TX2入门教程软件篇-安装MXNet(jetpack3.3)

TX2入门教程软件篇-安装MXNet(jetpack3.3)

说明：

• 介绍如何在tx2下安装MXNet
• 测试环境：jetpack3.3

步骤：

• 新建目录

mkdir -p ~/dl/Mxnet
cd ~/dl/Mxnet

• 安装依赖：

sudo apt-get update
sudo apt-get -y install git build-essential libatlas-base-dev libopencv-dev graphviz python-pip
sudo pip install pip --upgrade
sudo pip install setuptools numpy --upgrade
sudo pip install graphviz jupyter

• 下载mxnet：

git clone https://github.com/dmlc/mxnet.git --recursive
cd mxnet

• 修改配置增加CUDA支持，无CUDA可不改：
• 进入mxnet里面的make文件夹，更改config.mk

cp make/config.mk .
echo "USE_CUDA=1" >> config.mk    
echo "USE_CUDA_PATH=/usr/local/cuda" >> config.mk
echo "USE_CUDNN=1" >> config.mk

• 修改3rdparty/mshadow/make/mshadow.mk

vim 3rdparty/mshadow/make/mshadow.mk 

• 修改内容：
• 先找到122行，增加如下

MSHADOW_CFLAGS += -DMSHADOW_USE_PASCAL=1

• 回到mxnet文件夹里面：

 make -j $(nproc)

• Python2安装MXNet Python Bindings

cd python
pip install -e . --user

• Python3安装MXNet Python Bindings

cd python
pip3 install -e . --user

测试：

• python2 执行命令：

python example/image-classification/train_mnist.py

• python3 执行命令：

python3 example/image-classification/train_mnist.py

• 输出：

INFO:root:start with arguments Namespace(batch_size=64, disp_batches=100, gpus=None, kv_store='device', load_epoch=None, lr=0.05, lr_factor=0.1, lr_step_epochs='10', model_prefix=None, mom=0.9, monitor=0, network='mlp', num_classes=10, num_epochs=20, num_examples=60000, num_layers=None, optimizer='sgd', test_io=0, top_k=0, wd=0.0001)
INFO:root:Epoch[0] Batch [100]  Speed: 948.66 samples/sec   accuracy=0.777692
INFO:root:Epoch[0] Batch [200]  Speed: 1030.88 samples/sec  accuracy=0.915781
INFO:root:Epoch[0] Batch [300]  Speed: 1034.63 samples/sec  accuracy=0.931250
INFO:root:Epoch[0] Batch [400]  Speed: 983.33 samples/sec   accuracy=0.932031
INFO:root:Epoch[0] Batch [500]  Speed: 968.70 samples/sec   accuracy=0.943906
INFO:root:Epoch[0] Batch [600]  Speed: 1004.47 samples/sec  accuracy=0.940625
INFO:root:Epoch[0] Batch [700]  Speed: 990.10 samples/sec   accuracy=0.950313
INFO:root:Epoch[0] Batch [800]  Speed: 941.56 samples/sec   accuracy=0.951094
INFO:root:Epoch[0] Batch [900]  Speed: 992.14 samples/sec   accuracy=0.955000
INFO:root:Epoch[0] Train-accuracy=0.957770
INFO:root:Epoch[0] Time cost=64.217
INFO:root:Epoch[0] Validation-accuracy=0.963276
INFO:root:Epoch[1] Batch [100]  Speed: 942.83 samples/sec   accuracy=0.964728
INFO:root:Epoch[1] Batch [200]  Speed: 925.19 samples/sec   accuracy=0.962187

参考：

• https://elinux.org/Jetson_Zoo#PyTorch_.28Caffe2.29