论文地址:《ReSeg: A Recurrent Neural Network-based Model for Semantic Segmentation》
代码地址1:PyTorch Implementation of ReSeg
代码地址2:Theano Implementation of ReSeg
PyTorch实现的安装运行
安装ReSeg前需要下载:
Pascal VOC 2010 Dataset:一个用于分割的数据集
Pascal-Part Annotations:这里需要有额外的步骤才可以下载成功。
标注数据集的下载方式(参考):
1.sudo apt-get install ctorrent
2.下载标注数据集的.torrent文件
3.ctorrent trainval.tar.gz-f86670296bff85bcdffea6c4fc2e791446f9fb5e.torrent
出现如下画面
然后就下载好了trainval.tar.gz 即标注数据集
1.下载github源码
git clone --recursive https://github.com/Wizaron/reseg-pytorch.git
(我的安放位置为ubuntu系统的/home/xxx/Document下)
2.将下载好的数据集和标注放在reseg-pytorch/data/raw下并且提取文件(原先是tar文件和tar.gz文件)
3.cd reseg-pytorch/code/pytorch
4.下载anaconda(注意Anaconda和ROS 安装的时候可能出现冲突,会在执行python脚本的时候报错)
5.Create environment : conda env create -f pytorch_conda_environment.yml
第五步里面有一个cudatoolkit=8.0=3需要换源然后才可以找得到
(找channel里面pkg的全部版本号可以通过conda search pkg)
最后OK 的时候提示的是
To activate this environment, use
$ conda activate reseg-pytorch
现在的问题:标注集解压不了
Anaconda与ROS共存问题
1.切换python版本(base)
临时:输入$ conda deactivate,(base)消失
永久:conda config --set auto_activate_base false
2.在anaconda版本的python中下载库
3.修改系统软链接
$ conda install setuptools
$ pip install -U rosdep rosinstall_generator wstool rosinstall six vcstools
安装教程:https://www.cnblogs.com/shfanzie/p/10276000.html
WARNING:
You currently have a PYTHONPATH environment variable set. This may cause
unexpected behavior when running the Python interpreter in Anaconda2.
For best results, please verify that your PYTHONPATH only points to
directories of packages that are compatible with the Python interpreter
in Anaconda2: /home/asber/Document/anaconda2
Do you wish the installer to initialize Anaconda2
by running conda init? [yes|no]yes
==> For changes to take effect, close and re-open your current shell. <==
If you'd prefer that conda's base environment not be activated on startup,
set the auto_activate_base parameter to false:
conda config --set auto_activate_base false
Thank you for installing Anaconda2!
发现cm之后暂时没有报错(因为不知道为什么cm的对象一直只有ros_caffe)
换源:https://mirrors.tuna.tsinghua.edu.cn/help/anaconda/
参考:https://www.zhihu.com/question/50963250
https://blog.csdn.net/m0_37579176/article/details/86536411
Theano实现的安装运行
需要python版本大于3.5
conda create -n py3.6 python=3.6
1.theano下载
在anaconda的base中pip install Theano
Successfully installed Theano-1.0.4
2.pip install lasagne matplotlib Pillow progressbar2 pydot-ng retrying scikit-image scikit-learn tabulate
3.Download the CamVid dataset
Download the CamVid dataset from :http://mi.eng.cam.ac.uk/research/projects/VideoRec/CamVid/
Download video from: ftp://svr-ftp.eng.cam.ac.uk/pub/eccv 直接下载avi
Labeled Images 那里可以下载labeled images
存放在./datasets/camvid/下
You can change this path modifying camvid.py if you want.
4.Download the VGG-16 weights
下载不到
论文解读:
参考:
1.【阅读笔记】《ReSeg: A Recurrent Neural Network-based Model for Semantic Segmentation》
2.【图像分割模型】用BRNN做分割—ReSeg
一开始想要实现reseg是因为在<基于深度学习的语义分割问题研究综述>中看到Reseg是接受depth的,但是看了论文的确是没有的,大失所望。
终止。