Work | Date |
---|---|
Install dependency Libraries | 2019.05.08–05.09 |
Learn the Python(The chapters before GUI) 算法笔记:Prim、Kruskal、拓扑排序、关键路径 |
2019.05.10–05.17 |
答辩PPT+答辩+颓废 | 2019.05.18–05.31 |
算法笔记:树与二叉树、二叉查找树BST、平衡二叉树、堆、哈夫曼树 | 2019.06.01–06.05 |
重庆旅游+回家+回校拿证+毕业典礼 | 2019.06.06–06.19 |
长沙租房+领通知书+实验室占坑 | 2019.06.20–06.21 |
1.算法:树、查找树、平很二叉树、图论、动态规划、并查集、堆、哈夫曼树、字符串篇 2.Python学习:Numpy Matplotlib 3.目标检测(Object Detection): - 基于图的图像分割(Graph-Based Image Segmentation): PDF - 图像梯度 - Sobel算子: PDF - 梯度直方图:Article Link - 选择性搜索(Selective Search):Article Link - 支持向量机SVM(Support Vector Machine) 4.Linux学习 5.Conda 6.Caffe搭建(失败) |
2019.06.22–07.01 |
1.动手学深度学习—-CV部分完成(Tag: DL) 2.三个项目实验 3.五篇论文精读(见Tag: papers) 4.八月份开始《算法笔记实战》截止10月6号完成PAT甲级90题(Tag: Algorithm/PAT甲级) |
2019.07.02-10.06 |
1.SSD: Single Shot MultiBox Detector:https://blog.csdn.net/u010167269/article/details/52563573
2.Object Detection:https://handong1587.github.io/deep_learning/2015/10/09/object-detection.html
3.RCNN算法详解:https://blog.csdn.net/shenxiaolu1984/article/details/51066975
4.Fast RCNN算法详解:https://blog.csdn.net/shenxiaolu1984/article/details/51036677
5.Faster RCNN算法详解:https://blog.csdn.net/shenxiaolu1984/article/details/51152614
6.R-FCN:基于区域的全卷积网络来检测物体:https://blog.csdn.net/shadow_guo/article/details/51767036
7.Selective Search for Object Recognition(IJCV, 2013):https://www.cnblogs.com/zhao441354231/p/5941190.html
8.Relation Networks for Object Detection:https://blog.csdn.net/u014380165/article/details/80779432
9.Single-Shot Refinement Neural Network for Object Detection:https://blog.csdn.net/u014380165/article/details/79502308
10.基于图的图像分割(Graph-Based Image Segmentation):https://blog.csdn.net/guoyunfei20/article/details/78727972
11.卷积神经网络笔记:https://zhuanlan.zhihu.com/p/22038289?refer=intelligentunit
12.Backpropagation In Convolutional Neural Networks:https://www.jefkine.com/general/2016/09/05/backpropagation-in-convolutional-neural-networks/
13.Neural Networks and Deep Learning小节:https://www.cnblogs.com/zyly/p/8638856.html
14.卷积神经网络之AlexNet:https://www.cnblogs.com/wangguchangqing/p/10333370.html
15.AlexNet论文(ImageNet Classification with Deep Convolutional Neural Networks)(译):https://blog.csdn.net/hongbin_xu/article/details/80271291
16.受限波尔兹曼网络及及代码实现:https://www.cnblogs.com/zyly/p/9055616.html
17.自编码网络介绍及代码实现:https://www.cnblogs.com/zyly/p/9072595.html
18.SkipNet: Learning Dynamic Routing in Convolutional Networks:http://openaccess.thecvf.com/content_ECCV_2018/html/Xin_Wang_SkipNet_Learning_Dynamic_ECCV_2018_paper.html
19.Dilated Residual Networks:http://openaccess.thecvf.com/content_cvpr_2017/html/Yu_Dilated_Residual_Networks_CVPR_2017_paper.html
