Skip to content

Latest commit

 

History

History
58 lines (52 loc) · 2.65 KB

dtu_README.md

File metadata and controls

58 lines (52 loc) · 2.65 KB
Raw

INSTALLATION

if you've noticed your python3 bin doens't point to your conda env when using --prefix to point to your scratch dir, then you need to do the following:

  • conda config --set always_copy True
  • conda config --show | grep always_copy now continue as normal:
  • conda create --prefix /MotionDiffuse/env python=3.7
  • conda activate /MotionDiffuse/env
  • double check your GCC is 5+ by running gcc --version; if not, do module load gcc/5.4.0
  • module load cuda/10.1 # you must run these icuda commands before installing torch otherwise it will say version not found!!
  • module load cudnn/v7.6.5.32-prod-cuda-10.1
  • conda install pytorch=1.7.1 torchvision=0.8.2 cudatoolkit=10.1 -c pytorch
  • python3 -m pip install "mmcv-full>=1.3.17,<=1.5.3" -f https://download.openmmlab.com/mmcv/dist/cu101/torch1.7.1/index.html
  • python3 -m pip install -r requirements.txt
  • python3 -m pip install --upgrade protobuf==3.20.0

fyi there is an annoying warning in the logs (https://stackoverflow.com/questions/57381430/synonym-of-type-is-deprecated-in-a-future-version-of-numpy-it-will-be-underst) that can be silenced by downgrading numpy:to 1.16.4 BUT this is incompatible with the other package versions, so don't do it

fyi: (/work3/s222376/MotionDiffuseNew) s222376@n-62-20-1 /work3/s222376/MotionDiffuse/text2motion (train_baseline)$ module list Currently Loaded Modulefiles:

  1. latex/TeXLive19(default) 3) cudnn/v7.6.5.32-prod-cuda-10.1 5) gcc/5.4.0
  2. cuda/10.1 4) binutils/2.29(default)

TRAINING

  • download KIT-ML data from <> and put the zip for it in text2motion/data/
  • cd text2motion/data && unzip KIT-ML-20231122T121619Z-001.zip
  • cd KIT-ML && unrar x new_joint_vecs.rar
  • unrar x new_joints.rar
  • unrar x texts.rar
  • dirs should look like
text2motion/data/KIT-ML
├── new_joint_vecs
│   ├─�
├── new_joints
│   ├─�
└── texts
    ├─�
--all.txt
--<etc>
  • voltash (dtu hpc command to go to interactive gpu node)
  • make train
  • verify above works without errors and then kill training because you're on interactive gpu, you will likely run out of memory anyway (can decrease --batchsize but then it's slow)
  • to do full training, edit jobscript.sh to use your email and submit job via "make queue"

INFERENCE with pretrained model

  • download...checkpoints?? idk look at their README.md

Changes I made

  • ignore standardization
  • tokens are [] empty...
  • reusing kit_chain thing lol
  • only training on one sequence from grab

TO KEEP IN MIND:

  • they specify best way to train in readme somewhere -- follow this when doing real training!
  • need to add the emotion text to the caption!!