Object detection: object_ Detection API +ssd target detection model

Environmental Science

window10, anaconda Python 3.5.6, tensorflow 1.8.0

tensorflow1.4later model download ：https://download.csdn.net/download/weixin_42748604/12647262

install object_detection API 

Refer to the official installation documentation

This document uses WIN10 System , Some changes have been made in some places , Please refer to .

Install dependency packages

pip install --user Cython
pip install --user contextlib2
pip install --user pillow
pip install --user lxml
pip install --user jupyter
pip install --user matplotlib
pip install --user tf_slim

install COCO API

git This installation COCO

This method requires the computer to install git

pip install git+https://github.com/philferriere/cocoapi.git#subdirectory=PythonAPI

python install COCO

git clone https://github.com/pdollar/coco.git

cd coco/PythonAPI

python setup.py build_ext --inplace

python setup.py build_ext install

test COCO API Installation

Get into python,import pycocotools See if there's a mistake .

compile Protobuf

Google official use proto, It needs to be compiled as .py file

stay models-master/research/ Execute under directory ：

for /f %i in ('dir /b object_detection\protos\*.proto') do protoc object_detection\protos\%i --python_out=.

Add environment variables

stay models-master\research Execute under directory ：

SET PYTHONPATH=%cd%;%cd%\slim

Run setup.py

stay models\research Run under folder ：

python setup.py build

python setup.py install

test API

Use Jupyter notebook open models-master/research/object_detection Under the object_detection_tutorial.ipynb file ,

It means that API Environment installed successfully .（ notes ：Jupyter notebook To run in the same environment ）

Test picture in research\object_detection\test_images Under the table of contents , You can add your own pictures , With image3... name .

Establish the target detection model

Data preparation

Use labellmage Label pictures , Generate xml file . Then from xml File generation csv file , Again by csv File generation tfrecord file , Google tensorflow Use this format file for training .

labellmage File acquisition and use

Collected relatively new versions , Please click on Download link .

Usage method ：

When the label is finished , stay object_detection New under folder images Folder , stay images New under file train,test Folder , Put the picture files into them respectively .

from xml File generation csv file

stay research\object_detection Execute the xml2csv Program .

train,test The two folders need to be generated separately

# xml2csv.py

import os
import glob
import pandas as pd
import xml.etree.ElementTree as ET

# Change to store train,test The address of 
os.chdir('C:/Users/lishu/Desktop/tensorflow/tf1/models-1.4later/models-adfd5a3aca41638aa9fb297c5095f33d64446d8f/research/object_detection/images/test')
path = 'C:/Users/lishu/Desktop/tensorflow/tf1/models-1.4later/models-adfd5a3aca41638aa9fb297c5095f33d64446d8f/research/object_detection/images/test'

def xml_to_csv(path):
    xml_list = []
    for xml_file in glob.glob(path + '/*.xml'):
        tree = ET.parse(xml_file)
        root = tree.getroot()
        for member in root.findall('object'):
            value = (root.find('filename').text,
                     int(root.find('size')[0].text),
                     int(root.find('size')[1].text),
                     member[0].text,
                     int(member[4][0].text),
                     int(member[4][1].text),
                     int(member[4][2].text),
                     int(member[4][3].text)
                     )
            xml_list.append(value)
    column_name = ['filename', 'width', 'height', 'class', 'xmin', 'ymin', 'xmax', 'ymax']
    xml_df = pd.DataFrame(xml_list, columns=column_name)
    return xml_df


def main():
    image_path = path
    xml_df = xml_to_csv(image_path)
    # Modify to generate csv File address 
    xml_df.to_csv('C:/Users/lishu/Desktop/tensorflow/tf1/models-1.4later/models-adfd5a3aca41638aa9fb297c5095f33d64446d8f/research/object_detection/data/test.csv', index=None)
    print('Successfully converted xml to csv.')


main()

from csv File generation tfrecord file

stay research\object_detection Execute the following code ：

# Create train data:
  python generate_tfrecord.py --csv_input=data/train.csv  --output_path=data/train.record
  # Create test data:
  python generate_tfrecord.py --csv_input=data/test.csv  --output_path=data/test.record

# generate_tfrecord.py

# -*- coding: utf-8 -*-


import os
import io
import pandas as pd
import tensorflow as tf

from PIL import Image
from object_detection.utils import dataset_util
from collections import namedtuple, OrderedDict

# Modify the path 
os.chdir('C:/Users/lishu/Desktop/tensorflow/tf1/models-1.4later/models-adfd5a3aca41638aa9fb297c5095f33d64446d8f/research/object_detection')

flags = tf.app.flags
flags.DEFINE_string('csv_input', '', 'Path to the CSV input')
flags.DEFINE_string('output_path', '', 'Path to output TFRecord')
FLAGS = flags.FLAGS


