Support for mmcv 1.7, CUDA 11.8 and PyTorch 1.13?

[mexicantexan]

Wondering if this could get support for mmcv 1.7, mmsegmentation 0.29, CUDA 11.8 and PyTorch 1.13? Currently, when attempting to use these dependencies several errors are thrown on import.

[zhyever]

Could you please share the errors so that I could try to help you?

[mexicantexan]

@zhyever just wanted to say great repo :) I ended up getting the repo to work with mmcv 1.7, CUDA 11.7, Pytorch 1.13, and mmsegmentation 0.29. My environment was Linux 22.04 with Python 3.9. Here is the script I used to set everything up.

#!/bin/bash

INSTALL_DIR="./Monocular-Depth-Estimation-Toolbox"
if [ -d "$INSTALL_DIR" ]; then
  cd Monocular-Depth-Estimation-Toolbox
  VENV_DIR="./venv"

  if [ ! -d "$INSTALL_DIR" ]; then
    # pip install pre-build reqs
    python3.9 venv venv
  fi

  . ./venv/bin/activate

  pip install wheel setuptools

  # install latest pytorch (at time of writing -> CUDA 11.7, torch 1.13)
  pip install torch torchvision torchaudio

  # install matching mmcv-full
  pip install openmim
  mim install mmcv-full

  # install matching mmsegmentation
  pip install mmsegmentation

  # install runtime reqs
  pip install -r ./requirements/runtime.txt
  # install test reqs
  pip install -r ./requirements/tests.txt
  # install docs reqs
  pip install -r ./requirements/docs.txt

  # install optional reqs
  pip install -r ./requirements/optional.txt

  # install training requirements
  pip install future tensorboard

  # install other reqs as needed
  pip install timm

  pip install -e .

  # create data directory
  if [ ! -d "./data" ]; then
    # pip install pre-build reqs
    mkdir data
  fi
  cd data

  # create nyu directory
  if [ ! -d "./nyu" ]; then
    mkdir nyu
  fi
  # create kitti directory
  if [ ! -d "./kitti" ]; then
    mkdir kitti
  fi
  # create cityscapes directory
  if [ ! -d "./cityscapes" ]; then
    mkdir cityscapes
  fi
  if [ ! -d "./cityscapesExtra" ]; then
    mkdir cityscapesExtra
  fi
  # create sunrgbd directory
  if [ ! -d "./SUNRGBD" ]; then
    mkdir SUNRGBD
  fi
  # create custom dataset
  if [ ! -d "./custom_dataset" ]; then
    mkdir custom_dataset
  fi
  cd custom_dataset
  if [ ! -d "./train" ]; then
    mkdir train
  fi
  if [ ! -d "./val" ]; then
    mkdir val
  fi
  cd ..

  # copy splits into directories
  cd ../splits
  ls . | grep "cityscapes" | xargs -I '{}' cp '{}' ../data/cityscapes
  ls . | grep "cityscapes" | xargs -I '{}' cp '{}' ../data/cityscapesExtra
  ls . | grep "kitti" | xargs -I '{}' cp '{}' ../data/kitti
  ls . | grep "nyu" | xargs -I '{}' cp '{}' ../data/nyu
  ls . | grep "SUNRGBD" | xargs -I '{}' cp '{}' ../data/SUNRGBD
  cd ..

  # optional for final test
  cp ../test.png ./data/custom_dataset/val
  cp ../test_split.txt ./data/custom_dataset
  mv ./data/custom_dataset/test_split.txt ./data/custom_dataset/nyu_test.txt

  # create checkpoint dir
  if [ ! -d "./checkpoints" ]; then
    mkdir checkpoints
  fi
  cd checkpoints

  if [ ! -f "binsformer_swinl_22k_w7_nyu.pth" ]; then
    echo "Downloading binsformer pretrained model..."
    wget -O binsformer_swinl_22k_w7_nyu.pth "https://drive.google.com/u/0/uc?id=1j1FmtXKSOD5e6HWBBd_3cwI2M11Jd_nB&export=download&confirm=t&uuid=087cc973-fe39-4ba9-9898-8186149f7cf8&at=AHV7M3di_wo97d4H-uuZZyGpWzHT:1669749786439" >/dev/null 2>&1
    echo "Done Downloading pretrained model..."
  fi

  cd ..

  if [ ! -f "./data/custom_dataset/val/test.png" ]; then
      python tools/test.py configs/binsformer/binsformer_swinl_22k_w7_nyu.py checkpoints/binsformer_swinl_22k_w7_nyu.pth --show-dir "./results"
  fi
else
  git clone https://github.com/zhyever/Monocular-Depth-Estimation-Toolbox.git

  # need to edit the following files to run the script in one go, may have to run "pip install -e ." afterwards
  echo ""
  echo "###########"
  echo "Need to edit the following:"
  echo "    - setup.py file comment out all requires"
  echo "    - setup.py add in 'Python 3.9' after 'Python3.8'"
  echo "    - depth/__init__.py change -> MMCV_MAX = '1.5.0' to MMCV_MAX = '1.7.0'"
  echo "    - configs/binsformer/binsformer_swinl_22k_w7_nyu.py change ln 79 -> data_root = 'data/nru/' to data_root = 'data/custom_dataset/'"
  echo "Once that is done rerun '. ./install'"
fi

Hopefully, this helps others.

[mexicantexan]

The errors I was getting were stemming from having conflicting libs i.e. mmcv-full being installed with mmcv in that instance only one is supposed to be installed