使用Tensorflow1.9用自己的数据集训练ssd-mobilenet
1.环境搭建
电脑装了ubuntu20.04，nv驱动的安装比较简单了，直接在设置里就额外驱动里就可以找到，如果那里没有，就下载一个nv驱动，点击安装却并不真地安装，这样而外驱动的选项卡里就会有了
1.1安装docker
就按照官网的步骤安装
1.2使用docker安装tensorflow
按照官网教程安装
https://tensorflow.google.cn/install/docker
1.3使用docker安装cuda支持
按照教程安装，注意保证网络通畅
https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/install-guide.html#docker
上述都安装完毕以后，就可以打开支持gpu的tensorflow image了，这里选择了tensorflow 1.9 + protobuf 3.6的组合
2.制作训练集
2.1数据采集
首先获取训练集图片，可以自己写一段程序来截图，也可以录制视频再转换成图片，这里使用第二种方法走一遍整个流程。
使用ubuntu自带程序cheese连接相机录制视频（如果没有cheese安装一个）
使用ffmpeg将视频转换为图片
sudo apt-get install cheese
sudo apt-get install ffmpeg// 视频路径 每隔几秒截取一张图片 图片路径
ffmpeg -i ./test/video.mpg -r 1 -f image2 temp/%05d.png
参考链接：https://blog.csdn.net/felaim/article/details/73656049
2.2数据标注
使用LabelImg标注工具：https://github.com/tzutalin/labelImg
如果环境安装不明白，强烈推荐直接用docker
docker run -it \
--user $(id -u) \
-e DISPLAY=unix$DISPLAY \
--workdir=$(pwd) \
--volume="/home/$USER:/home/$USER" \
--volume="/etc/group:/etc/group:ro" \
--volume="/etc/passwd:/etc/passwd:ro" \
--volume="/etc/shadow:/etc/shadow:ro" \
--volume="/etc/sudoers.d:/etc/sudoers.d:ro" \
-v /tmp/.X11-unix:/tmp/.X11-unix \
tzutalin/py2qt4参考连接：https://blog.csdn.net/xunan003/article/details/78720189
2.3数据格式转换
1)将xml转为csv
import os
import glob
import pandas as pd
import xml.etree.ElementTree as ET
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():
for folder in ['train','test']:
image_path = os.path.join(os.getcwd(), ('images/' + folder)) #这里就是需要访问的.xml的存放地址
xml_df = xml_to_csv(image_path) #object_detection/images/train or test
xml_df.to_csv(('images/' + folder + '_labels.csv'), index=None)
print('Successfully converted xml to csv.')
main()
通过修改为自己的xml路径与输出路径，就可以完成转换。这个代码对应的目录结构为：
生成train_labels.csv 和 test_labels.csv
2)将csv转换为TFrecord
TFrecord是google专门设计的一种数据格式，速度快，占用内存少。
我是用的别的博客上的python脚本转换的
"""
Usage:
# From tensorflow/models/
# Create train data:
python generate_tfrecord.py --csv_input=images/train_labels.csv --image_dir=images/train --output_path=train.record
# Create test data:
python generate_tfrecord.py --csv_input=images/test_labels.csv --image_dir=images/test --output_path=test.record
"""
from __future__ import division
from __future__ import print_function
from __future__ import absolute_import
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
flags = tf.app.flags
flags.DEFINE_string('csv_input', '', 'Path to the CSV input')
flags.DEFINE_string('image_dir', '', 'Path to the image directory')
flags.DEFINE_string('output_path', '', 'Path to output TFRecord')
FLAGS = flags.FLAGS
# M1，this code part need to be modified according to your real situation
def class_text_to_int(row_label):
if row_label == 'nine':
return 1
elif row_label == 'ten':
return 2
elif row_label == 'jack':
return 3
elif row_label == 'queen':
return 4
elif row_label == 'king':
return 5
elif row_label == 'ace':
return 6
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(), FLAGS.image_dir)
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()
