②anaconda + tensorflow2.6.2 + cuda11.2 + cudnn8.1安装使用过程

①测试tensorflow的gpu是否好用

在python解释器中输入如下代码进行测试

import tensorflow as tf

tf.test.is_gpu_available()

如果最后打印出来True，那就说明你成功了。

如果打印出来False，那么可能是你的cuda、cudnn、环境变量、独立显卡、tensorflow 版本等问题引起的，一般不会输出错误信息，这时，你就需要继续寻找教程了，拜拜

如果缺少什么cusolver64_11.dll等 dll文件，百度下载放在指定位置就可以了（如果是 anaconda环境，注意cuda和cuudn版本，conda list查看cudatoolkit 和cudnn版本与 tensorflow版本是否对应）

②anaconda + tensorflow2.6.2 + cuda11.2 + cudnn8.1安装使用过程

（建议配合官网教程）

1.安装 Anaconda

2.下载官网的模型Tensorflow model

链接1（Tensorflow model）GitHub – tensorflow/models: Models and examples built with TensorFlow

解压到C:\tensorflow2并重命名为models

3.下载官网的ssd预训练模型

链接2（SSD）GitHub – tensorflow/models: Models and examples built with TensorFlow

下载解压到C:\tensorflow1\models\research\object_detection

4.下载文件库（tensorflow训练的例子）

链接3（文件库）

https://github.com/EdjeElectronics/TensorFlow-Object-Detection-API-Tutorial-Train-Multiple-Objects-Windows-10

下载解压到C:\tensorflow1\models\research\object_detection

5.下载protuf

链接4（protuf）https://github.com/google/protobuf/releases

安装protuf  (非常容易出错，我这里选择的是win版 3.4)         将bin文件夹中的【protoc.exe】放到C:\Windows 并cmd进入models\research\  这个目录下

实施

protoc object_detection/protos/*.proto –python_out=.

三．设置conda虚拟环境

C:\>  conda create -n tensorflow2pip python=3.6.7

C:\>  activate tensorflow2

(tensorflow1) C:\> conda install tensorflow-gpu=2.6.2

(tensorflow1) C:\> conda install -c anaconda protobuf

(tensorflow1) C:\> pip install pillow

(tensorflow1) C:\> pip install lxml

(tensorflow1) C:\> pip install Cython

(tensorflow1) C:\> pip install contextlib2

(tensorflow1) C:\> pip install jupyter

(tensorflow1) C:\> pip install matplotlib

(tensorflow1) C:\> pip install pandas

(tensorflow1) C:\> pip install opencv-python

设置python环境变量,在终端执行：

SetPYTHONPATH=C:\tensorflow1\models;C:\tensorflow1\models\research;C:\tensorflow1\models\research\slim

四．编译Protobuf

cd C:\tensorflow1\models\research

protoc object_detection/protos/*.proto –python_out=.

复制C:\tensorflow2\models\research\object_detection\packages\tf2里面的setup.py

到C:\tensorflow1\models\research目录下

实施：

python setup.py build

python setup.py install

（如果python setup.py install执行时下载包太慢或者下载不了，到阿里云镜像官网找对应的包的.whl文件下载后放在C:\anaconda\pkgs目录下）

终端执行pip installC:\anaconda\pkgs\包名 进行安装

之后执行 pip installtf-nightly，自动下载tensorflow执行环境

五号。注释图片
把训练图片和测试图片放进去

C:\tensorflow1\models\research\object_detection\images\
中的train和test文件夹下

要是使用例子的图片集，注意xml的图片路径要修改（不会图片集标注百度吧）

六。生成训练数据

跑步

(tensorflow1) C:\tensorflow1\models\research\object_detection> python xml_to_csv.py

会在 \object_detection\images中生成两个文件

train_labels.csv 和 test_labels.csv

修改object_detection\目录下的generate_tfrecord.py

要修改的部分是

# TO-DO replace this with label map

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

将这部分内容中的row_label==’ ’ 修改为要训练的类别，请根据自己的情况适当增删

例如

# TO-DO replace this with label map

def class_text_to_int(row_label):

if row_label == ‘basketball’:

return 1

elif row_label == ‘shirt’:

return 2

elif row_label == ‘shoe’:

return 3

else:

