caffe
1. 개요
출처 : Caffe 실습, SNU 패턴 인식 및 컴퓨터 지능 연구실, 박성헌, 황지혜, 유재영
- Caffe : Convolutional Architecture for Fast Feature Embedding
- Developed by the Berkeley Vision and Learning Center (BVLC)
- Yangqing Jia, Evan Shelhamer, Travor Darrell
1.1 특징
- Pure C++/CUDA architecture
- Command line, Python, MATLAB interfaces
- Fast, well-tested code
- Pre-processing and deployment tools, reference models and examples
- Image data management
- Seamless GPU acceleration
- Large community of contributors to the open-source project
1.2 기능
Data pre-processing and management : $CAFFE_ROOT/build/tools
- Conversion from CSV and Images to LMDB
A. Data ingest formats
- LevelDB or LMDB database
- In-memory (C++ and Python only)
- HDF5
- Image files
B. Pre-processing tools
- LevelDB/LMDB creation from raw images
- Training and validation set creation with shuffling
- Mean-image generation
C. Data transformations(tools.data_augmentation)
- Image cropping, resizing, scaling and mirroring
- Mean subtraction
1.3 이미지 처리
Caffe expects the images (i.e. the dataset) to be stored as blob of size (N, C, H, W)
- N being the dataset size
- C the number of channels
- H the height of the images
- W the width of the images.
1.4 LMDB I/O and Pre-processing
데이터를 LMDB에 넣어 처리 하는것을 선호
- import lmdb :
2. 설치
현재('17.03월) python2 만 지원
WITH_PYTHON_LAYER=1 option설치 필요
.bashrc 설정 필요
export OPENBLAS_NUM_THREADS=(4)
export CAFFE_ROOT=/home/david/caffe
export LD_LIBRARY_PATH=/usr/local/cuda/lib64:$LD_LIBRARY_PATH
export PYTHONPATH=/home/david/caffe/python:$PYTHONPATH
참고 : caffe-installation
- 스크립트 이용하여 설치 : CPU Only, Ubuntu 14.04
Caffe 실습: [추천_pdf] 서울대학교 융합과학기술대학원, 패턴 인식 및 컴퓨터 지능 연구실