这个脚本用的是tensorflow1，所以如果你用的是tensorflow2那么这个脚本就用不了
这个脚本运行时也有一些坑，主要有两个
一个是数据集路径设置一定要正确
还有一个是class_text_to_int这个函数里的标签设置必须和自己制作的数据集对应
python generate_tfrecord.py --csv_input=images/train_labels.csv --image_dir=images/train --output_path=train.record
python generate_tfrecord.py --csv_input=images/test_labels.csv --image_dir=images/test --output_path=test.record
这里代码对应的目录结构为：
object_detection >> images >> train + test
这句代码是在object_detection目录中执行的，可以顺利生成TFrecord文件
3.训练模型
这里许多博客都是说下载自己需要的模型，然后解压到object_detection API的目录下，例如自己下载了一个ssd_model，那么就把它解压到如下目录里
models/research/object_detection/ssd_model
但是那些博客给的模型下载连接都失效了，应给是tf2以后这里发生了比较大的变化，后来我在一篇博客里下载到了一个，但是忘记下载链接了。所谓的模型就是以下几个文件：
这个连接里给的模型我没有试，里边只有一个frozen_inference_graph.pb: https://pan.baidu.com/s/17hyoo-7zrVOc8ZAX8XEh3w 提取码：noff
3.1拷贝并且修改两个配置文件
cp object_detection/data/pascal_label_map.pbtxt object_detection/ssd_model/
cp object_detection/samples/configs/ssd_mobilenet_v1_pets.config object_detection/ssd_model/.pbtxt文件里存的是识别的类别标签，需要根据自己的数据集进行更改，下边是文件原来的样子
item {
id: 1
name: 'aeroplane'
}
item {
id: 2
name: 'bicycle'
}
item {
id: 3
name: 'bird'
}
item {
id: 4
name: 'boat'
}
item {
id: 5
name: 'bottle'
}
.config文件是和训练有关的参数，我们必须要改的是5个路径：
// 这个路径要改成自己的
fine_tune_checkpoint: "PATH_TO_BE_CONFIGURED/model.ckpt"
from_detection_checkpoint: true
# Note: The below line limits the training process to 200K steps, which we
# empirically found to be sufficient enough to train the pets dataset. This
# effectively bypasses the learning rate schedule (the learning rate will
# never decay). Remove the below line to train indefinitely.
num_steps: 200000
data_augmentation_options {
random_horizontal_flip {
}
}
data_augmentation_options {
ssd_random_crop {
}
}
}
// 下边路径要改成自己的
train_input_reader: {
tf_record_input_reader {
input_path: "PATH_TO_BE_CONFIGURED/mscoco_train.record-?????-of-00100"
}
label_map_path: "PATH_TO_BE_CONFIGURED/mscoco_label_map.pbtxt"
}
eval_config: {
num_examples: 8000
# Note: The below line limits the evaluation process to 10 evaluations.
# Remove the below line to evaluate indefinitely.
max_evals: 10
}
eval_input_reader: {
tf_record_input_reader {
input_path: "PATH_TO_BE_CONFIGURED/mscoco_val.record-?????-of-00010"
}
label_map_path: "PATH_TO_BE_CONFIGURED/mscoco_label_map.pbtxt"
shuffle: false
num_readers: 1
}注意把from_detection_checkpoint: true打开
3.2训练模型
这里是使用legacy/train.py来训练，下面是代码里的注释：
r"""Training executable for detection models.
This executable is used to train DetectionModels. There are two ways of
configuring the training job:
1) A single pipeline_pb2.TrainEvalPipelineConfig configuration file
can be specified by --pipeline_config_path.
Example usage:
./train \
--logtostderr \
--train_dir=path/to/train_dir \
--pipeline_config_path=pipeline_config.pbtxt
2) Three configuration files can be provided: a model_pb2.DetectionModel
configuration file to define what type of DetectionModel is being trained, an
input_reader_pb2.InputReader file to specify what training data will be used and
a train_pb2.TrainConfig file to configure training parameters.