# TO-DO replace this with label map
def class_text_to_int(row_label):
    if row_label == 'mlj':     #  Need to change 
        return 1
    else:
        None


def split(df, group):
    data = namedtuple('data', ['filename', 'object'])
    gb = df.groupby(group)
    return [data(filename, gb.get_group(x)) for filename, x in zip(gb.groups.keys(), gb.groups)]


def create_tf_example(group, path):
    with tf.gfile.GFile(os.path.join(path, '{}'.format(group.filename)), 'rb') as fid:
        encoded_jpg = fid.read()
    encoded_jpg_io = io.BytesIO(encoded_jpg)
    image = Image.open(encoded_jpg_io)
    width, height = image.size

    filename = group.filename.encode('utf8')
    image_format = b'jpg'
    xmins = []
    xmaxs = []
    ymins = []
    ymaxs = []
    classes_text = []
    classes = []

    for index, row in group.object.iterrows():
        xmins.append(row['xmin'] / width)
        xmaxs.append(row['xmax'] / width)
        ymins.append(row['ymin'] / height)
        ymaxs.append(row['ymax'] / height)
        classes_text.append(row['class'].encode('utf8'))
        classes.append(class_text_to_int(row['class']))

    tf_example = tf.train.Example(features=tf.train.Features(feature={
        'image/height': dataset_util.int64_feature(height),
        'image/width': dataset_util.int64_feature(width),
        'image/filename': dataset_util.bytes_feature(filename),
        'image/source_id': dataset_util.bytes_feature(filename),
        'image/encoded': dataset_util.bytes_feature(encoded_jpg),
        'image/format': dataset_util.bytes_feature(image_format),
        'image/object/bbox/xmin': dataset_util.float_list_feature(xmins),
        'image/object/bbox/xmax': dataset_util.float_list_feature(xmaxs),
        'image/object/bbox/ymin': dataset_util.float_list_feature(ymins),
        'image/object/bbox/ymax': dataset_util.float_list_feature(ymaxs),
        'image/object/class/text': dataset_util.bytes_list_feature(classes_text),
        'image/object/class/label': dataset_util.int64_list_feature(classes),
    }))
    return tf_example


def main(_):
    writer = tf.python_io.TFRecordWriter(FLAGS.output_path)
    path = os.path.join(os.getcwd(), 'C:/Users/lishu/Desktop/tensorflow/tf1/models-1.4later/models-adfd5a3aca41638aa9fb297c5095f33d64446d8f/research/object_detection/images/test')         #   Need to change 
    examples = pd.read_csv(FLAGS.csv_input)
    grouped = split(examples, 'filename')
    for group in grouped:
        tf_example = create_tf_example(group, path)
        writer.write(tf_example.SerializeToString())

    writer.close()
    output_path = os.path.join(os.getcwd(), FLAGS.output_path)
    print('Successfully created the TFRecords: {}'.format(output_path))


if __name__ == '__main__':
    tf.app.run()

Model and profile settings

stay research\object_detection New under the directory training Folder , take research\object_detection\samples\configs Under folder

ssd_mobilenet_v1_pets.config File copy to training Under the folder , And modify some code ：

1、 take num_classes Change according to the actual situation , In this paper 23;

2、batch_size Change it to 1 or 2, If the computer has good performance , It can be changed to 2 The power of ;

3、input_path Change it to your own train.tfrecord route , if necessary eval Test words , You also need to modify the path .

4、 stay object_detection/data Copy one randomly under the folder .pbtxt The file copy is named hg.pbtxt, Change the code inside as follows ：

item {
name: "hb"
id: 1
}

...

representative label The name of and the corresponding number , take label_map_path Set to the corresponding hg.pbtxt route , I am here data/hg.pbtxt.

5、 Setup of pre training model fine_tune_checkpoint: "C:/Users/lishu/Desktop/tensorflow/tf1/models/research/object_detection/ssd_mobilenet_v1_coco_2017_11_17/model.ckpt"
  from_detection_checkpoint: true

The pre training model needs to be downloaded online according to the model you choose , This article USES ssd_mobilenet_v1_coco_2017_11_17.

Training models

stay research\object_detection Execute under directory ：

python train.py --logtostderr --train_dir=training/ --pipeline_config_path=training/ssd_mobilenet_v1_coco.config

Use tensorboard Check the training status

Enter the same training environment and directory research\object_detection Next , perform ：

tensorboard --logdir=training

Save the model

Use research\object_detection Under the export_inference_graph.py File save model file .

In the code

--trained_checkpoint_prefix training/model.ckpt-10317 Save training to 10317 Step saved model

--output_directory mlj Represents the folder where the model is saved

python export_inference_graph.py --input_type image_tensor --pipeline_config_path training/ssd_mobilenet_v1_coco.config --trained_checkpoint_prefix training/model.ckpt-10317 --output_directory hg

After execution mlj A series of files will be formed under the folder ,frozen_inference_graph.pb Files required for testing the model .

test model

Use eval.py test

stay research\object_detection Under the table of contents , Enter this command :

python eval.py --logtostderr --checkpoint_dir=training --eval_dir=eval --pipeline_config_path=training/ssd_mobilenet_v1_coco.config

Use Jupyter notebook test

Use Jupyter notebook open research\object_detection In the catalog object_detection_tutorial.ipynb file ,
modify Variables The content below is ：

MODEL_NAME = 'hg'
PATH_TO_CKPT = MODEL_NAME + '/frozen_inference_graph.pb'
PATH_TO_LABELS = os.path.join('data', 'hg.pbtxt')
NUM_CLASSES = 23

Delete Download Model All codes under the module ,

The test picture is on research\object_detection\test_images Under the folder , Picture with image1... name ,

Run the ipynb file , Display test results .

 

Welcome to exchange and study .