See project in github : https://github.com/FreeApe/EmbCaffe
Embedded Caffe
Make the depth learning model running on Embedded Systems and multi-platform.
Features
- Based on caffe
- Compact,Only forward calculation
- Support faster rcnn
- Support ssd
- Support GPU/CPU model(Windows only support CPU model now)
- Cross-platform
- Small
- Remove the gtest
- Remove the test
- Remove the python wrapper
Requirements
Linux
System : Linux / Ubuntu 14.04
Cmake : 3.4+
C++ compiler : 4.8+
BLAS : Atlas
Boost : 1.54
glog : Y
gflags : Y
protobuf : 2.6.1
CUDA : 7.5
cuDNN : 5.0.5
OpenCV : N (for examples)
Windows
System : Windows
Cmake-gui : 3.8+
C++ compiler : vc140(VS2015)
embcaffe_3rdparty : https://github.com/FreeApe/embcaffe_3rdparty
Build
Build EmbCaffe on Linux
- Way 1
# Note: Test examples with this way
$ cd EmbCaffe
$ mkdir cmake_build
$ cd cmake_build
$ cmake ..
$ make all -j
- Way 2
$ cd EmbCaffe
$ make all -j
Build EmbCaffe on Windows
1. with CMake-gui3.8+
2. git clone https://github.com/FreeApe/embcaffe_3rdparty(You can also compile these third-party libraries yourself)
2. Configure and Generate(You should configure CMAKE_INCLUDE_PATH and CMAKE_LIBRARY_PATH)
3. Open Caffe.sln with VS2015
4. build solutions
Build examples
Linux:
$ cd examples
$ mkdir build
$ cd build
$ cmake ..
$ make -j
Windows:
1. with CMake-gui3.8+
2. Configure and Generate
3. Open Caffe-example.sln with VS2015
4. build solutions
5. run demos
Running the examples, the results show as :
Build Errors
1. error ‘type name’ declared as function returning an array escape
Make sure the CUDA version is 7.5
TODO / Targets
- Remove Backward calculation
- Support GPU model on windows
- Optimize the calculation
- ……