Example usage:
./train \
--logtostderr \
--train_dir=path/to/train_dir \
--model_config_path=model_config.pbtxt \
--train_config_path=train_config.pbtxt \
--input_config_path=train_input_config.pbtxt
"""python object_detection/legacy/train.py --logtostderr --train_dir=object_detection/ssd_mobilenet_v1_coco_2018_01_28/training --model_dir=object_detection/ssd_mobilenet_v1_coco_2018_01_28 --pipeline_config_path=object_detection/ssd_mobilenet_v1_coco_2018_01_28/ssd_mobilenet_v1_coco.config--train_dir是存放这次训练出来的网络的
--model_dir是模型存放预训练模型的
--pipelin_config_path是要用的config文件路径，就是我们上边改的那个配置文件
训练过程中如果遇到OOM(out of memory)问题，多半是因为显存不够，修改config里的batch_size应该能够解决，实在不行就换个显卡吧
通过nvidia-smi可以监控显卡运行状态：
https://blog.csdn.net/qq_24878901/article/details/81660309
https://blog.csdn.net/ytusdc/article/details/85229771
训练过程可以使用tensorboard监控，但是我使用的时候网页打不开，可能是docker的端口设置的不对，之前用jupyter时也出现过类似的问题
tensorboard --logdir=/train_dir=object_detection/ssd_mobilenet_v1_coco_2018_01_28/training把训练模型转换为pb文件
python object_detection/export_inference_graph.py --input_type image_tensor --pipeline_config_path object_detection/ssd_mobilenet_v1_coco_2018_01_28/ssd_mobilenet_v1_coco.config --trained_checkpoint_prefix object_detection/ssd_mobilenet_v1_coco_2018_01_28/training/model.ckpt-200000 --output_directory object_detection/ssd_mobilenet_v1_coco_2018_01_28/trained_model这里 --trained_checkpoint_prefix 这个参数要在checkpoint文件里找到一个checkpoint
完成之后trained_model文件夹下会生成如下文件：
pb文件在saved_model里
4.模型测试
使用以下测试代码测试时提示OOM错误
import cv2
import numpy as np
import tensorflow as tf
from object_detection.utils import label_map_util
from object_detection.utils import visualization_utils as vis_util
class TOD(object):
def __init__(self):
self.PATH_TO_CKPT = '/home/lyf/models/research/object_detection/ssd_model/model/frozen_inference_graph.pb'
self.PATH_TO_LABELS = '/home/lyf/models/research/object_detection/ssd_model/pascal_label_map.pbtxt'
self.NUM_CLASSES = 1
self.detection_graph = self._load_model()
self.category_index = self._load_label_map()
def _load_model(self):
detection_graph = tf.Graph()
with detection_graph.as_default():
od_graph_def = tf.GraphDef()
with tf.gfile.GFile(self.PATH_TO_CKPT, 'rb') as fid:
serialized_graph = fid.read()
od_graph_def.ParseFromString(serialized_graph)
tf.import_graph_def(od_graph_def, name='')
return detection_graph
def _load_label_map(self):
label_map = label_map_util.load_labelmap(self.PATH_TO_LABELS)
categories = label_map_util.convert_label_map_to_categories(label_map,
max_num_classes=self.NUM_CLASSES,
use_display_name=True)
category_index = label_map_util.create_category_index(categories)
return category_index
def detect(self, image):
with self.detection_graph.as_default():
with tf.Session(graph=self.detection_graph) as sess:
# Expand dimensions since the model expects images to have shape: [1, None, None, 3]
image_np_expanded = np.expand_dims(image, axis=0)
image_tensor = self.detection_graph.get_tensor_by_name('image_tensor:0')
boxes = self.detection_graph.get_tensor_by_name('detection_boxes:0')
scores = self.detection_graph.get_tensor_by_name('detection_scores:0')
classes = self.detection_graph.get_tensor_by_name('detection_classes:0')
num_detections = self.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.squeeze(boxes),
np.squeeze(classes).astype(np.int32),
np.squeeze(scores),
self.category_index,
use_normalized_coordinates=True,
line_thickness=8)
cv2.namedWindow("detection", cv2.WINDOW_NORMAL)
cv2.imshow("detection", image)
cv2.waitKey(0)
if __name__ == '__main__':
image = cv2.imread('/home/lyf/test/000084.jpg')
detecotr = TOD()
detecotr.detect(image)无奈使用object_detection API进行测试
用jupyter-notebook时，由于我是用的docker，所以要在/object_detection文件夹下打开jupyter，要不然看不到object_detection_tutorial.ipynb
