安装Delf
Tensorflow
conda create -n tf python=3.6
conda activate tf
conda install tensorflow-gpu=1.14
使用conda安装的tf1.14后在要运行的代码加上如下片段,不然会卷积错误异常
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
sess = tf.Session(config=config)
Protobuf
The DELF library uses protobuf (the python version) to configure feature extraction and its format. You will need the protoc compiler, version >= 3.3. The easiest way to get it is to download directly. For Linux, this can be done as (see here for other platforms):
wget https://github.com/google/protobuf/releases/download/v3.3.0/protoc-3.3.0-linux-x86_64.zip
unzip protoc-3.3.0-linux-x86_64.zip
PATH_TO_PROTOC=`pwd`
Python 依赖
pip install matplotlib numpy scikit-image scipy
tensorflow/models
注意 :object_detection library requires you to add tensorflow/models/research/ to your PYTHONPATH
git clone https://github.com/tensorflow/models
# First, install slim's "nets" package.
cd models/research/slim/
pip install -e .
# Second, setup the object_detection module by editing PYTHONPATH.
cd ..
# From tensorflow/models/research/
export PYTHONPATH=$PYTHONPATH:`pwd`
在这一步配置环境变量可以
sudo vim /etc/profile
export PYTHONPATH="$PYTHONPATH:{yourpath}"
Then, compile DELF’s protobufs. Use PATH_TO_PROTOC
as the directory where you downloaded the protoc compiler.
# From tensorflow/models/research/delf/
${PATH_TO_PROTOC?}/bin/protoc delf/protos/*.proto --python_out=.
Finally, install the DELF package.
# From tensorflow/models/research/delf/
pip install -e . # Install "delf" package.
# At this point, running
python -c 'import delf'
最后会返回一些type异常,不用管他,安装结束。