这是机器未来的第4篇文章

写在前面：

• 博客简介：专注AIoT领域，追逐未来时代的脉搏，记录路途中的技术成长！

• 专栏简介：记录博主从0到1掌握物体检测工作流的过程，具备自定义物体检测器的能力

• 面向人群：具备深度学习理论基础的学生或初级开发者

• 专栏计划：接下来会逐步发布跨入人工智能的系列博文，敬请期待

• Python零基础快速入门系列

• 快速入门Python数据科学系列

• 人工智能开发环境搭建系列

• 机器学习系列

• 物体检测快速入门系列

• 自动驾驶物体检测系列

• ......

@[toc]

1. 概述

tensorflow object detection api一个框架，它可以很容易地构建、训练和部署对象检测模型，并且是一个提供了众多基于COCO数据集、Kitti数据集、Open Images数据集、AVA v2.1数据集和iNaturalist物种检测数据集上提供预先训练的对象检测模型集合。

tensorflow object detection api是目前最主流的目标检测框架之一，主流的目标检测模型如图所示： snipaste20220513_094828

2. 预置条件

为了顺利按照本手册安装tensroflow object detection api，请参考Windows部署Docker GPU深度学习开发环境安装必备的工具。

若自行创建安装条件，请确保已经满足以下条件

• 支持python3.8以上版本
• 支持cuda、cudnn（可选）
• 支持git

本手册使用docker运行环境。

3. 安装步骤

3.1 Docker环境

3.1.1 启动docker

启动docker桌面客户端，如图所示： 1

3.1.2 启动容器

在windows平台可以启动命令行工具或者windows terminal工具(App Store下载)，这里使用terminal工具。

输入以下命令，查看当前存在的images列表

PS C:\Users\xxxxx> docker images
REPOSITORY               TAG                 IMAGE ID       CREATED       SIZE
docker/getting-started   latest              bd9a9f733898   5 weeks ago   28.8MB
tensorflow/tensorflow    2.8.0-gpu-jupyter   cc9a9ae2a5af   6 weeks ago   5.99GB

可以看到之前安装的tensorflow-2.8.0-gpu-jupyter镜像，现在基于这个镜像启动容器

docker run --gpus all -itd -v e:/dockerdir/docker_work/:/home/zhou/ -p 8888:8888 -p 6006:6006 --ipc=host cc9a9ae2a5af  jupyter notebook --no-browser --ip=0.0.0.0 --allow-root --NotebookApp.token= --notebook-dir='/home/zhou/'

命令释义： docker run：表示基于镜像启动容器 --gpus all：不加此选项,nvidia-smi命令会不可用 -i: 交互式操作。 -t: 终端。 -d:后台运行，需要使用【docker exec -it 容器id /bin/bash】进入容器 -v e:/dockerdir/docker_work/:/home/zhou/:将windows平台的e:/dockerdir/docker_work/目录映射到docker的ubuntu系统的/home/zhou/目录下，实现windows平台和docker系统的文件共享 -p 8888:8888 -p 6006:6006：表示将windows系统的8888、6006端口映射到docker的8888、6006端口，这两个端口分别为jupyter notebook和tensorboard的访问端口 --ipc=host:用于多个容器之间的通讯 cc9a9ae2a5af:tensorflow-2.8.0-gpu-jupyter镜像的IMAGE ID jupyter notebook --no-browser --ip=0.0.0.0 --allow-root --NotebookApp.token= --notebook-dir='/home/zhou/': docker开机启动命令，这里启动jupyter

3.1.3 使用vscode访问docker container

启动vscode后，选择docker工具栏，在启动的容器上，右键选择附着到VsCode2

3.1.4 更换docker容器ubuntu系统的安装源为国内源

在vscode软件界面上，选择【文件】-【打开文件夹】，选择根目录【/】，找到【/etc/apt/sources.list】,将ubuntu的安装源全部切换为aliyun源，具体操作为：将【archive.ubuntu.com】修改为【mirrors.aliyun.com】即可，修改后如下：

