Install

References

• Official
  • wiki
    • https://github.com/BVLC/caffe/wiki
  • Homepage
    • http://caffe.berkeleyvision.org/
• Others
  • https://hpc.uiowa.edu/sites/hpc.uiowa.edu/files/wysiwyg_uploads/ModifiedCaffeTutorial_0.pdf

Prerequirements

• DB (http://caffe.berkeleyvision.org/tutorial/data.html)
  • level db
    • fast key-value storage library written at Google
    • https://github.com/google/leveldb
  • lmdb
    • lightning memory-mapped database
    • https://en.wikipedia.org/wiki/Lightning_Memory-Mapped_Database
• protobuf
  • for serialize/deserialize
  • https://en.wikipedia.org/wiki/Protocol_Buffers
• hdf5
  • file format, Hierarchical Data Format
  • https://en.wikipedia.org/wiki/Hierarchical_Data_Format
• glog
  • logging library from google
  • https://github.com/google/glog
• gflags
  • command line flag parser
  • https://github.com/gflags/gflags

• for python
  • for pygraphviz, which pydot is dependent on
    • PyGraphviz doesn't work without Graphviz
    • sudo apt-get install graphviz
  • pip3 install pydot scikit-image pytest

sudo apt-get install libprotobuf-dev libleveldb-dev libsnappy-dev libhdf5-serial-dev protobuf-compiler
sudo apt-get install --no-install-recommends libboost-all-dev
sudo apt-get install libatlas-base-dev
sudo apt-get install libgflags-dev libgoogle-glog-dev liblmdb-dev

Fix

diff --git a/Makefile b/Makefile
index 4d32416..ad173f8 100644
--- a/Makefile
+++ b/Makefile
@@ -178,7 +178,7 @@ ifneq ($(CPU_ONLY), 1)
    LIBRARIES := cudart cublas curand
 endif

-LIBRARIES += glog gflags protobuf boost_system boost_filesystem m hdf5_hl hdf5
+LIBRARIES += glog gflags protobuf boost_system boost_filesystem m hdf5_serial_hl hdf5_serial

 # handle IO dependencies
 USE_LEVELDB ?= 1

Configurations

Makefile.config

• for Multi-GPU

  USE_NCCL := 1

  • install nccl (http://www.nvidia.com/object/caffe-installation.html)

    $ git clone https://github.com/NVIDIA/nccl.git
    $ cd nccl
    $ sudo make install -j4

• for cudnn

  USE_CUDNN:=1

• for opencv

  OPENCV_VERSION:=3

• for python3

  PYTHON_LIBRARIES := boost_python-py35 python3.5m
  PYTHON_INCLUDE := /usr/include/python3.5m \
                 /usr/lib/python3.5/dist-packages/numpy/core/include

• Custumize include and library dir
  • e.g. I installed OpenCV with prefix=/opt/ so I need /opt/include, /opt/lib

    INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial/ /opt/include/
    LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /opt/lib

Makefile

• for python3

@@ -533,7 +533,7 @@ runtest: $(TEST_ALL_BIN)
    $(TEST_ALL_BIN) $(TEST_GPUID) --gtest_shuffle $(TEST_FILTER)

 pytest: py
-   cd python; python -m unittest discover -s caffe/test
+   cd python; python3 -m unittest discover -s caffe/test

 mattest: mat
    cd matlab; $(MATLAB_DIR)/bin/matlab -nodisplay -r 'caffe.run_tests(), exit()'

How to run my own code

• export PYTHONPATH=~/work/caffe/ssd/python/
• Easy way
  • put a single *.cpp file in ./examples
• Build own project by linking caffe.so???
  • 아직 모른다.

save/load caffemodel

• save

  • declared in solver.prototxt

    # snapshot intermediate results
    snapshot: 5000
    snapshot_prefix: "examples/mnist/lenet"

  • you can see *.caffemodel, *.solverstate

    -rw-rw-r--  1 rofox rofox 1725006 May 25 10:59 lenet_iter_10000.caffemodel
    -rw-rw-r--  1 rofox rofox 1724471 May 25 10:59 lenet_iter_10000.solverstate
    -rw-rw-r--  1 rofox rofox 1725006 May 25 10:58 lenet_iter_5000.caffemodel
    -rw-rw-r--  1 rofox rofox 1724470 May 25 10:58 lenet_iter_5000.solverstate

• load *

Multi GPU Test

• cifar10 test total batch 100