorflow_install/models/research/slim

  • 验证

#From tensorflow/models/research
python object_detection/builders/model_builder_test.py

验证前先确保setup.py编译安装，一些依赖库是否安装，我在验证中遇到很多错误，安装以下依赖库解决：

  • python setup.py build
  • python setup.py install
  • pip install matplotlib
  • yum install -y tkinter
  • pip install image
  • pip install pillow

1.4. 准备数据

参考：https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/preparing_inputs.md

以PASCAL VOC 2012为例：

  • 下载并解压

#From tensorflow/models/research/object_detection
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar
tar -xvf VOCtrainval_11-May-2012.tar

  • 生成TFRecord（得到pascal_train.record和pascal_val.record）

#From tensorflow/models/research
mkdir object_detection/VOC2012
python object_detection/dataset_tools/create_pascal_tf_record.py
--label_map_path=object_detection/data/pascal_label_map.pbtxt
--data_dir=VOCdevkit --year=VOC2012 --set=train
--output_path=object_detection/VOC2012/pascal_train.record
python object_detection/dataset_tools/create_pascal_tf_record.py
--label_map_path=object_detection/data/pascal_label_map.pbtxt
--data_dir=VOCdevkit --year=VOC2012 --set=val
--output_path=object_detection/VOC2012/pascal_val.record

如果需要用自己的数据，则参考create_pascal_tf_record.py编写处理数据生成TFRecord的脚本。（参考：https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/using_your_own_dataset.md）在下一篇文章介绍。

1.5. （可选）下载模型

官方提供了不少与训练模型（https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md），这里以ssd_mobilenet_v1_coco以例：

#From tensorflow/models/research/object_detection
wget http://download.tensorflow.org/models/object_detection/ssd_mobilenet_v1_coco_2017_11_17.tar.gz
tar -xzvf ssd_mobilenet_v1_coco_2017_11_17.tar.gz

2. 训练

如果使用现有模型进行预测则不需要训练。
文件结构：

models
├── research
│   ├── object_detection
│   │   ├── VOC2012
│   │   │   ├── ssd_mobilenet_train_logs
│   │   │   ├── ssd_mobilenet_val_logs
│   │   │   ├── ssd_mobilenet_v1_voc2012.config
│   │   │   ├── pascal_label_map.pbtxt
│   │   │   ├── pascal_train.record
│   │   │   └── pascal_val.record
│   │   ├── infer.py
│   │   └── create_pascal_tf_record.py
│   ├── eval_voc2012.sh
│   └── train_voc2012.sh

2.1. 配置

参考：https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/configuring_jobs.md

配置分为5个部分：

  • model config：定义了什么类型的模型将被训练（如元架构，特征提取器）
  • train_config：决定应该使用哪些参数来训练模型参数（如SGD参数，输入预处理和特征提取器初始化值）
  • eval_config
  • train_input_config：定义了模型应该训练的数据集
  • eval_input_config：定义了模型将被评估的数据集

在object_detection/samples/model_configs文件夹中提供了示例模型配置。这些配置文件的内容可以粘贴到框架配置的模型区域。应该注意的是，num_classes字段应该改为适合正在训练的数据集的值。

这里使用ssd_mobilenet:

#From tensorflow/models/research

  • cp object_detection/samples/configs/ssd_mobilenet_v1_pets.config object_detection/VOC2012/ssd_mobilenet_v1_voc2012.config
  • 修改9行为
    num_classes:20
  • 修改158行为
    fine_tune_checkpoint: "object_detection/ssd_mobilenet_v1_coco_2017_11_17/model.ckpt"
  • 修改177行为
    input_path: "object_detection/VOC2012/pascal_train.record"
  • 修改179行和193行为：
    label_map_path: "object_detection/data/pascal_label_map.pbtxt"
  • 修改191行为：input_path: "object_detection/VOC2012/pascal_val.record"

2.2. 训练

新建research/train_voc2012.sh，添加以下内容：

python object_detection/train.py \
    --logtostderr \
    --pipeline_config_path=object_detection/VOC2012/ssd_mobilenet_v1_voc2012.config \
    --train_dir=object_detection/VOC2012/ssd_mobilenet_train_logs \
    2>&1 | tee object_detection/VOC2012/ssd_mobilenet_train_logs.txt &