None

然后生成TFRecord 文件

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

会在object_detection\目录下生成train.record和test.record文件

七．创建Label Map

label map 文件为object_detection\training\labelmap.pbtxt

修改它的内容，原来的内容是

item {

id: 1

name: ‘nine’

}

item {

id: 2

name: ‘ten’

}

item {

id: 3

name: ‘jack’

}

item {

id: 4

name: ‘queen’

}

item {

id: 5

name: ‘king’

}

item {

id: 6

name: ‘ace’

}

修改为自己的训练类别和对应ID，ID就是第7步中每个类别返回的数字

例如

if row_label == ‘basketball’:

return 1

则在labelmap.pbtxt中修改为

item {

id: 1

name: ‘basketball’

}

八。配置训练文件

将C:\tensorflow2\models\research\object_detection\configs\tf2里面的

ssd_mobilenet_v2_320x320_coco17_tpu-8.config文件复制到object_detection\training文件夹中

然后修改文件

1.num_classes: 标签种类个数

1. fine_tune_checkpoint: “C:/tensorflow2/models/research/object_detection/ssd_mobilenet_v2_320x320_coco17_tpu-8/checkpoint/ckpt-0”

1. fine_tune_checkpoint_type:”detection”

3.train_input_reader: {

label_map_path: “C:/tensorflow2/models/research/object_detection/trainning/labelmap.pbtxt”

tf_record_input_reader {

input_path: “C:/tensorflow2/models/research/object_detection/train.record”

}

}

4.eval_input_reader: {

label_map_path: “C:/tensorflow2/models/research/object_detection/trainning/labelmap.pbtxt”

shuffle: false

num_epochs: 1

tf_record_input_reader {

input_path: “C:/tensorflow2/models/research/object_detection/test.record”

}

}

batch_size 调整为1-6，太大带不起

• 开始训练
  重新训练时需要清除training\中除了labelmap和ssd_mobilenet。。。以外的所有文件，否则会报错

训练执行：

python model_main_tf2.py  –logtostderr –model_dir=training/ –pipeline_config_path=training/ssd_mobilenet_v2_320x320_coco17_tpu-8.config

继续训练，将fine_tune_checkpoint:

“C:/tensorflow2/models/research/object_detection/training/ckpt-21”

十．默认设定的迭代次数是20000次，当迭代次数达到最大后，模型训练结束，我们接着进入到training文件夹下可以看到有如下生成的文件：

检查训练模型的收敛程度：

D:\Program Files\models\research\object_detection>

tensorboard –logdir training –bind_all

从上图还是可以看出Loss呈下降趋势 ，实际效果如何还需要进一步评估。

十一：导出.pb文件

D:\Program Files\models-master\research\object_detection>

创建文件夹inference_graph

（1）tf1版本：导出frozen_inference_graph.pb文件。

python export_inference_graph.py –input_type image_tensor –pipeline_config_path training/ssd_mobilenet_v2_320x320_coco17_tpu-8.config –trained_checkpoint_prefix training/ckpt-40  –output_directory inference_graph

（2）tf2版本：导出saved_model.pb，该文件中包含了我们训练好的检测器以及网络架构信息和参数信息等。

python exporter_main_v2.py –input_type  image_tensor  –pipeline_config_path=training/ssd_mobilenet_v2_320x320_coco17_tpu-8.config –trained_checkpoint_dir=training   –output_directory=training/train_export

运行结果如下图所示：

十二．测试Tensorflow模型

写一个Object_detection_image_LLL.py

######## Image Object Detection Using Tensorflow-trained Classifier #########
#
# Author: Evan Juras
# Date: 1/15/18
# Description:
# This program uses a TensorFlow-trained neural network to perform object detection.
# It loads the classifier and uses it to perform object detection on an image.
# It draws boxes, scores, and labels around the objects of interest in the image.