20.Effective Use of Synthetic Data for Urban Scene Semantic Segmentation:https://link.springer.com/chapter/10.1007/978-3-030-01216-8_6
21.DetNet: Design Backbone for Object Detection:https://blog.csdn.net/ChuiGeDaQiQiu/article/details/82056223
22.DetNet代码解析:https://blog.csdn.net/qq_41438431/article/details/85233248
23.ECCV 2018论文节选:https://blog.csdn.net/amusi1994/article/details/82356417
24.计算机视觉论文速递:https://github.com/amusi/daily-paper-computer-vision/blob/master/2019/03/12.md
25.【计算机视觉论文速递】2019-01-01~01-04:https://github.com/amusi/daily-paper-computer-vision/blob/master/2019/01/01-04.md
26.【计算机视觉论文速递】2018-12-31:https://github.com/amusi/daily-paper-computer-vision/blob/master/2018/12/31.md
27.Summary: 有4篇论文速递信息:https://github.com/amusi/daily-paper-computer-vision/blob/master/2018/05/16.md
28.Diverse Image-to-Image Translation via Disentangled Representations:https://www.cnblogs.com/SuperLab/p/9837664.html
29.视频物体检测(VID) Deep Feature Flow for Video Recognition:https://blog.csdn.net/u012426298/article/details/80487505
30.FlowNet 2.0: Evolution of Optical Flow Estimation with Deep Networks:https://blog.csdn.net/bea_tree/article/details/67049373
31.Object Detection:https://handong1587.github.io/deep_learning/2015/10/09/object-detection.html
Anaconda安装与配置
Anaconda指的是一个开源Python发行版本,其包含了Conda、Python等180多个科学包及其依赖项。
1 | wget https://mirrors.tuna.tsinghua.edu.cn/anaconda/archive/Anaconda3-5.3.1-Linux-x86_64.sh |
Conda使用
方便管理环境
国内镜像
方便包和依赖库的快速安装
Pip镜像
(a)Linux下,修改 ~/.pip/pip.conf (没有就创建一个文件夹及文件。文件夹要加“.”,表示是隐藏文件夹)
内容如下:
1 | [global] |
(b) windows下,直接在user目录中创建一个pip目录,如:C:\Users\xx\pip,然后新建文件pip.ini,即 %HOMEPATH%\pip\pip.ini,在pip.ini文件中输入以下内容:
1 | [global] |
Conda镜像
1 | conda config --add channels https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/ |
Apt镜像
/etc/apt/sources.list.d/下所有文件的cuda和navida源注释
然后修改sources.list
1 | #deb http://mirrors.tuna.tsinghua.edu.cn/ubuntu/ xenial main restricted universe multiverse |
最后apt-get update即可
vim配置
1 | set nu |
Tensorflow安装
Google旗下开源深度学习框架
1 | pip install tensorflow |
服务器传文件
1 | scp -P port_number [-R] file_path/dir_path username@ip:dst_dir_path |
多GPU训练踩坑
注意多GPU训练的时候,系统默认在可见GPU索引0位置进行相关参数更新和梯度计算,或许是bug吧,暂时还没找到具体原因,pytorch和mxnet都是,只能暂时通过设置可见设备进行避免
1 | os.environ['CUDA_VISIBLE_DEVICES']='gpu_idx,...' |
GPU显存隐藏进程占用
1 | nvidia-smi |
Linux下查看目录文件大小
1 | du -h --max-depth=0 * |
Linux下查看目录/文件个数
1 | ls -l |grep "^-"|wc -l # 当前目录下文件个数 |
CUDA与CUDNN安装
1 | conda install cudatoolkit=10.1 -c https://mirrors.tuna.tsinghua.edu.cn/anaconda/pkgs/free/linux-64 |
Linux修改root密码
1 | passwd 用户名 |
安装事项
1 | bash Anaconda3-xxx.sh |
注意框架冲突问题,遇到过mxnet装了后再装tensorflow报错,有个wrapt错误,网上方法试了后,能装了,但是我的mxnet程序就出问题了,还是老老实实分不同conda环境好了
代理设置
网络设置手动代理 设置ip:127.0.0.1, port:7890
然后ghelper下好clash配置好订阅就行了。