执行以下命令即可训练：

./train_voc2012.sh

2.3. 验证

可一边训练一边验证，注意使用其它的GPU或合理分配显存。
新建tensorflow/models/eval_voc2012.sh，内容以下：

    --logtostderr \
    --pipeline_config_path=object_detection/VOC2012/ssd_mobilenet_v1_voc2012.config \
    --checkpoint_dir=object_detection/VOC2012/ssd_mobilenet_train_logs \
    --eval_dir=object_detection/VOC2012/ssd_mobilenet_val_logs &

进入tensorflow/models/research，运行CUDA_VISIBLE_DEVICES="1" （这里就不需要设置了，我们使用CPU，不用GPU）./train_voc2012.sh即可验证。

3. 测试

3.1. 导出模型

训练完成后得到一些checkpoint文件在ssd_mobilenet_train_logs中，如：

  • graph.pbtxt
  • model.ckpt-200000.data-00000-of-00001
  • model.ckpt-200000.info
  • model.ckpt-200000.meta
    其中meta保存了graph和metadata，ckpt保存了网络的weights。
    而进行预测时只需模型和权重，不需要metadata，故可使用官方提供的脚本生成推导图：
python object_detection/export_inference_graph.py \
    --input_type image_tensor \
    --pipeline_config_path object_detection/VOC2012/ssd_mobilenet_v1_voc2012.config \
    --trained_checkpoint_prefix object_detection/VOC2012/ssd_mobilenet_train_logs/model.ckpt-200000 \
    --output_directory object_detection/VOC2012

3.2. 测试图片

  • 运行object_detection_tutorial.ipynb并修改其中的各种路径即可。
  • 或自写编译inference脚本，如tensorflow/models/research/object_detection/infer.py
import numpy as np
import os
import six.moves.urllib as urllib
import sys
import tarfile
import tensorflow as tf
import zipfile

from collections import defaultdict
from io import StringIO
from matplotlib import pyplot as plt
from PIL import Image

## This is needed to display the images.
#%matplotlib inline

# This is needed since the notebook is stored in the object_detection folder.
sys.path.append("..")

from utils import label_map_util

from utils import visualization_utils as vis_util

# What model to download.
MODEL_NAME = 'ssd_mobilenet_v1_coco_2017_11_17'
MODEL_FILE = MODEL_NAME + '.tar.gz'
DOWNLOAD_BASE = 'http://download.tensorflow.org/models/object_detection/'

# Path to frozen detection graph. This is the actual model that is used for the object detection.
PATH_TO_CKPT = MODEL_NAME + '/frozen_inference_graph.pb'

# List of the strings that is used to add correct label for each box.
PATH_TO_LABELS = os.path.join('data', 'mscoco_label_map.pbtxt')

NUM_CLASSES = 90

#download model
opener = urllib.request.URLopener()
opener.retrieve(DOWNLOAD_BASE + MODEL_FILE, MODEL_FILE)
#tar_file = tarfile.open(MODEL_FILE)
#for file in tar_file.getmembers():
#  file_name = os.path.basename(file.name)
#  if 'frozen_inference_graph.pb' in file_name:
#    tar_file.extract(file, os.getcwd())

#Load a (frozen) Tensorflow model into memory.
detection_graph = tf.Graph()
with detection_graph.as_default():
  od_graph_def = tf.GraphDef()
  with tf.gfile.GFile(PATH_TO_CKPT, 'rb') as fid:
    serialized_graph = fid.read()
    od_graph_def.ParseFromString(serialized_graph)
    tf.import_graph_def(od_graph_def, name='')
#Loading label map
label_map = label_map_util.load_labelmap(PATH_TO_LABELS)
categories = label_map_util.convert_label_map_to_categories(label_map, max_num_classes=NUM_CLASSES, use_display_name=True)
category_index = label_map_util.create_category_index(categories)
#Helper code
def load_image_into_numpy_array(image):
  (im_width, im_height) = image.size
  return np.array(image.getdata()).reshape(
      (im_height, im_width, 3)).astype(np.uint8)


