Introduction

BVLC/caffe: https://github.com/BVLC/caffe.git
SegNet-Tutorial: https://github.com/alexgkendall/SegNet-Tutorial.git Rename to SegNet
vgg-caffe: https://github.com/davidgengenbach/vgg-caffe.git
caffe-segnet-cudnn5: https://github.com/TimoSaemann/caffe-segnet-cudnn5.git Located at SegNet/caffe-segnet
ENET: A Deep Neural Network Architecture for Real-Time Semantic Segmentation

System Environment

Linux ubuntu 16.04

$ uname -a
Linux yubao-Z370M-S01 4.15.0-51-generic #55~16.04.1-Ubuntu SMP Thu May 16 09:24:37 UTC 2019 x86_64 x86_64 x86_64 GNU/Linux

Cuda

10.0

cuDNN

v7.6.1+cuda 10.0

Python

2.7

External Files

Weights

VGG 16-layer mode: http://www.robots.ox.ac.uk/~vgg/research/very_deep/

Python pacages

Refer SegNet/requirements.txt

Reference

Segnet tutorial: http://mi.eng.cam.ac.uk/projects/segnet/tutorial.html
Install caffe: Caffe-Installation Guide
Install cuDNN: Installing cuDNN on Linux
Download pretrained model: https://github.com/alexgkendall/SegNet-Tutorial/blob/master/Example_Models/segnet_model_zoo.md
guanfuchen/semseg 常用的语义分割架构结构综述以及代码复现

Build caffe

Dependences

CUDA

library version 7+ and the latest driver version are recommended, but 6.* is fine too 5.5, and 5.0 are compatible but considered legacy

Check current cuda version:

nvidia-smi

cuDNN

cuDNN for GPU acceleration (v6)
download: https://developer.nvidia.com/rdp/cudnn-download
- Download cuDNN v7.6.1 (June 24, 2019), for CUDA 10.1
- Download cuDNN v7.6.1 (June 24, 2019), for CUDA 10.0
- Download cuDNN v7.6.1 (June 24, 2019), for CUDA 9.2
- Download cuDNN v7.6.1 (June 24, 2019), for CUDA 9.0

# cuDNN Runtime Library for Ubuntu16.04 (Deb)
wget -c https://developer.nvidia.com/compute/machine-learning/cudnn/secure/v7.6.1.34/prod/10.0_20190620/Ubuntu16_04-x64/libcudnn7_7.6.1.34-1%2Bcuda10.0_amd64.deb
# cuDNN Developer Library for Ubuntu16.04 (Deb)
wget -c 
https://developer.nvidia.com/compute/machine-learning/cudnn/secure/v7.6.1.34/prod/10.0_20190620/Ubuntu16_04-x64/libcudnn7-dev_7.6.1.34-1%2Bcuda10.0_amd64.deb
# cuDNN Code Samples and User Guide for Ubuntu16.04 (Deb)
https://developer.nvidia.com/compute/machine-learning/cudnn/secure/v7.6.1.34/prod/10.0_20190620/Ubuntu16_04-x64/libcudnn7-doc_7.6.1.34-1%2Bcuda10.0_amd64.deb

Remove previous version

sudo dpkg -r libcudnn7 libcudnn7-dev

Install using dpkg

sudo dpkg -i <new-cudnn-runtime>.deb
sudo dpkg -i <new-cudnn-dev>.deb
sudo ldconfig

My cuda version is 10.0 so I should install v7.61 for CUDA 10.0

avoid auto upgrade when run apt upgrade [very important]

sudo apt-mark hold libcudnn7 libcudnn7-dev

I’m using version 7 + cuda 10 Check current cudnn version

$ cat /usr/include/cudnn.h | grep CUDNN_MAJOR -A 2
#define CUDNN_MAJOR 7

compile caffe and install

cd caffe
mkdir build
cd build
cmake -DCMAKE_INSTALL_PREFIX=<INSTALL_PATH> ..
make all
make install
make runtest

