caffe-ssd下载:https://share.weiyun.com/50NyOiw
需要安装boost 1.54版本、python2.7、protobuf 2.6.1
boost 1.54下载:https://share.weiyun.com/5kwTbx1
protobuf 2.6.1下载:https://share.weiyun.com/5HzTe0d
安装protobuf:
# 安装protobuf:
cd protobuf-2.6.1
./configure
make all
sudo make install
# 检查protobuf是否安装成功
protoc --version
安装boost:
tar zxvf boost_1_58_0.tar.gz
#设置编译器和所选库
cd boost_1_58_0
./bootstrap.sh --with-libraries=all --with-toolset=gcc
#编译boost
./b2 toolset=gcc
#安装boost
./b2 install --prefix=/usr
ldconfig
#卸载boost的话,直接删除
rm -rf /usr/lib/boost_*
rm -rf /usr/include/boost
配置caffe-ssd:
将文件 upsample_layer.hpp 放置,caffe_ssd/include/caffe/layers/upsample_layer.hpp
将文件upsample_layer.cpp放置,caffe_ssd/src/caffe/layers/upsample_layer.cpp
将文件upsample_layer.cu放置,caffe_ssd/src/caffe/layers/upsample_layer.cu
编辑Makefile文件:
修改opencv这项为:
ifeq ($(USE_OPENCV), 1)
LIBRARIES += opencv_core opencv_highgui opencv_imgproc
ifeq ($(OPENCV_VERSION), 3)
LIBRARIES += opencv_imgcodecs opencv_videoio
endif
endif
LIBRARIES修改为:
LIBRARIES += glog gflags protobuf boost_system boost_filesystem boost_regex boost_thread stdc++ m hdf5_hl hdf5
编辑Makefile.config文件:
CPU_ONLY := 1
USE_OPENCV := 1
OPENCV_VERSION := 3
INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial/ /usr/include/opencv2/
LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/x86_64-linux-gnu/hdf5/serial/ /usr/lib/x86_64-linux-gnu/
编译caffe-ssd:
sudo make all -j8
sudo make py
开始转化:
安装pytorch:
#安装pytorch
pip install “自行去官网,选择合适的pytorch版本”
pip install torchvision
pip install numpy
#测试pytorch是否安装成功,进入python环境
import torch
import torchvision
print(torch.cuda.is_available())
借助pytorch转化工具:https://share.weiyun.com/58a6JNC
yolov3,在darknet下的配置文件和权重文件,下载链接:https://share.weiyun.com/5MdbRss
将yolov3-tiny-half_3cls64.cfg文件里的:
#[maxpool]
#size=2
#stride=1
注释掉。
最后执行:
sudo python yolov3_darknet2caffe.py