Buy May 4 on the cloud server, toss for nearly two weeks. Experienced tensorflow, pytorch, caffe configuration, run faster-rcnn, ssd-tensorflow, ssd-caffe, Ga-RPN, RefineDet, DetNet, RFB-Net codes and other papers. Linux operating system, familiar point of the environment with not so afraid of. However, it must be promptly recorded, otherwise white buy.
Cloud Server Configuration: Tesla P4 GPU
hardware problems:
1. How to select and configure cloud servers? See the Favorites tab.
The software problem:
the graphics driver version Installation nvidia 1. Check command
https://blog.csdn.net/jizhidexiaoming/article/details/79549837
input command as above, show:
graphics driver version: 384.111
Many places will use graphics driver version:
lspci | grep -i nvidia
cat / proc / driver / nvidia / version
nvidia-smi
2.Ubuntu installation cuda
because the cloud server can be installed directly cuda9.0, so I did not install their own, so also do save me a lot of things.
But still retain the link to prevent future study, work use.
https://blog.csdn.net/wanzhen4330/article/details/81699769
3.Ubuntu installation cudnn
I use this link: https://blog.csdn.net/weixin_41864878/article/details/79621210
4. Review and cudnn cuda version:
https://blog.csdn.net/baidu_32936911/article/details/79774289
Many places will need to see cuda and cudnn version number:
View cuda version: cat /usr/local/cuda/version.txt
my cuda is 9.0.176
View cudnn version: cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2
My cudnn 7.0 .5
5. Install the anaconda (python version 3.6):
Currently the official website is the latest version of the anaconda, the corresponding python version 3.7 if you want to download 3.6, refer to the following picture:
but now can not open this image stood open-source software, Tsinghua University .
6. Install tensorflow-gpu version: pip install tensorflow-gpu == 1.6
See blog Faster RCNN code to run
7. dataset question:
VOC2007, VOC2012, COCO: save the download link in the bookmarks inside. Thunder download is also backed up. May be first compressed to the server, then decompressed data decompressed command set. After downloading something you can use Thunder download, it will be very fast.
8. install different versions pytorch: See ga-rpn blog code running, select a different operating system, version pytorch Python official website, pip conda installation or mounting, enter the appropriate command. But I want to install pytorch previous version, see the official website of other commands. Conda with pip command or commands are possible, see the specific DetNet before and RFB-Net blog.
9. Install caffe: see previous blog it.
10. Pending