吉浦迅-Jetson TX安装Caffe
400-600-6945
点击这里给我发消息

当前位置:首页 >> 实验室报告 >>Jetson TX安装Caffe

NEWS
Jetson TX安装Caffe
作者/来源:原创     发布于:2018-05-12     点击数:140

CAFFE作为深度学习的常用框架,也是NVIDIA一开始就密切合作的重要应用,目前除了BVLC的开源CAFFE安装包之外,NVIDIA也提供 GPU 特调板 -- NVCAFFE。根据NVIDIA提供的讯息,最新NVCAFFE 0.17版提供以下功能:

NVIDIA Caffe (NVIDIA Corporation ©2017) is an NVIDIA-maintained fork of BVLC Caffe tuned for NVIDIA GPUs, particularly in multi-GPU configurations. Here are the major features: 

▶ 16 bit (half) floating point train and inference support.

▶ Mixed-precision support. It allows to store and/or compute data in either 64, 32 or 16 bit formats. Precision can be defined for every layer (forward and backward passes might be different too), or it can be set for the whole Net.

▶ Layer-wise Adaptive Rate Control (LARC) and adaptive global gradient scaler for better accuracy, especially in 16-bit training.

▶ Integration with cuDNN.

▶ Automatic selection of the best cuDNN convolution algorithm.

▶ Integration with v2.2 of NCCL library for improved multi-GPU scaling. (这部分对 TX 没有用!)

▶ Optimized GPU memory management for data and parameters storage, I/O buffers and workspace for convolutional layers.

▶ Parallel data parser, transformer and image reader for improved I/O performance.

▶ Parallel back propagation and gradient reduction on multi-GPU systems.

▶ Fast solvers implementation with fused CUDA kernels for weights and history update.

▶ Multi-GPU test phase for even memory load across multiple GPUs.

▶ Backward compatibility with BVLC Caffe and NVCaffe 0.15 and higher.

▶ Extended set of optimized models (including 16 bit floating point examples).

虽然 NVACFFE 并不算 CAFFE 的一支(branch),但是安装、编译方式完全一样。

1、取得安装包:

    a) NVCAFFE下载地址:https://github.com/NVIDIA/caffe

        或 git clone https://github.com/NVIDIA/caffe.git

        使用手册在 https://github.com/NVIDIA/caffe/blob/caffe-0.17/NVCaffe-User-Guide.pdf

    b) BVLC CAFFE下载地址:https://github.com/BVLC/caffe

        或 git clone https://github.com/BVLC/caffe.git

2、依赖库:

    $ sudo apt-get install libprotobuf-dev libleveldb-dev libsnappy-dev libhdf5-serial-dev protobuf-compiler libjpeg-turbo* -y

    $ sudo apt-get install --no-install-recommends libboost-all-dev -y

    # 安装 BLAS 库
    $ sudo apt-get install libatlas-base-dev libopenblas-dev -y
    # 其他 Remaining Dependencies
    $ sudo apt-get install libgflags-dev libgoogle-glog-dev liblmdb-dev -y
    $ sudo apt-get install python-dev python-numpy -y

3、开始安装

    $ cd caffe 

    $ mkdir build && cd build 

    $ sudo cmake -DCUDA_USE_STATIC_CUDA_RUNTIME=OFF ..

    $ sudo make -j4 all        # 大约 40 分钟

    $ sudo make -j4 pycaffe

    # 如果您的时间充裕:

    $ sudo make -j6 runtest

4、独立测试:在 build 目录下

    $ tools/caffe time --model=../models/bvlc_alexnet/deploy.prototxt --gpu=0  #GPU

    $ tools/caffe time --model=../models/bvlc_alexnet/deploy.prototxt               #CPU

【完】


  • 正品行货
  • 正规发票
  • 修养保障
  • 技术支持

最新消息  |  关于吉浦迅

  • 技术QQ群
  • OpenACC
  • Q群:195055206
  • OpenCL
  • Q群:142754832
  • GPU Matlab
  • Q群:62833093
苏州吉浦迅科技有限公司苏ICP备09073381号