3D-Caffe

The 3D Caffe library with supporting the 3D operations (Conv and Deconv).

Modification

• ND support for CuDNN engine in "Convolution" layer
• CuDNN engine for layer "Deconvolution"
• Support for ND pooling in 'Pooling' layer
• Add randomly cropping operation in "HDF5DataLayer"

Besides, This version also includes the following modification:

• Add a weighted softmax loss layer
• Add ImageSeg dataLayer to read image and segmentation mask

Installation

1. Clone the respository and checkout 3D-Caffe branch

git clone https://github.com/yulequan/3D-Caffe/
cd 3D-Caffe
git checkout 3D-Caffe

2. Build Caffe

cp Makefile.config.example Makefile.config
 vim Makefile.config
##uncomment USE_CUDNN := 1 if you want to use CuDNN
make -j8

For other installation issues, please follow the offical instructions of Caffe.

It has been tested successfully on Ubuntu 14.04 with a) CUDA 8.0 and CuDNN 5.0; b) CUDA 7.5 and CuDNN4.0.

Note

• Remember to use git checkout 3D-Caffe. Otherwise, you only compile the offical Caffe version.

• We use HDF5DataLayer to read data when doing 3D operation. You need to generate the hdf5 data from original data type. You can refer this to generate your own h5 data using Matlab.

• You can refer the HeartSeg project as a demo on usage of this 3D-Caffe and generating the h5 file.

Reference code

• U-Net code