Skip to content

faster rcnn cuda 10

2018.10.06 01:09

WHRIA 조회 수:2084

https://gist.github.com/acmiyaguchi/bc535ba23eabd3564edd73e491763d50

 


이 사람의 rebased 를 download 하고 나서 아래처럼 patch

 

Instructions

  • install cuda-9.0 (https://developer.nvidia.com/cuda-toolkit-archive)
  • install cudnn-7 (https://developer.nvidia.com/cudnn)
  • install dependencies for caffe
    • see the caffe installation guide [1] and wiki page [2]
  • install boost-1.65 or higher (http://www.boost.org/users/history/version_1_65_1.html) [3]
  • update caffe-fast-rcnn
    • Easiest way is to use a rebased branch on BVLC/master. See the notes below if you would rather merge the changes yourself.
    $ cd $FRCN_ROOT/caffe-fast-rcnn
    $ git remote add acmiyaguchi https://github.com/acmiyaguchi/caffe-fast-rcnn.git
    $ git checkout acmiyaguchi/faster-rcnn-rebased
    
    • optional: rebase against upstream master
    $ git remote add caffe https://github.com/BVLC/caffe.git
    $ git rebase caffe/master
    
  • set variables in Makefile.config
    • USE_CUDNN := 1
    • WITH_PYTHON_LAYER := 1
    • OPENCV_VERSION := 3
    • CUSTOM_CXX := g++-6
      • I symlink gcc-6 and g++-6 to gcc and g++ in /usr/local/cuda-9-0/bin/ instead of this setting, but either should work
    • remove compute_20 from CUDA_ARCH since it's been depreciated in cuda-9-0
    • add hdf5 for linking since it's been renamed in 17.10
      • INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/include/hdf5/serial/
      • LIBRARY_DIRS := $(PYTHON_LIB) /usr/lib/x86_64-linux-gnu/hdf5/serial/
  • build caffe
    • make -j${N_THREADS} all && make pycaffe
  • change self.param_str_ to self.param_str in $FRCN_ROOT/lib/rpn/proposal_layer.py.
    • see rebased-caffe.patch below for the full diff
  • validate that ./tools/demo.py works

Notes on Rebasing

BVLC/caffe and rbgirshick/caffe have diverged since this workaround [4] in March 2017. I've rebased the faster-rcnnpatchset against the master caffe branch for convenience. [5]

This branch should merge cleanly into BVLC/caffe with the exception of the license.

If you rather merge the upstream changes yourself, the patch in [6] should outline the changes you need afterwards. -X theirs may not be the best merge policy. In particular, changes to the src/caffe/caffe.proto schema will overwrite the ROIPoolingParameter.

rebased-caffe.patch

diff --git a/lib/rpn/proposal_layer.py b/lib/rpn/proposal_layer.py
index b157160..e6d28cc 100644
--- a/lib/rpn/proposal_layer.py
+++ b/lib/rpn/proposal_layer.py
@@ -23,7 +23,7 @@ class ProposalLayer(caffe.Layer):
 
     def setup(self, bottom, top):
         # parse the layer parameter string, which must be valid YAML
-        layer_params = yaml.load(self.param_str_)
+        layer_params = yaml.load(self.param_str)
 
         self._feat_stride = layer_params['feat_stride']
         anchor_scales = layer_params.get('scales', (8, 16, 32))
번호 제목 글쓴이 날짜 조회 수
» faster rcnn cuda 10 WHRIA 2018.10.06 2084
1550 Intro WHRIA 2018.10.01 3969
1549 ambient-light-sensor WHRIA 2018.09.29 250
1548 register WHRIA 2018.09.21 1892
1547 Deep learning 기반 DEMO WHRIA 2018.09.16 7103
1546 slide note 삭제 슬라이드 노트 삭제 WHRIA 2018.09.12 4393
1545 faster rcnn resnet 101 WHRIA 2018.09.09 169
1544 나이 40에... WHRIA 2018.09.01 130
1543 letencrypt - win WHRIA 2018.08.12 34
1542 달러 환율 file WHRIA 2018.06.16 1262
1541 Interpretation of the Outputs of Deep Learning Model Trained with Skin Cancer Dataset [1] WHRIA 2018.06.02 7678
1540 skin cancer [1] WHRIA 2018.05.23 1454
1539 deep learning framework WHRIA 2018.05.17 1767
1538 mxnet [2] WHRIA 2018.05.17 3192
1537 intel 내장 그래픽 / CUDA WHRIA 2018.05.12 3113

Powered by Xpress Engine / Designed by Sketchbook

sketchbook5, 스케치북5

sketchbook5, 스케치북5

나눔글꼴 설치 안내


이 PC에는 나눔글꼴이 설치되어 있지 않습니다.

이 사이트를 나눔글꼴로 보기 위해서는
나눔글꼴을 설치해야 합니다.

설치 취소