# See http://help.ubuntu.com/community/UpgradeNotes for how to upgrade to
# newer versions of the distribution.
deb http://mirrors.aliyun.com/ubuntu/ focal main restricted
# deb-src http://mirrors.aliyun.com/ubuntu/ focal main restricted
## Major bug fix updates produced after the final release of the
## distribution.
deb http://mirrors.aliyun.com/ubuntu/ focal-updates main restricted
# deb-src http://mirrors.aliyun.com/ubuntu/ focal-updates main restricted
## N.B. software from this repository is ENTIRELY UNSUPPORTED by the Ubuntu
## team. Also, please note that software in universe WILL NOT receive any
## review or updates from the Ubuntu security team.
deb http://mirrors.aliyun.com/ubuntu/ focal universe
# deb-src http://mirrors.aliyun.com/ubuntu/ focal universe
deb http://mirrors.aliyun.com/ubuntu/ focal-updates universe
# deb-src http://mirrors.aliyun.com/ubuntu/ focal-updates universe
## N.B. software from this repository is ENTIRELY UNSUPPORTED by the Ubuntu
## team, and may not be under a free licence. Please satisfy yourself as to
## your rights to use the software. Also, please note that software in
## multiverse WILL NOT receive any review or updates from the Ubuntu
## security team.
deb http://mirrors.aliyun.com/ubuntu/ focal multiverse
# deb-src http://mirrors.aliyun.com/ubuntu/ focal multiverse
deb http://mirrors.aliyun.com/ubuntu/ focal-updates multiverse
# deb-src http://mirrors.aliyun.com/ubuntu/ focal-updates multiverse
## N.B. software from this repository may not have been tested as
## extensively as that contained in the main release, although it includes
## newer versions of some applications which may provide useful features.
## Also, please note that software in backports WILL NOT receive any review
## or updates from the Ubuntu security team.
deb http://mirrors.aliyun.com/ubuntu/ focal-backports main restricted universe multiverse
# deb-src http://mirrors.aliyun.com/ubuntu/ focal-backports main restricted universe multiverse
## Uncomment the following two lines to add software from Canonical's
## 'partner' repository.
## This software is not part of Ubuntu, but is offered by Canonical and the
## respective vendors as a service to Ubuntu users.
# deb http://archive.canonical.com/ubuntu focal partner
# deb-src http://archive.canonical.com/ubuntu focal partner
deb http://security.ubuntu.com/ubuntu/ focal-security main restricted
# deb-src http://security.ubuntu.com/ubuntu/ focal-security main restricted
deb http://security.ubuntu.com/ubuntu/ focal-security universe
# deb-src http://security.ubuntu.com/ubuntu/ focal-security universe
deb http://security.ubuntu.com/ubuntu/ focal-security multiverse
# deb-src http://security.ubuntu.com/ubuntu/ focal-security multiverse

• 执行如下命令，更新配置

apt-get update;apt-get -f install; apt-get upgrade

• 更多aliyun的源配置访问：阿里云安装源传送门

3.1.5 验证GPU是否加载成功（在电脑有Nvidia显卡的情况下）

• 输入nvidia-smi查看GPU使用情况，nvcc -V查询cuda版本

root@cc58e655b170:/home/zhou# nvidia-smi
Tue Mar 22 15:08:57 2022       
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 470.85       Driver Version: 472.47       CUDA Version: 11.4     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|                               |                      |               MIG M. |
|===============================+======================+======================|
|   0  NVIDIA GeForce ...  Off  | 00000000:01:00.0 Off |                  N/A |
| N/A   48C    P8     9W /  N/A |    153MiB /  6144MiB |    ERR!      Default |
|                               |                      |                  N/A |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes:                                                                  |
|  GPU   GI   CI        PID   Type   Process name                  GPU Memory |
|        ID   ID                                                   Usage      |
|=============================================================================|
|  No running processes found                                                 |
+-----------------------------------------------------------------------------+
root@cc58e655b170:/home/zhou# nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Sun_Feb_14_21:12:58_PST_2021
Cuda compilation tools, release 11.2, V11.2.152
Build cuda_11.2.r11.2/compiler.29618528_0

从nvcc -V的日志，可以看出cuda版本为11.2

• 输入以下命令，查询cuDNN版本

python -c "import tensorflow as tf;print(tf.reduce_sum(tf.random.normal([1000, 1000])))"

输出结果如下：

root@cc58e655b170:/usr# python -c "import tensorflow as tf;print(tf.reduce_sum(tf.random.normal([1000, 1000])))"
2022-03-22 15:26:13.281719: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1525] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 3951 MB memory:  -> device: 0, name: NVIDIA GeForce GTX 1660 Ti, pci bus id: 0000:01:00.0, compute capability: 7.5
tf.Tensor(-2613.715, shape=(), dtype=float32)

从输出日志，可以看到GPU：NVIDIA GeForce GTX 1660 Ti已经加载到docker，cuDNN版本为7.5

3.2 Windows开发环境

同Docker环境，验证cuda和cuDNN安装情况。

3.3 下载tensorflow object detection api项目源码

• 在home/zhou目录下创建tensorflow的目录

cd /home/zhou; mkdir tensorflow; cd tensorflow

• 下载源码

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

下载完毕后，默认文件名名称为models-master, 将文件名重命名为models,保持文件名和平台一致

mv models-matser models

如果网速不好，直接下载zip压缩包吧 3下载完毕后的文档结构如图所示：

tensorflow/
└─ models/
   ├─ community/
   ├─ official/
   ├─ orbit/
   ├─ research/
   └── ...