## Some of the code is copied from Google’s example at
## https://github.com/tensorflow/models/blob/master/research/object_detection/object_detection_tutorial.ipynb

## and some is copied from Dat Tran’s example at
## https://github.com/datitran/object_detector_app/blob/master/object_detection_app.py

## but I changed it to make it more understandable to me.

# Import packages
importos

importcv2
importnumpyasnp
fromPILimportImage
importtkinter
importmatplotlib

matplotlib.use(‘TkAgg’)
importmatplotlib.pyplotasplt

importtime
importtensorflowastf

fromobject_detection.utilsimportlabel_map_util
fromobject_detection.utilsimportconfig_util
fromobject_detection.utilsimportvisualization_utilsasviz_utils
fromobject_detection.buildersimportmodel_builder
fromsiximportBytesIO
importsys

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

# Import utilites
fromutilsimportlabel_map_util
fromutilsimportvisualization_utilsasvis_util

#前向传播函数
defget_model_detection_function(model):
“””Get a tf.function for detection.”””

@tf.function
defdetect_fn(image):
“””Detect objects in image.”””

image,shapes = model.preprocess(image)
prediction_dict = model.predict(image,shapes)
detections = model.postprocess(prediction_dict,shapes)

returndetections,prediction_dict,tf.reshape(shapes,[-1])

returndetect_fn

# Grab path to current working directory
CWD_PATH = os.getcwd()

# Name of the directory containing the object detection module we’re using
IMAGE_NAME =’test3.jpg’

# Path to image
PATH_TO_IMAGE = os.path.join(CWD_PATH,IMAGE_NAME)

#构建检测模型并加载trained_model权值
pipeline_config = os.path.join(CWD_PATH,’training\\train_export\\’,’pipeline.config’)
model_dir =’training\\train_export\\checkpoint\\’
# Load pipeline config and build a detection model
#加载pipeline config然后读取检测模型
configs = config_util.get_configs_from_pipeline_file(pipeline_config)
model_config = configs[‘model’]
detection_model = model_builder.build(model_config=model_config,is_training=False)
# 恢复模型
ckpt = tf.compat.v2.train.Checkpoint(model=detection_model)
ckpt.restore(os.path.join(CWD_PATH,model_dir,’ckpt-0′)).expect_partial()

#加载模型权重
detect_fn = get_model_detection_function(detection_model)

#从pipeline config读取label_map信息
label_map_path = configs[‘eval_input_config’].label_map_path
label_map = label_map_util.load_labelmap(label_map_path)
categories = label_map_util.convert_label_map_to_categories(
label_map,
max_num_classes=label_map_util.get_max_label_map_index(label_map),
use_display_name=True)
category_index = label_map_util.create_category_index(categories)
label_map_dict = label_map_util.get_label_map_dict(label_map,use_display_name=True)

#阅读图片
image = cv2.imread(PATH_TO_IMAGE)
image_rgb = cv2.cvtColor(image,cv2.COLOR_BGR2RGB)
input_tensor = tf.convert_to_tensor(np.expand_dims(image_rgb,0),dtype=tf.float32)

#image 在模型中前向传播
detections,predictions_dict,shapes = detect_fn(input_tensor)

# Draw the results of the detection (aka ‘visulaize the results’)
#绘制结果
vis_util.visualize_boxes_and_labels_on_image_array(
image,
detections[‘detection_boxes’][0].numpy(),
detections[‘detection_classes’][0].numpy().astype(int),
detections[‘detection_scores’][0].numpy(),
category_index,
use_normalized_coordinates=True,
max_boxes_to_draw=200,
min_score_thresh=0.6,
agnostic_mode=False)

# All the results have been drawn on image. Now display the image.
cv2.imshow(‘Object detector’,image)

# Press any key to close the image
cv2.waitKey(0)

# Clean up
cv2.destroyAllWindows()

注意修改读取路径

③模型转换.pb 到 .tflite

查看 tensorflow Lite官网文档