deeplearning_tutorial

Syllabus

1. Machine Learning Basic
2. Neural Network & Training Techniques
3. Convlutional Neural Network & Recurrent Neural Network
4. Detection & Segmentation ( Yolo2 )
5. Reinforcement Learning ( DQN & gym )
6. Reinforcement Learning ( A3C )
7. Generative Model ( Auto-encoder & GAN & DCGAN )
8. Jetson TX2 setup & Test TensorRT
9. Paper Review

Docker Setup

Docker on Ubuntu with CPU

docker run --name pytorch --rm \
    -v "/tmp/.X11-unix:/tmp/.X11-unix:rw" \
    -v "$HOME/.Xauthority:/root/.Xauthority:rw" \
    -e QT_X11_NO_MITSHM=1 \
    -e DISPLAY=$DISPLAY \
    -p 8888:8888 \
    --net=host \
    --mount 'type=bind,src=/home/ghryou/Workspace/deeplearning_tutorial,dst=/app' \
    -it ghryou/pytorch:cpu bash

jupyter notebook --allow-root

Docker on Ubuntu with GPU

docker run --name pytorch --rm \
   --runtime=nvidia \
   -v "/tmp/.X11-unix:/tmp/.X11-unix:rw" \
   -v "$HOME/.Xauthority:/root/.Xauthority:rw" \
   -e QT_X11_NO_MITSHM=1 \
   -e DISPLAY=$DISPLAY \
   --net=host \
   --mount 'type=bind,src=/home/naverlabs/Workspace/deeplearning_tutorial,dst=/app' \
   -it ghryou/pytorch:gpu bash​

Docker on Windows

1. setup shared memory at docker settings Right click docker icon -> Settings -> Shared Drives -> Check C Drive -> Apply -> Type password -> Restart docker

2. Install Xming X Server 3. Follow this link to setup Xming X Server

4. Run Xming 5. Open windows 10 powershell & Run this code

docker run --name pytorch --rm `
-p 8888:8888 `
-p 6006:6006 `
-e DISPLAY:<IP_Address>:0 `
--mount type=bind,src=C:\USERS\naverlabs\Desktop\deeplearning_tutorial,dst=/app `
-it ghryou/pytorch:cpu bash​

Docker on Mac

1. Install homebrew
2. brew install xquartx
3. Follow this link for X11 security Setup
4. Run this code

open -a XQuartz
./docker/run.mac.sh ghryou/pytorch:cpu bash

Ubuntu Commands

sudo chown <User name> -R <directory path>

Docker Cheat Sheet

docker build -t <tag name> -f <Dockerfile path> <Dockerfile directory>
docker exec -it pytorch bash
docker images
docker ps -al
docker stop <image id>
docker rmi <image id>
docker system prune

Push images to Docker Cloud

Jetson Setup

1. Download JetPack from Nvidia website
2. Makes run files executable with sudo chmod +x JetPack-L4T-3.3-linux-x64_b39.run
3. Run ./JetPack-L4T-3.3-linux-x64_b39.run
4. Follow the Commands of JetPack Installer (Install every package)
5. Select Device accesses Internet via Router/Switch
6. Connect jetson with host PC and follow the instructions
7. Connect display to jetson and setup internet connection on jetson
8. Check IP address of jetson with ifconfig
9. Run ./JetPack-L4T-3.3-linux-x64_b39.run again
10. Disable Flash OS Image to Target
11. Select Device accesses Internet via Router/Switch
12. Enter jetson's IP address and follow the instructions

Jetson Pytorch Setup

1. Install pip3

sudo apt-get update && sudo apt-get upgrade
sudo apt-get install python3-pip
pip3 install -U pip3

2. clone pyTorch repo

git clone http://github.com/pytorch/pytorch
cd pytorch
git submodule update --init

3. install prereqs

sudo pip3 install -U setuptools
sudo pip3 install -r requirements.txt

4. Develop Mode:

python3 setup.py build_deps
sudo python3 setup.py develop

5. Verify CUDA (from python3 interactive terminal)

import torch
print(torch.__version__)
print(torch.cuda.is_available())

Reference: https://gist.github.com/dusty-nv/ef2b372301c00c0a9d3203e42fd83426