3.4 安装配置protobuf

Tensorflow对象检测API使用Protobufs来配置模型和训练参数。在使用框架之前，必须下载并编译Protobuf库。

cd /home/zhou

• 下载protobuf 这里下载的已经预编译好的protobuf

wget -c https://github.com/protocolbuffers/protobuf/releases/download/v3.19.4/protoc-3.19.4-linux-x86_64.zip

• 解压先执行mkdir protoc-3.19.4创建目录，然后执行unzip protoc-3.19.4-linux-x86_64.zip -d protoc-3.19.4/解压到制定目录protoc-3.19.4

root@cc58e655b170:/home/zhou# mkdir protoc-3.19.4
root@cc58e655b170:/home/zhou# unzip protoc-3.19.4-linux-x86_64.zip -d protoc-3.19.4/
Archive:  protoc-3.19.4-linux-x86_64.zip
   creating: protoc-3.19.4/include/
   creating: protoc-3.19.4/include/google/
   creating: protoc-3.19.4/include/google/protobuf/
  inflating: protoc-3.19.4/include/google/protobuf/wrappers.proto  
  inflating: protoc-3.19.4/include/google/protobuf/source_context.proto  
  inflating: protoc-3.19.4/include/google/protobuf/struct.proto  
  inflating: protoc-3.19.4/include/google/protobuf/any.proto  
  inflating: protoc-3.19.4/include/google/protobuf/api.proto  
  inflating: protoc-3.19.4/include/google/protobuf/descriptor.proto  
   creating: protoc-3.19.4/include/google/protobuf/compiler/
  inflating: protoc-3.19.4/include/google/protobuf/compiler/plugin.proto  
  inflating: protoc-3.19.4/include/google/protobuf/timestamp.proto  
  inflating: protoc-3.19.4/include/google/protobuf/field_mask.proto  
  inflating: protoc-3.19.4/include/google/protobuf/empty.proto  
  inflating: protoc-3.19.4/include/google/protobuf/duration.proto  
  inflating: protoc-3.19.4/include/google/protobuf/type.proto  
   creating: protoc-3.19.4/bin/
  inflating: protoc-3.19.4/bin/protoc  
  inflating: protoc-3.19.4/readme.txt

• 配置protoc 在~/.bashrc文件的末尾添加如下代码

export PATH=$PATH:/home/zhou/protoc-3.19.4/bin

执行如下命令，使其生效

source ~/.bashrc

执行echo $PATH查看是否生效

root@cc58e655b170:/home/zhou/protoc-3.19.4/bin# echo $PATH
/home/zhou/protoc-3.19.4/bin:/home/zhou/protoc-3.19.4/bin:/home/zhou/protoc-3.19.4/bin:/root/.vscode-server/bin/c722ca6c7eed3d7987c0d5c3df5c45f6b15e77d1/bin/remote-cli:/usr/local/nvidia/bin:/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/home/zhou/protoc-3.19.4/bin

可以看到protoc的安装目录/home/zhou/protoc-3.19.4/bin已经添加到PATH了。

3.5 将proto后缀文件转换为python可识别格式

cd /home/zhou/tensorflow/models/research/

ls object_detection/protos/

• 4转换proto文件格式为python可识别序列化文件

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

• 转换后，如下所示

ls object_detection/protos/

3.6 安装coco api

从TensorFlow 2.x开始, pycocotools包被列为对象检测API的依赖项。理想情况下，这个包应该在安装对象检测API时安装，如下面安装对象检测API一节所述，但是由于各种原因，安装可能会失败，因此更简单的方法是提前安装这个包，在这种情况下，稍后的安装将被跳过。

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

默认指标是基于Pascal VOC评估中使用的那些指标。要使用COCO对象检测指标，在配置文件的eval_config消息中添加metrics_set: "coco_detection_metrics"。要使用COCO实例分割度量，在配置文件的eval_config消息中添加metrics_set: "coco_mask_metrics"。

3.7 安装object detection api

• 当前的工作路径应为

root@cc58e655b170:/home/zhou/tensorflow/models/research# pwd
/home/zhou/tensorflow/models/research

• 安装object detection api

cp object_detection/packages/tf2/setup.py .
python -m pip install --use-feature=2020-resolver .