# For the sake of simplicity we will use only 2 images:
# image1.jpg
# image2.jpg
# If you want to test the code with your images, just add path to the images to the TEST_IMAGE_PATHS.
PATH_TO_TEST_IMAGES_DIR = 'test_images'
#TEST_IMAGE_PATHS = [ os.path.join(PATH_TO_TEST_IMAGES_DIR, 'image{}.jpg'.format(i)) for i in range(1, 3) ]
TEST_IMAGE = sys.argv[1]
print 'the test image is:', TEST_IMAGE

# Size, in inches, of the output images.
IMAGE_SIZE = (12, 8)

with detection_graph.as_default():
  with tf.Session(graph=detection_graph) as sess:
    # Definite input and output Tensors for detection_graph
    image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
    # Each box represents a part of the image where a particular object was detected.
    detection_boxes = detection_graph.get_tensor_by_name('detection_boxes:0')
    # Each score represent how level of confidence for each of the objects.
    # Score is shown on the result image, together with the class label.
    detection_scores = detection_graph.get_tensor_by_name('detection_scores:0')
    detection_classes = detection_graph.get_tensor_by_name('detection_classes:0')
    num_detections = detection_graph.get_tensor_by_name('num_detections:0')
    #for image_path in TEST_IMAGE_PATHS:
    image = Image.open(TEST_IMAGE)
    # the array based representation of the image will be used later in order to prepare the
    # result image with boxes and labels on it.
    image_np = load_image_into_numpy_array(image)
    # Expand dimensions since the model expects images to have shape: [1, None, None, 3]
    image_np_expanded = np.expand_dims(image_np, axis=0)
    image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
    # Each box represents a part of the image where a particular object was detected.
    boxes = detection_graph.get_tensor_by_name('detection_boxes:0')
    # Each score represent how level of confidence for each of the objects.
    # Score is shown on the result image, together with the class label.
    scores = detection_graph.get_tensor_by_name('detection_scores:0')
    classes = detection_graph.get_tensor_by_name('detection_classes:0')
    num_detections = detection_graph.get_tensor_by_name('num_detections:0')
    # Actual detection.
    (boxes, scores, classes, num_detections) = sess.run(
        [boxes, scores, classes, num_detections],
        feed_dict={image_tensor: image_np_expanded})
    # Visualization of the results of a detection.
    vis_util.visualize_boxes_and_labels_on_image_array(
        image_np,
        np.squeeze(boxes),
        np.squeeze(classes).astype(np.int32),
        np.squeeze(scores),
        category_index,
        use_normalized_coordinates=True,
        line_thickness=8)

    print(scores)  
    print(classes)  
    print(category_index) 

    count = 0
    for i in range(100):
        if scores is None or final_score[i] > 0.5:
            count = count + 1
    print 'the count of objects is: ', count  

    plt.figure(figsize=IMAGE_SIZE)
    plt.imshow(image_np)
    plt.show()

其中，这段代码是统计识别的物体个数：

count = 0
    for i in range(100):
        if scores is None or final_score[i] > 0.5:
            count = count + 1
    print 'the count of objects is: ', count

运行 infer.py test_images/image2.jpg，效果如图：

源码安装tensorflow
git clone --recurse-submodules https://github.com/tensorflow/tensorflow
wget https://copr.fedorainfracloud.org/coprs/vbatts/bazel/repo/epel-7/vbatts-bazel-epel-7.repo
cp vbatts-bazel-epel-7.repo /etc/yum.repos.d/
yum install -y bazel
yum install -y python-numpy swig python-dev python-wheel

cd tensorflow
./configure

bazel build -c opt --config=cuda //tensorflow/tools/pip_package:build_pip_package

