CaffeMex

v2.3:

A multi-GPU & memory-reduced MAT-Caffe on LINUX and WINDOWS

For now it is an unstable version has worked well in plenty of works.

Contributors & Bug report: [Yu Liu] liuyuisanai@gmail.com [Hongyang Li] yangli@ee.cuhk.edu.hk

What's New in Version 2.3?

1.Reduce 30% GPU memory usage

2.More stable for detection task (fix 0-count issue in a few layers)

3.Add center loss layer

Installation for LINUX

1.cp Makefile.config.example Makefile.config and customize your own config.

2.make -j && make matcaffe

Installation for WINDOWS

1.Find 'windows/' and you will know. By merging R-FCN https://github.com/daijifeng001/R-FCN

Feature

1.Support both windows and linux platforms

2.Reduce 30% GPU memory usage (by merging Yuanjun Xiong's caffe http://yjxiong.me/)

3.Lastest Matlab interface

4.Compatible with original caffe

Running on single machine with single/multiple GPU(s)