安装过程会持续一段时间，安装完毕后，可以执行如下代码，测试安装是否完成。

python object_detection/builders/model_builder_tf2_test.py

输出如下：

......
I0322 16:48:09.677789 140205126002496 efficientnet_model.py:144] round_filter input=192 output=384
I0322 16:48:10.876914 140205126002496 efficientnet_model.py:144] round_filter input=192 output=384
I0322 16:48:10.877072 140205126002496 efficientnet_model.py:144] round_filter input=320 output=640
I0322 16:48:11.294571 140205126002496 efficientnet_model.py:144] round_filter input=1280 output=2560
I0322 16:48:11.337533 140205126002496 efficientnet_model.py:454] Building model efficientnet with params ModelConfig(width_coefficient=2.0, depth_coefficient=3.1, resolution=600, dropout_rate=0.5, blocks=(BlockConfig(input_filters=32, output_filters=16, kernel_size=3, num_repeat=1, expand_ratio=1, strides=(1, 1), se_ratio=0.25, id_skip=True, fused_conv=False, conv_type='depthwise'), BlockConfig(input_filters=16, output_filters=24, kernel_size=3, num_repeat=2, expand_ratio=6, strides=(2, 2), se_ratio=0.25, id_skip=True, fused_conv=False, conv_type='depthwise'), BlockConfig(input_filters=24, output_filters=40, kernel_size=5, num_repeat=2, expand_ratio=6, strides=(2, 2), se_ratio=0.25, id_skip=True, fused_conv=False, conv_type='depthwise'), BlockConfig(input_filters=40, output_filters=80, kernel_size=3, num_repeat=3, expand_ratio=6, strides=(2, 2), se_ratio=0.25, id_skip=True, fused_conv=False, conv_type='depthwise'), BlockConfig(input_filters=80, output_filters=112, kernel_size=5, num_repeat=3, expand_ratio=6, strides=(1, 1), se_ratio=0.25, id_skip=True, fused_conv=False, conv_type='depthwise'), BlockConfig(input_filters=112, output_filters=192, kernel_size=5, num_repeat=4, expand_ratio=6, strides=(2, 2), se_ratio=0.25, id_skip=True, fused_conv=False, conv_type='depthwise'), BlockConfig(input_filters=192, output_filters=320, kernel_size=3, num_repeat=1, expand_ratio=6, strides=(1, 1), se_ratio=0.25, id_skip=True, fused_conv=False, conv_type='depthwise')), stem_base_filters=32, top_base_filters=1280, activation='simple_swish', batch_norm='default', bn_momentum=0.99, bn_epsilon=0.001, weight_decay=5e-06, drop_connect_rate=0.2, depth_divisor=8, min_depth=None, use_se=True, input_channels=3, num_classes=1000, model_name='efficientnet', rescale_input=False, data_format='channels_last', dtype='float32')
INFO:tensorflow:time(__main__.ModelBuilderTF2Test.test_create_ssd_models_from_config): 33.12s
I0322 16:48:11.521103 140205126002496 test_util.py:2373] time(__main__.ModelBuilderTF2Test.test_create_ssd_models_from_config): 33.12s
[       OK ] ModelBuilderTF2Test.test_create_ssd_models_from_config
[ RUN      ] ModelBuilderTF2Test.test_invalid_faster_rcnn_batchnorm_update
INFO:tensorflow:time(__main__.ModelBuilderTF2Test.test_invalid_faster_rcnn_batchnorm_update): 0.0s
I0322 16:48:11.532667 140205126002496 test_util.py:2373] time(__main__.ModelBuilderTF2Test.test_invalid_faster_rcnn_batchnorm_update): 0.0s
[       OK ] ModelBuilderTF2Test.test_invalid_faster_rcnn_batchnorm_update
[ RUN      ] ModelBuilderTF2Test.test_invalid_first_stage_nms_iou_threshold
INFO:tensorflow:time(__main__.ModelBuilderTF2Test.test_invalid_first_stage_nms_iou_threshold): 0.0s
I0322 16:48:11.535152 140205126002496 test_util.py:2373] time(__main__.ModelBuilderTF2Test.test_invalid_first_stage_nms_iou_threshold): 0.0s
[       OK ] ModelBuilderTF2Test.test_invalid_first_stage_nms_iou_threshold
[ RUN      ] ModelBuilderTF2Test.test_invalid_model_config_proto
INFO:tensorflow:time(__main__.ModelBuilderTF2Test.test_invalid_model_config_proto): 0.0s
I0322 16:48:11.535965 140205126002496 test_util.py:2373] time(__main__.ModelBuilderTF2Test.test_invalid_model_config_proto): 0.0s
[       OK ] ModelBuilderTF2Test.test_invalid_model_config_proto
[ RUN      ] ModelBuilderTF2Test.test_invalid_second_stage_batch_size
INFO:tensorflow:time(__main__.ModelBuilderTF2Test.test_invalid_second_stage_batch_size): 0.0s
I0322 16:48:11.539124 140205126002496 test_util.py:2373] time(__main__.ModelBuilderTF2Test.test_invalid_second_stage_batch_size): 0.0s
[       OK ] ModelBuilderTF2Test.test_invalid_second_stage_batch_size
[ RUN      ] ModelBuilderTF2Test.test_session
[  SKIPPED ] ModelBuilderTF2Test.test_session
[ RUN      ] ModelBuilderTF2Test.test_unknown_faster_rcnn_feature_extractor
INFO:tensorflow:time(__main__.ModelBuilderTF2Test.test_unknown_faster_rcnn_feature_extractor): 0.0s
I0322 16:48:11.542018 140205126002496 test_util.py:2373] time(__main__.ModelBuilderTF2Test.test_unknown_faster_rcnn_feature_extractor): 0.0s
[       OK ] ModelBuilderTF2Test.test_unknown_faster_rcnn_feature_extractor
[ RUN      ] ModelBuilderTF2Test.test_unknown_meta_architecture
INFO:tensorflow:time(__main__.ModelBuilderTF2Test.test_unknown_meta_architecture): 0.0s
I0322 16:48:11.543226 140205126002496 test_util.py:2373] time(__main__.ModelBuilderTF2Test.test_unknown_meta_architecture): 0.0s
[       OK ] ModelBuilderTF2Test.test_unknown_meta_architecture
[ RUN      ] ModelBuilderTF2Test.test_unknown_ssd_feature_extractor
INFO:tensorflow:time(__main__.ModelBuilderTF2Test.test_unknown_ssd_feature_extractor): 0.0s
I0322 16:48:11.545147 140205126002496 test_util.py:2373] time(__main__.ModelBuilderTF2Test.test_unknown_ssd_feature_extractor): 0.0s
[       OK ] ModelBuilderTF2Test.test_unknown_ssd_feature_extractor
----------------------------------------------------------------------
Ran 24 tests in 42.982s
OK (skipped=1)

