论文:https://arxiv.org/pdf/1802.02611.pdf
源码:https://github.com/tensorflow/models/tree/master/research/deeplab
环境:ubuntu16.04 + cuda9.0 + cudnn7.6 + tensorflow-gpu1.11
cuda9.0 + cudnn7.6安装配置可参照:https://blog.csdn.net/wgshun616/article/details/81019182
多个版本cuda安装切换可参照:https://blog.csdn.net/yinxingtianxia/article/details/80462892
// 查看GPU型号
lspci | grep -i nvidia
// 查看NVIDIA驱动版本
sudo dpkg --list | grep nvidia-*
// 查看CUDA版本,需一致
cat /usr/local/cuda/version.txt
nvcc -V
// 查看CUDNN版本
cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2
deeplabv3+依赖库安装参照:https://github.com/tensorflow/models/blob/master/research/deeplab/g3doc/installation.md
安装tensorflow-gpu时输入:
pip install tensorflow-gpu==1.11
记得在gedit ~/.bashrc配置文件中加入:
export PYTHONPATH=$PYTHONPATH:`pwd`:`pwd`/slim
运行网上数据集过程请参照:https://github.com/tensorflow/models/blob/master/research/deeplab/g3doc/pascal.md
运行自己数据集:
1. 建立数据集
先在datasets目录下建立如下目录结构:
+ pascal_voc_seg
+ VOCdevkit
+ VOC2012
+ JPEGImages
+ SegmentationClass
+ ImageSets
+ Mai