最后编辑于 ：
©著作权归作者所有,转载或内容合作请联系作者
  • 人面猴
    序言：七十年代末，一起剥皮案震惊了整个滨河市，随后出现的几起案子，更是在滨河造成了极大的恐慌，老刑警刘岩，带你破解...
    沈念sama阅读 204,684评论 6赞 478
  • 死咒
    序言：滨河连续发生了三起死亡事件，死亡现场离奇诡异，居然都是意外死亡，警方通过查阅死者的电脑和手机，发现死者居然都...
    沈念sama阅读 87,143评论 2赞 381
  • 救了他两次的神仙让他今天三更去死
    文/潘晓璐 我一进店门，熙熙楼的掌柜王于贵愁眉苦脸地迎上来，“玉大人，你说我怎么就摊上这事。” “怎么了？”我有些...
    开封第一讲书人阅读 151,214评论 0赞 337
  • 道士缉凶录：失踪的卖姜人
    文/不坏的土叔 我叫张陵，是天一观的道长。 经常有香客问我，道长，这世上最难降的妖魔是什么？ 我笑而不...
    开封第一讲书人阅读 54,788评论 1赞 277
  • 港岛之恋（遗憾婚礼）
    正文 为了忘掉前任，我火速办了婚礼，结果婚礼上，老公的妹妹穿的比我还像新娘。我一直安慰自己，他们只是感情好，可当我...
    茶点故事阅读 63,796评论 5赞 368
  • 恶毒庶女顶嫁案：这布局不是一般人想出来的
    文/花漫 我一把揭开白布。 她就那样静静地躺着，像睡着了一般。 火红的嫁衣衬着肌肤如雪。 梳的纹丝不乱的头发上，一...
    开封第一讲书人阅读 48,665评论 1赞 281
  • 城市分裂传说
    那天，我揣着相机与录音，去河边找鬼。 笑死，一个胖子当着我的面吹牛，可吹牛的内容都是我干的。 我是一名探鬼主播，决...
    沈念sama阅读 38,027评论 3赞 399
  • 双鸳鸯连环套：你想象不到人心有多黑
    文/苍兰香墨 我猛地睁开眼，长吁一口气：“原来是场噩梦啊……” “哼！你这毒妇竟也来了？” 一声冷哼从身侧响起，我...
    开封第一讲书人阅读 36,679评论 0赞 258
  • 万荣杀人案实录
    序言：老挝万荣一对情侣失踪，失踪者是张志新（化名）和其女友刘颖，没想到半个月后，有当地人在树林里发现了一具尸体，经...
    沈念sama阅读 41,346评论 1赞 299
  • 护林员之死
    正文 独居荒郊野岭守林人离奇死亡，尸身上长有42处带血的脓包…… 初始之章·张勋 以下内容为张勋视角 年9月15日...
    茶点故事阅读 35,664评论 2赞 321
  • 白月光启示录
    正文 我和宋清朗相恋三年，在试婚纱的时候发现自己被绿了。 大学时的朋友给我发了我未婚夫和他白月光在一起吃饭的照片。...
    茶点故事阅读 37,766评论 1赞 331
  • 活死人
    序言：一个原本活蹦乱跳的男人离奇死亡，死状恐怖，灵堂内的尸体忽然破棺而出，到底是诈尸还是另有隐情，我是刑警宁泽，带...
    沈念sama阅读 33,412评论 4赞 321
  • 日本核电站爆炸内幕
    正文 年R本政府宣布，位于F岛的核电站，受9级特大地震影响，放射性物质发生泄漏。R本人自食恶果不足惜，却给世界环境...
    茶点故事阅读 39,015评论 3赞 307
  • 男人毒药：我在死后第九天来索命
    文/蒙蒙 一、第九天 我趴在偏房一处隐蔽的房顶上张望。 院中可真热闹，春花似锦、人声如沸。这庄子的主人今日做“春日...
    开封第一讲书人阅读 29,974评论 0赞 19
  • 一桩弑父案，背后竟有这般阴谋
    文/苍兰香墨 我抬头看了看天上的太阳。三九已至，却和暖如春，着一层夹袄步出监牢的瞬间，已是汗流浃背。 一阵脚步声响...
    开封第一讲书人阅读 31,203评论 1赞 260
  • 情欲美人皮
    我被黑心中介骗来泰国打工， 没想到刚下飞机就差点儿被人妖公主榨干…… 1. 我叫王不留，地道东北人。 一个月前我还...
    沈念sama阅读 45,073评论 2赞 350
  • 代替公主和亲
    正文 我出身青楼，却偏偏与公主长得像，于是被迫代替她去往敌国和亲。 传闻我的和亲对象是个残疾皇子，可洞房花烛夜当晚...
    茶点故事阅读 42,501评论 2赞 343