Config python path

export PYTHONPATH=<CAFFE_PATH>/python:$PYTHONPATH

Install caffe using conda

I have not yet tried this method Create ‘caffe’ virtual environment

conda create -n caffe_gpu -c defaults python=3.6 caffe-gpu

Test python caffe

$ python
Python 2.7.12 (default, Nov 12 2018, 14:36:49)
[GCC 5.4.0 20160609] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import caffe
>>> exit()

SegNet: SegNet-Tutorial

Official: SegNet-Tutorial

Structure

$ tree -L 1
├── caffe-segnet
├── CamVid
├── docker
├── Example_Models
├── Models
├── README.md
├── requirement.txt
└── Scripts

Build caffe-segnet

mkdir build
cd build
cmake ..
make
make install

Getting Started with Live Demo

First open Scripts/webcam_demo.py and edit line 14 to match the path to your installation of SegNet.

python Scripts/webcam_demo.py --model Example_Models/segnet_model_driving_webdemo.prototxt --weights /Example_Models/segnet_weights_driving_webdemo.caffemodel --colours /Scripts/camvid12.png

Setting Up Caffe and the Dataset

Modify path of

CamVid/test.txt
CamVid/train.txt

vim:

:%s#SegNet#home/yubao/data/Project/semseg/SegNet-Tutorial-alexgkendall#g

Modify file path in:
- Models/segnet_train.prototxt

7   dense_image_data_param {
8     source: "/home/yubao/Project/semseg/SegNet-Tutorial-alexgkendall/CamVid/train.txt"  # Change this to the absolute      path to your data file
9     batch_size: 4               # Change this number to a batch size that will fit on your GPU

Models/segnet_inference.prototxt
Models/segnet_solver.prototxt

Create a folder to store your training weights and solver details with mkdir /SegNet/Models/Training

SegNet

Training CamVid using SegNet-basic

Dataset

CamVid dataset contains 367 training and 233 testing images of road scenes.

Training

./SegNet/caffe-segnet/build/tools/caffe train -gpu 0 -solver /SegNet/Models/segnet_solver.prototxt  # This will begin training SegNet on GPU 0
./SegNet/caffe-segnet/build/tools/caffe train -gpu 0 -solver ./SegNet/Models/segnet_basic_solver.prototxt  # This will begin training SegNet-Basic on GPU 0
./SegNet/caffe-segnet/build/tools/caffe train -gpu 0 -solver ./SegNet/Models/segnet_solver.prototxt -weights /SegNet/Models/VGG_ILSVRC_16_layers.caffemodel  # This will begin training SegNet on GPU 0 with a pretrained encoder

Testing

python /Segnet/Scripts/compute_bn_statistics.py /SegNet/Models/segnet_train.prototxt /SegNet/Models/Training/segnet_iter_10000.caffemodel /Segnet/Models/Inference/  # compute BN statistics for SegNet
python /Segnet/Scripts/compute_bn_statistics.py /SegNet/Models/segnet_basic_train.prototxt /SegNet/Models/Training/segnet_basic_iter_10000.caffemodel /Segnet/Models/Inference/  # compute BN statistics for SegNet-Basic

Result of basics

Training Pascal VOC

Dataset

Pascal VOC

Training

Testing

Result

I0630 21:12:15.934836  5739 sgd_solver.cpp:112] Iteration 599, lr = 0.0001
I0630 21:12:17.073256  5739 solver.cpp:239] Iteration 600 (0.878424 iter/s, 1.1384s/1 iters), loss = 1.95768
I0630 21:12:17.073312  5739 solver.cpp:258]     Train net output #0: loss = 1.8794 (* 1 = 1.8794 loss)
I0630 21:12:17.073334  5739 sgd_solver.cpp:112] Iteration 600, lr = 0.0001
itt:600 accuracy:0.9596
Baseline accuracy:0.9240

