tf
阅读和验证该文章前：请先搭建tensorflow环境：
tensorflow环境搭建
TensorFlow对象检测API是一个建立在TensorFlow之上的开源框架，可以轻松构建，训练和部署对象检测模型。 github地址
下载TensorFlow Model模型库
TensorFlow 模型库包含了很多开源的模型，包括图像分类、检测、自然语言处理NLP、视频预测、图像理解等等，我们要学习的对象检测ＡＰＩ也包括在这里面；
首先，在Home目录创建tensorflow, 将代码clone到本地
git clone https://github.com/tensorflow/models.git
使用以上命令，直接拉去的是tensorflow的master分支，也就是最新的分支代码
注意：你要使用与你的tensorflow版本对应的版本，例如你的tensorflow是1，那么你应该modle-r1.xx版本；如果你是使用的是tensorflow2，你应该使用modle-2.xx版本;所以，你应该根据你的tensorflow的版本，选择合适的物体检测API分支
(base) jiadongfeng:~/tensorflow/models/research$ git branch -a
* master
...
remotes/origin/guptapriya-patch-6
...
...
remotes/origin/r1.10.0
remotes/origin/r1.11
remotes/origin/r1.12.0
remotes/origin/r1.13.0
remotes/origin/r1.4.0
remotes/origin/r1.5
remotes/origin/r1.6.0
remotes/origin/r1.7.0
remotes/origin/r1.8.0
remotes/origin/r1.8.1
remotes/origin/r1.9.0
remotes/origin/r2.1.0
remotes/origin/r2.1_model_reference
...
因为我们使用的tesorflow版本是1.14.0，因此model branch我们使用remotes/origin/r1.13.0
//你的tensorflow必须是v1.9以上版本,以下是物体检测api中做的限制
if StrictVersion(tf.__version__) < StrictVersion('1.9.0'):
raise ImportError('Please upgrade your TensorFlow installation to v1.9.* or later!')
Protobuf 编译
Tensorflow Object Detection API 用 Protobufs 来配置模型和训练参数. 在用这个框架之前,必须先编译Protobuf 库，切换到这个目录下： tensorflow/models/research/
cd ~/tensorflow/models/research/
执行：
protoc object_detection/protos/*.proto --python_out=.
添加环境变量 PYTHONPATH
- 方法1：
tensorflow/models/research/ 和 slim 目录 需要添加到PYTHONPATH环境变量中. 从终端中，切换到tensorflow/models/research/目录，执行:
# From tensorflow/models/research/
export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim
注意: 这条命令在新打开的终端中需要重新执行一次才会在新终端中生效;
- 方法2
如果不想那么麻烦，就用下面的命令编辑 ~/.bashrc 文件，把上面的语句添加到末尾.
gedit ~/.bashrc
- 方法3
也可以在anaconda安装目录下的site-packages文件夹下建立一个tensorflow_model.pth文件，文件内容写models的路径，自己slim的路径，博主的路径是：
/home/jiadongfeng/tensorflow/models/research
/home/jiadongfeng/tensorflow/models/research/slim
测试model
执行，得到ＯＫ则说明安装成功:
(base) jiadongfeng:~/tensorflow/models/research$ export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim
(base) jiadongfeng:~/tensorflow/models/research$ python object_detection/builders/model_builder_test.py
WARNING:tensorflow:
The TensorFlow contrib module will not be included in TensorFlow 2.0.
For more information, please see:
* https://github.com/tensorflow/community/blob/master/rfcs/20180907-contrib-sunset.md
* https://github.com/tensorflow/addons
* https://github.com/tensorflow/io (for I/O related ops)
If you depend on functionality not listed there, please file an issue.
