SoonminHwang / caffe-for-seg-det

Merged version of caffe for segmentation and detection

Caffe for Dilation Network & Faster R-CNN

This is a fork of Caffe that support the training of dilation network described in this ICLR conference paper. It is updated with the master branch of BVLC/caffe at commit f28f5a (June 27, 2016). The original README of Caffe is at README_caffe.md.

Added layer from caffe for faster r-cnn

	- RoI pooling Layer
	- Smooth L1 Layer
	- Modified caffe.proto

Build

This fork perserves the compatibililty with vanilla Caffe. For help of compiling this code, please turn to Caffe installation guide.

Makefile in this fork is changed slightly so that it is easier to build caffe from different git branches at the same time. There are some changes in the resulting build dir:

• git branch name is appended after ${BUILD_DIR}. So the actual build dir is ${BUILD_DIR}_${GIT_BRANCH}, such as ./build_master.
• ${BUILD_DIR}_${GIT_BRANCH}/python should be added to PYTHONPATH instead of the python dir in the caffe source code.

Usage

Please check fyu/dilation for how to use this code. The built caffe binary is used directly by dilation/train.py.