Possible Errors

ImportError: No module named caffe

t$ python Scripts/compute_bn_statistics.py Example_Models/segnet_pascal.prototxt Example_Models/segnet_pascal.caffemodel Models/Inference/
Traceback (most recent call last):
  File "Scripts/compute_bn_statistics.py", line 14, in <module>
    import caffe
ImportError: No module named caffe

Solution:

export PYTHONPATH=<CAFFE_PATH>/python:$PYTHONPATH
eg:
 export PYTHONPATH=/home/yubao/data/Project/semseg/SegNet/caffe-segnet/build/install/python:$PYTHONPATH

If still not work, check relative files. For example:

Scripts/compute_bn_statistics.py

9
10 caffe_root = '/home/yubao/data/Project/semseg/SegNet/caffe-segnet/build/install/python'          # Change this to the absolute     directory to SegNet Caffe

Cudnn error when compile caffe-segnet-cudnn5

Error message:

(caffe_gpu) yubao@yubao-Z370M-S01:~/data/Project/semseg/caffe-segnet-cudnn5-TimoSaemann/build$ make
Scanning dependencies of target proto
[  0%] Building CXX object src/caffe/CMakeFiles/proto.dir/__/__/include/caffe/proto/caffe.pb.cc.o
[  0%] Linking CXX static library ../../lib/libproto.a
[  1%] Built target proto
[  1%] Building NVCC (Device) object src/caffe/CMakeFiles/cuda_compile_1.dir/layers/cuda_compile_1_generated_absval_layer.cu.o
/home/yubao/data/Project/semseg/caffe-segnet-cudnn5-TimoSaemann/include/caffe/util/cudnn.hpp(112): error: too few arguments in function call

1 error detected in the compilation of "/tmp/tmpxft_000074ef_00000000-4_absval_layer.cpp4.ii".
CMake Error at cuda_compile_1_generated_absval_layer.cu.o.Release.cmake:279 (message):
  Error generating file
  /home/yubao/data/Project/semseg/caffe-segnet-cudnn5-TimoSaemann/build/src/caffe/CMakeFiles/cuda_compile_1.dir/layers/./cuda_compile_1_generated_absval_layer.cu.o


src/caffe/CMakeFiles/caffe.dir/build.make:499: recipe for target 'src/caffe/CMakeFiles/cuda_compile_1.dir/layers/cuda_compile_1_generated_absval_layer.cu.o' failed
make[2]: *** [src/caffe/CMakeFiles/cuda_compile_1.dir/layers/cuda_compile_1_generated_absval_layer.cu.o] Error 1
CMakeFiles/Makefile2:271: recipe for target 'src/caffe/CMakeFiles/caffe.dir/all' failed
make[1]: *** [src/caffe/CMakeFiles/caffe.dir/all] Error 2
Makefile:129: recipe for target 'all' failed
make: *** [all] Error 2

Solution:

It’s happening due to cudnn.hpp (Location: include/caffe/util/cudnn.hpp) . Update cudnn.hpp file. It is not considering the current cuDNN versions.

Update the cudnn.hpp from github repo, BVLC/caffe/include/caffe/util/cudnn.hpp Or, go to this link https://github.com/BVLC/caffe/blob/master/include/caffe/util/cudnn.hpp [Copy this and replace the old one]