Running tests under Python 2.7.16: /home/jiadongfeng/anaconda2/bin/python
[ RUN ] ModelBuilderTest.test_create_experimental_model
[ OK ] ModelBuilderTest.test_create_experimental_model
[ RUN ] ModelBuilderTest.test_create_faster_rcnn_model_from_config_with_example_miner
[ OK ] ModelBuilderTest.test_create_faster_rcnn_model_from_config_with_example_miner
[ RUN ] ModelBuilderTest.test_create_faster_rcnn_models_from_config_faster_rcnn_with_matmul
[ OK ] ModelBuilderTest.test_create_faster_rcnn_models_from_config_faster_rcnn_with_matmul
[ RUN ] ModelBuilderTest.test_create_faster_rcnn_models_from_config_faster_rcnn_without_matmul
[ OK ] ModelBuilderTest.test_create_faster_rcnn_models_from_config_faster_rcnn_without_matmul
[ RUN ] ModelBuilderTest.test_create_faster_rcnn_models_from_config_mask_rcnn_with_matmul
[ OK ] ModelBuilderTest.test_create_faster_rcnn_models_from_config_mask_rcnn_with_matmul
[ RUN ] ModelBuilderTest.test_create_faster_rcnn_models_from_config_mask_rcnn_without_matmul
[ OK ] ModelBuilderTest.test_create_faster_rcnn_models_from_config_mask_rcnn_without_matmul
[ RUN ] ModelBuilderTest.test_create_rfcn_model_from_config
[ OK ] ModelBuilderTest.test_create_rfcn_model_from_config
[ RUN ] ModelBuilderTest.test_create_ssd_fpn_model_from_config
[ OK ] ModelBuilderTest.test_create_ssd_fpn_model_from_config
[ RUN ] ModelBuilderTest.test_create_ssd_models_from_config
[ OK ] ModelBuilderTest.test_create_ssd_models_from_config
[ RUN ] ModelBuilderTest.test_invalid_faster_rcnn_batchnorm_update
[ OK ] ModelBuilderTest.test_invalid_faster_rcnn_batchnorm_update
[ RUN ] ModelBuilderTest.test_invalid_first_stage_nms_iou_threshold
[ OK ] ModelBuilderTest.test_invalid_first_stage_nms_iou_threshold
[ RUN ] ModelBuilderTest.test_invalid_model_config_proto
[ OK ] ModelBuilderTest.test_invalid_model_config_proto
[ RUN ] ModelBuilderTest.test_invalid_second_stage_batch_size
[ OK ] ModelBuilderTest.test_invalid_second_stage_batch_size
[ RUN ] ModelBuilderTest.test_session
[ SKIPPED ] ModelBuilderTest.test_session
[ RUN ] ModelBuilderTest.test_unknown_faster_rcnn_feature_extractor
[ OK ] ModelBuilderTest.test_unknown_faster_rcnn_feature_extractor
[ RUN ] ModelBuilderTest.test_unknown_meta_architecture
[ OK ] ModelBuilderTest.test_unknown_meta_architecture
[ RUN ] ModelBuilderTest.test_unknown_ssd_feature_extractor
[ OK ] ModelBuilderTest.test_unknown_ssd_feature_extractor
----------------------------------------------------------------------
Ran 17 tests in 0.383s
OK (skipped=1)
运行demo
进入/tensorflow/models/research/object_detection目录，运行：
jupyter notebook
一定要在object_detection目录，因为代码的数据读取和保存都是在这个目录的。model中的测试代码是从当前目录读取数据的。当然，你也可以修改代码，重新指定目录读取数据
在jupyter中打开以下文件
~/tensorflow/models/research/object_detection/object_detection_tutorial.ipynb
在window系统中，可以直接打开以下网址查看：
按照步骤，获取测试demo代码
以上官方提供的demo，如果你完全按照代码拷贝，是运行不了的。需要稍微修改，以下为本地处理了各种报错的代码（代码较多，放到了单独的页面）
运行结果
image1_ret.png
image2_ret.png
报错处理
- TypeError: load() takes at least 3 arguments (1 given)
error locaton:
tf.saved_model.load(str(model_dir))
TypeError: load() takes at least 3 arguments (1
这个是tensorflow 版本和物体检测版本不对应引起的。
如果你使用tensorflow1,使用model的r1版本，使用git checkout直接拉取即可
(base) jiadongfeng:~/tensorflow/models/research$ git branch -a
* master
emotes/origin/r1.10.0
remotes/origin/r1.11
remotes/origin/r1.12.0
remotes/origin/r1.13.0
remotes/origin/r1.4.0
remotes/origin/r1.5
remotes/origin/r1.6.0
remotes/origin/r1.7.0
remotes/origin/r1.8.0
remotes/origin/r1.8.1
remotes/origin/r1.9.0
remotes/origin/r2.1.0
remotes/origin/r2.1_model_reference
(base) jiadongfeng:~/tensorflow/models/research$ git checkout remotes/origin/r1.13.0
不要使用model的最新版本或者2.0版本，因为model的2.0版本有使用tensorflow1的接口，无论如果都编译不过
当我们将tensorflow更新到v2时，提示tf.contrib 被 deprecated 。也就是说在最新的支持tensorflow2的物体检测api中，又使用了tensorflow1的接口。
/home/jiadongfeng/anaconda2/lib/python2.7/site-packages/ipykernel_launcher.py:10: UserWarning:
This call to matplotlib.use() has no effect because the backend has already
been chosen; matplotlib.use() must be called before pylab, matplotlib.pyplot,
or matplotlib.backends is imported for the first time.
解决方案：在头部新增import matplotlib和matplotlib.use('Agg')
#####import
import numpy as np
import os
import six.moves.urllib as urllib
import sys
import tarfile
import tensorflow as tf
import zipfile
#顺序不能反
import matplotlib
matplotlib.use('Agg')
- 运行物体检测Demo，无运行报错，但是不显示图片
参考：
https://blog.csdn.net/weixin_43345331/article/details/87924551
vis_util.visualize_boxes_and_labels_on_image_array(
image_np,
output_dict['detection_boxes'],
output_dict['detection_classes'],
output_dict['detection_scores'],
category_index,
instance_masks=output_dict.get('detection_masks'),
use_normalized_coordinates=True,
line_thickness=8)
plt.figure(figsize=IMAGE_SIZE)
plt.imshow(image_np)
plt.imshow不显示图片，参照晚上调用plt.show()也不行，也没有报错，只能通过以下方式将处理结果保存下来查看
plt.figure(figsize=IMAGE_SIZE)
plt.imshow(image_np)
pylab.show()
//在当前目录保存处理后的图片
plt.savefig(image_path.split('.')[0]+'_ret')
网友评论