unreal_caffe is a self-maintained caffe version without pain in official repo. current version support some layer which needed by Faster-RCNN and R-FCN in C++. CUDNN version update to cudnn7.
- cudnn 7 support, add to cudnn7;
roi_pooling_layer
needed in Fast-RCNN series task;smooth_l1_loss_ohem_layer
needed in RFCN task;deformalable_conv_layer
which is the most advanced conv (still under test.);
Note: the default config in Makefile.config
not using CUDA and CUDNN, if you wanna build on CUDA, make sure you have CUDA9.0 & CUDNN 7 installed.
For installation, pls install depencies first incase you are not already for the installation:
./install_depencies.sh
make all -j32
sudo make install
If you wanna contribute to this project, you can add my wechat: jintianiloveu
. New layers and feature and examples are welcome!
Do something like this in your project.
link_directories("~/unreal_caffe/include/")
include_directories("~/unreal_caffe/include/")
find_library(caffe_LIBRARY
NAMES caffe
HINTS "~/unreal_caffe/build/lib/"
PATHS "~/unreal_caffe/build/lib/")
target_link_libraries(rfcn_caffe ${caffe_LIBRARY})
Original copyright:
@article{dai16rfcn,
Author = {Jifeng Dai, Yi Li, Kaiming He, Jian Sun},
Title = {{R-FCN}: Object Detection via Region-based Fully Convolutional Networks},
Journal = {arXiv preprint arXiv:1605.06409},
Year = {2016}
}