I was a junior student holding visual studio windows system from writing on a blog long gone, when registered in OPENCV and so complex components below shivering, believe it worried hpp not find it, no more dependent libraries or install a new system to play the whole thing and then collapse. (I believe this is the voice of the majority of their computer with windows developer of it)
After graduate student came into contact with the neural network, your computer must have a laboratory can not be used to provide a server, doing development in linux above, from the very beginning will pip install tensorflow, find dependencies around to the back finally learned to walk a lot conda create detour, go now can be considered to own a record of it, and preach it to the people who see blog Anaconda Dafa is good! .jpg
Anaconda in the end is what?
This problem is actually a lot of blog has been said, I am also not come up with very serious for him to be defined, for me is a python environment management software, he encapsulates the establishment of various dependencies under python when the user will need to use retrieved from the library anaconda out anaconda installed in the specified environment, with the recommended Tsinghua mirror source ( https://mirror.tuna.tsinghua.edu.cn/help/anaconda/ ) use together.
Anaconda solve the problem:
- Pip caused conflict between different python bag, conda in the installation package will detect the existing version and the version to be installed matches, as well as updated versions of packages associated with other existing packages will result in a conflict, The pip will not.
- Use Anaconda can create multiple environments on one system, and can only be modified pip on the native system. If we duplicate code is a requirement pytorch = 0.4.0, another requirement pytorch = 1.0.1, then simply use pip can choose to delete the existing pytorch to install another, or to re-open a server. But all I want to use Anaconda .jpg, use the source activate [name] command environment can easily switch between different environments.
- Depending on the version of python, pip update may introduce problems, conda command is more stable than the pip ( which is what I like one of the main conda of using pip in python2.7, I met only once prompted me to update pip after the update because the update files __main__ lead to pip other parameters need to be updated and the problem is not working, this time requires a rollback pip pip version or modify the parameters of the corresponding python file. when a system python2.7 and python3 when used with .6 that I feel really cool more acid = =
Overall, a good network or project development environment supports the need for a stable and manageable, and Anaconda It provides such a feature.
And very easy to install! So why not invincible Anaconda it!
Anaconda installation (Linux)
Anaconda in the following linux installation is very simple, basically sh download it to change to change the border variable ok fine somehow, just to summarize key steps.
1. Download Anaconda_xxxxxx.sh , I'm using Anaconda3. It can be downloaded directly from Tsinghua mirror, but updated version may be slightly delayed compared to the official website. Download Link: https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/
Search can download the latest date.
2. Use the command to install Anaconda, yes anyway, go on the right, in the process, he will ask you where to install Anaconda home, usually the default is in / Home / comptername / Anaconda3 , if needed after the change of their own to remember to change the path.
sh Anaconda3-2019.07-Linux-x86_64.sh
3. Then Anaconda has been installed, but when using conda command will tell you can not find the command, it is because there is no written to the environment variable.
Change the / etc / profile file (using vi vim gedit or subl own pleasure), which joined in:
#Anaconda
export PATH=$PATH:/home/computername/anaconda3/bin
4.Anaconda can officially use to see if there is a problem is not the path step 3 did not write right.
Use Anaconda configuration Pytorch_yolo3 environment
Code GitHub: https://github.com/azikk/PyTorch-YOLOv3
1. Create a new environment, the name is called yolo3, using python 3.6 version, and enter
conda create -n yolo3 python=3.6
source activate yolo3
2.安装pytorch，这里要求pytorch版本是1.0以上，我的电脑cuda版本是9.0。使用1.0.1是因为清华源中没找到更高版本的，如果制定更高版本就会去官方源下载了，太麻烦，就直接使用1.0.1
conda install pytorch=1.0.1 torchvision cudatoolkit=9.0
3.安装requirement.txt中写出的其他依赖，其中tqdm的包在官方源和清华源中都没有，要使用-c conda-forge从该仓库中下载。
conda install -c conda-forge tqdm
4.安装所有依赖后，就可以使用代码了。