看到结果为OK,则表示安装成功，接下来就可以开始物体检测之旅了。

• 《物体检测快速入门系列》快速导航：

• 物体检测快速入门系列(1)-基于Tensorflow2.x Object Detection API构建自定义物体检测器
• 物体检测快速入门系列(2)-Windows部署GPU深度学习开发环境
• 物体检测快速入门系列(3)-Windows部署Docker GPU深度学习开发环境
• 物体检测快速入门系列(4)-TensorFlow 2.x Object Detection API快速安装手册

一文掌握物体检测库TensorFlow 2.x Object Detection安装

1. 概述

2. 预置条件

3. 安装步骤

3.1 Docker环境

3.1.1 启动docker

3.1.2 启动容器

3.1.3 使用vscode访问docker container

3.1.4 更换docker容器ubuntu系统的安装源为国内源

3.1.5 验证GPU是否加载成功（在电脑有Nvidia显卡的情况下）

3.2 Windows开发环境

3.3 下载tensorflow object detection api项目源码

3.4 安装配置protobuf

3.5 将proto后缀文件转换为python可识别格式

3.6 安装coco api

3.7 安装object detection api

热门文章

最新文章

相关课程

相关电子书

相关实验场景

热门

活动广场

任务中心

开发者评测

高校计划

乘风者计划

训练营

阿里云MVP

话题

直播

下载

镜像站

技术资料

插件

一文掌握物体检测库TensorFlow 2.x Object Detection安装

1. 概述

2. 预置条件

3. 安装步骤

3.1 Docker环境

3.1.1 启动docker

3.1.2 启动容器

3.1.3 使用vscode访问docker container

3.1.4 更换docker容器ubuntu系统的安装源为国内源

3.1.5 验证GPU是否加载成功（在电脑有Nvidia显卡的情况下）

3.2 Windows开发环境

3.3 下载tensorflow object detection api项目源码

3.4 安装配置protobuf

3.5 将proto后缀文件转换为python可识别格式

3.6 安装coco api

3.7 安装object detection api

热门文章

最新文章

相关课程

相关电子书

相关实验场景