记录CenterNet代码编译成功运行

一开始没有安装anaconda，直接用的python,安包也是各种pip,遇到了非常多的bug,改到怀疑人生，强烈建议先安装anaconda,建立python3.6的虚拟环境，再进行网络的安装编译运行。

首先要检查自己的环境，环境不对的话，后期会出现奇奇怪怪的问题。程序所需环境为：cuda9.0 ,  torch0.4.1,cudnn7.0,我是在Ubuntu16.04中运行的。

1.创建python3.6的环境，并激活环境，可以在终端输入python，查看自己的版本。

 

 

 

 

 2.安装torch，或者输入python，import torch，print（torch__version__）可以查看torch的版本，没有的话用conda进行安装。

 

 

 

 运行nvcc -V命令，查看自己的cuda版本是否为9.0.

cuda9.0linux安装可以参考https://blog.csdn.net/qq_25241325/article/details/90753830

如果官网安装太慢的话，可以加清华源的镜像，去掉-c pytorch 进行安装。

3.把官网上的代码git下来，按照CenterNet官网里的INSTALL,进行requirements.txt的安装和COCOAPI的安装（注意：先安装requirements.txt,因为COCOAPI需要里面的CPython）。

 

 

 

 API需要make,按照官网操作即可

 

 4.编译DCNv2，这一步如果出现一大串的问题，基本都是cuda路径没设置对的原因。

5.编译成功后，即可运行demo.py,会出现以下报错：

 

 解决方法：

 

 还会出现错误：

 

 解决方法：（要提前mkdir output）

 

 最后就可以成功运行demo.py啦！！！放个效果图，识别的还是可以的

 

 

 新手一枚，之前因为环境的原因，改了许多奇怪的bug,这里记录一下，希望能够帮到小伙伴们。

————————————————————————————————————————————————————————————————————————————————————————————————————————————————

CenterNet官网里的INSTALL文档：

# Installation

The code was tested on Ubuntu 16.04, with [Anaconda](https://www.anaconda.com/download) Python 3.6 and [PyTorch]((http://pytorch.org/)) v0.4.1. NVIDIA GPUs are needed for both training and testing.
After install Anaconda:

0. [Optional but recommended] create a new conda environment.

~~~
conda create --name CenterNet python=3.6
~~~
And activate the environment.

~~~
conda activate CenterNet
~~~

1. Install pytorch0.4.1:

~~~
conda install pytorch=0.4.1 torchvision -c pytorch
~~~

And disable cudnn batch normalization(Due to [this issue](https://github.com/xingyizhou/pytorch-pose-hg-3d/issues/16)).

~~~
# PYTORCH=/path/to/pytorch # usually ~/anaconda3/envs/CenterNet/lib/python3.6/site-packages/
# for pytorch v0.4.0
sed -i "1194s/torch\.backends\.cudnn\.enabled/False/g" ${PYTORCH}/torch/nn/functional.py
# for pytorch v0.4.1
sed -i "1254s/torch\.backends\.cudnn\.enabled/False/g" ${PYTORCH}/torch/nn/functional.py
~~~

For other pytorch version, you can manually open `torch/nn/functional.py` and find the line with `torch.batch_norm` and replace the `torch.backends.cudnn.enabled` with `False`. We observed slight worse training results without doing so.

2. Install [COCOAPI](https://github.com/cocodataset/cocoapi):

~~~
# COCOAPI=/path/to/clone/cocoapi
git clone https://github.com/cocodataset/cocoapi.git $COCOAPI
cd $COCOAPI/PythonAPI
make
python setup.py install --user
~~~

3. Clone this repo:

~~~
CenterNet_ROOT=/path/to/clone/CenterNet
git clone https://github.com/xingyizhou/CenterNet $CenterNet_ROOT
~~~

4. Install the requirements

~~~
pip install -r requirements.txt
~~~


5. Compile deformable convolutional (from [DCNv2](https://github.com/CharlesShang/DCNv2/tree/pytorch_0.4)).

~~~
cd $CenterNet_ROOT/src/lib/models/networks/DCNv2
./make.sh
~~~
6. [Optional, only required if you are using extremenet or multi-scale testing] Compile NMS if your want to use multi-scale testing or test ExtremeNet.

~~~
cd $CenterNet_ROOT/src/lib/external
make
~~~

7. Download pertained models for [detection]() or [pose estimation]() and move them to `$CenterNet_ROOT/models/`. More models can be found in [Model zoo](MODEL_ZOO.md).

——————————————————————————————————————————————————————————————————————————————————————————————————————————————————————

参考博客：https://blog.csdn.net/m0_37798080/article/details/101196884#demopy_52

                 https://blog.csdn.net/mary_0830/article/details/103333639#comments_12208483

                 https://blog.csdn.net/weixin_38715903/article/details/98039181