notes.md

📝 Notes

🛠️ Set-up

• Guide: https://vulkan.telecom-sudparis.eu/help/
• Launch Python virtual environment wizard:

/space/tools/scripts/setupFramework.sh

• Select:
  • Virtualenv
  • self_supervised_learning
  • [keep default]
  • Basic
  • Tested
  • Python 3.8.15
  • GPU
  • CUDA 11.5
• Activate Python virtual environment:

cd /home/self_supervised_tmp/self_supervised_learning
source start

• Set-up project:

# Set-up Git project
mkdir dev
cd dev
sudo apt install -y git
git config --local user.name "FABallemand"
git config --local user.email "allemand.fabien@orange.fr"
git clone https://github.com/FABallemand/ProjetCassiopee
cd ProjetCassiopee

# Download utility
sudo apt install htop

# Download data
pip3 install gdown
sudo apt-get install -y p7zip-full p7zip-rar
nohup ./data/RGB-D_Object/download.sh &
nohup ./data/mocaplab/download.sh &

# Set-up Jupyter
# https://janakiev.com/blog/jupyter-virtual-envs/#add-virtual-environment-to-jupyter-notebook
# Deactivate virtual environment?
pip3 install --user ipykernel
python3 -m ipykernel install --user --name=self_supervised_learning
jupyter kernelspec list

# Download requirements
pip3 install -r requirement.txt
pip3 install torch==1.12.1+cu113 torchvision==0.13.1+cu113 torchaudio==0.12.1 --extra-index-url https://download.pytorch.org/whl/cu113

• https://pytorch.org/blog/deprecation-cuda-python-support/
• https://pytorch.org/get-started/previous-versions/

💾 Useful Commands

Data

tar -xvf file.tar
tar -xzvf file.tar.gz

7za x file.zip -ooutput/path

# Download Google Drive folder/file
gdown --folder link/to/folder -O /path/to/dir

# Copy folder/file from local to server
scp -r /path/to/local/dir user@remotehost:/path/to/remote/dir

Python

# Run program with nohup (see also tmux)
nohup python3 -u main.py &

Debugging

# Kill nohup process
# https://stackoverflow.com/questions/17385794/how-to-get-the-process-id-to-kill-a-nohup-process

# Run program with CUDA_LAUNCH_BLOCKING=1
CUDA_LAUNCH_BLOCKING=1 nohup python3 -u main.py &

# Reboot server
sudo reboot now

# Clear RAM cache
sudo sync; echo 1 > /proc/sys/vm/drop_caches

# Kill all process using pattern
pkill -f <pattern>

# "Restart" GPU on Ubuntu
# https://discuss.pytorch.org/t/cuda-fails-to-reinitialize-after-system-suspend/158108/3
sudo rmmod nvidia_uvm
sudo modprobe nvidia_uvm

# Debugging tool
dmesg | grep GPU

# Read end of file
# https://kb.iu.edu/d/acrj
tail [ +-[number][lbcr] ] [file]

🗂️ Dataset

• RGBD-SOD Dataset
• How2Sign (A Large-scale Multimodal Dataset for Continuous American Sign Language)
• HIC (Hands in Action)
• RGB-D Object Dataset
• List of RGBD datasets
• Mocaplab Google Drive
• Motion-X: A Large-scale 3D Expressive Whole-body Human Motion Dataset

🖥️ Code

• Tmux Tutorial Video
• Tmux Cheat Sheet
• PyTorch v1.12 Documentation
• PyTorch Tutorials
• SimCLR Article
• SimCLR in PyTorch
• SimCLR Tutorial
• SimCLR Tutorial Videos
• Demystify RAM Usage in Multi-Process Data Loaders

🧠 AI

• MMAction
• MMAction graph-based action recognition
• Two-Stream Adaptive Graph Convolutional Networks for Skeleton-Based Action Recognition

📋 ToDo

• [] ...