还需要对object_detection_tutorial.ipynb进行一些修改
这里是模型的路径以及自己设置的标签的路径，一定要仔细核对
# What model to download.
MODEL_NAME = 'ssd_mobilenet_v1_coco_2018_01_28'
#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_FROZEN_GRAPH = MODEL_NAME + '/trained_model/frozen_inference_graph.pb'
#这里一定要选择frozen这个，saved_model里那个.pb不行，我也不知道有啥区别
# List of the strings that is used to add correct label for each box.
PATH_TO_LABELS = os.path.join('ssd_mobilenet_v1_coco_2018_01_28', 'my_label_map.pbtxt')
donwload部分都注释掉
#opener = urllib.request.URLopener()
#opener.retrieve(DOWNLOAD_BASE + MODEL_FILE, MODEL_FILE)
#ar_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())自己的测试集
# 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 = 'ssd_mobilenet_v1_coco_2018_01_28/images/test'
TEST_IMAGE_PATHS = [ os.path.join(PATH_TO_TEST_IMAGES_DIR, '00076.jpg')]#.format(i)) for i in range(73, 83)
# Size, in inches, of the output images.
IMAGE_SIZE = (12, 8)然后就一路运行就可以看到结果了
5.模型转换为tflite
因为是在嵌入式设备上运行，把模型转换为tflite
detection API里有现成的转换工具
python export_tflite_ssd_graph.py --pipeline_config_path=ssd_mobilenet_v1_coco_2018_01_28/ssd_mobilenet_v1_coco.config --trained_checkpoint_prefix=ssd_mobilenet_v1_coco_2018_01_28/trained_model/model.ckpt --output_directory=ssd_mobilenet_v1_coco_2018_01_28/tflite_model --add_postprocessing_op=false这个脚本会生成以下文件：
那个pb文件就是我们熟悉的22.4Mb的网络模型了，但是我们最终想要的是.tflite文件，查看了一些博客，需要安装bazel编译工具，还要编译tensorflow源码，知难而退，看看有没有什么简单的办法，又参考了一片博客：https://blog.csdn.net/HK_Joe/article/details/102730752
import tensorflow as tf
in_path=r'D:\tmp_mobilenet_v1_100_224_classification_3\output_graph.pb'
out_path=r'D:\tmp_mobilenet_v1_100_224_classification_3\output_graph.tflite'
input_tensor_name=['Placeholder']
input_tensor_shape={'Placeholder':[1,224,224,3]}
class_tensor_name=['final_result']
convertr=tf.lite.TFLiteConverter.from_frozen_graph(in_path,input_arrays=input_tensor_name
,output_arrays=class_tensor_name
,input_shapes=input_tensor_shape)
# convertr=tf.lite.TFLiteConverter.from_saved_model(saved_model_dir=in_path,input_arrays=[input_tensor_name],output_arrays=[class_tensor_name])
tflite_model=convertr.convert()
with open(out_path,'wb') as f:
f.write(tflite_model)结果提示
AttributeError: module 'tensorflow' has no attribute 'lite'看来是tensorflow版本不对，查找到另外篇博客
https://zhuanlan.zhihu.com/p/138738210
先不转格式了，这都是小问题了
参考链接：
https://blog.csdn.net/qq_37643960/article/details/98069548