Python Caffe: Python 2.7 or Python 3.3+, numpy (>= 1.7), boost-provided boost.python
MATLAB Caffe: MATLAB with the mex compiler

cudnn error when testing neuron network

yubao@yubao-Z370M-S01:~/data/Project/semseg/caffe$ ./build/examples/cpp_classification/classification.bin ./models/bvlc_reference_caffenet/deploy.prototxt ./models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel ./data/ilsvrc12/imagenet_mean.binaryproto ./data/ilsvrc12/synset_words.txt ./examples/images/cat.jpg
F0630 09:11:41.399860 17359 cudnn_conv_layer.cpp:53] Check failed: status == CUDNN_STATUS_SUCCESS (1 vs. 0)  CUDNN_STATUS_NOT_INITIALIZED
Check failure stack trace: ***
    @     0x7fdf7114346d  google::LogMessage::Fail()
    @     0x7fdf71145a23  google::LogMessage::SendToLog()
    @     0x7fdf71142ffb  google::LogMessage::Flush()
    @     0x7fdf7114496e  google::LogMessageFatal::~LogMessageFatal()
    @     0x7fdf7170c0db  caffe::CuDNNConvolutionLayer<>::LayerSetUp()
    @     0x7fdf718293dc  caffe::Net<>::Init()
    @     0x7fdf7182c4a0  caffe::Net<>::Net()
    @           0x4085bb  Classifier::Classifier()
    @           0x4048c3  main
    @     0x7fdf6c775830  __libc_start_main
    @           0x405129  _start
Aborted (core dumped)

Solution:

Check my cuda version via nvidia-smi

cuda 10.0

Check cudnn version

yubao@yubao-Z370M-S01:~/data/Project/semseg/caffe$ dpkg -l | grep cuda |grep dnn
ii  libcudnn7                                                7.6.0.64-1+cuda10.1                                   amd64        cuDNN runtime libraries
ii  libcudnn7-dev                                            7.6.0.64-1+cuda10.1                                   amd64        cuDNN development libraries and headers

It looks like their version is inconsistent.

It is caused by apt-get upgrade

You can use

sudo apt-mark hold libcudnn7 libcudnn7-dev

to avoid auto upgrade.

remove current cudnn

My cudnn directory is here: /usr/lib/x86_64-linux-gnu/libcudnn.so

Remov all related libs

sudo dpkg -r libcudnn7 libcudnn7-dev

Install cuDNN again whith correct version

Finually succeed:

yubao@yubao-Z370M-S01:~/data/Project/semseg/caffe$ ./build/examples/cpp_classification/classification.bin ./models/bvlc_reference_caffenet/deploy.prototxt ./models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel ./data/ilsvrc12/imagenet_mean.binaryproto ./data/ilsvrc12/synset_words.txt ./examples/images/cat.jpg
---------- Prediction for ./examples/images/cat.jpg ----------
0.3134 - "n02123045 tabby, tabby cat"
0.2380 - "n02123159 tiger cat"
0.1235 - "n02124075 Egyptian cat"
0.1003 - "n02119022 red fox, Vulpes vulpes"
0.0715 - "n02127052 lynx, catamount"

Reference: http://nhoj62003.blogspot.com/2017/10/caffe.html

Name		Name	Last commit message	Last commit date
Latest commit History 18 Commits
LUT		LUT
SegNet		SegNet
caffe		caffe
docker		docker
images		images
.gitignore		.gitignore
LICENSE		LICENSE
README.org		README.org
run_docker.sh		run_docker.sh

License

yubaoliu/semseg

Folders and files

Latest commit

History

Repository files navigation

Introduction

System Environment

Linux ubuntu 16.04

Cuda

cuDNN

Python

External Files

Weights

Python pacages

Reference

Build caffe

Dependences

CUDA

cuDNN

compile caffe and install

Config python path

Install caffe using conda

Test python caffe

SegNet: SegNet-Tutorial

Structure

Build caffe-segnet

Getting Started with Live Demo

Setting Up Caffe and the Dataset

SegNet

Training CamVid using SegNet-basic

Dataset

Training

Testing

Result of basics

Training Pascal VOC

Dataset

Training

Testing

Result

Possible Errors

ImportError: No module named caffe

Cudnn error when compile caffe-segnet-cudnn5

cudnn error when testing neuron network

About

Resources

License

Stars

Watchers

Forks

Releases

Packages 0

Languages

Packages