Skip to content

Latest commit

 

History

History
28 lines (23 loc) · 2.2 KB

Install.md

File metadata and controls

28 lines (23 loc) · 2.2 KB
Raw

This fork aims to reproduce the results mentioned in the paper in 2023 from the pespective of a software developer

Developer environment:

  • Windows 10/11
  • Visual Studio 2022 17.5.1
  • anaconda(any version should be ok)
  • cuda SDK 11.8

The repository is inherited from [maskrcnn-benchmark](https://github.com/facebookresearch/maskrcnn-benchmark, which is completely outdated, and stopped to update since 4 years ago. Therefore we need to first get the closest pytorch installation that will make maskrcnn-benchmark build.

  1. conda create -n pytorch-1.0
  2. conda activate pytorch-1.0
  3. conda install python=3.7
  4. install pytorch 1.0 with whl
    pip install https://download.pytorch.org/whl/cu100/torch-1.0.0-cp37-cp37m-win_amd64.whl verify that the installation is sucessful python -c "import torch;print(torch.version)"
  5. install torchvision pip install https://download.pytorch.org/whl/torchvision-0.2.0-py2.py3-none-any.whl
  6. now start to build the cpp and cu files
    1. go to source code root dir
    2. python -c "import torch;from torch.utils.cpp_extension import CUDA_HOME;print (CUDA_HOME)" it is going to look for CUDA_PATH environment variable in the system, if it is not found, it outputs C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA\v10.2. pytorch 1.0 is likely built with cuda 10.2. But in my system it shows C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA\v11.8
    3. run "C:\Program Files\Microsoft Visual Studio\2022\Professional\VC\Auxiliary\Build\vcvars64.bat" so that cl.exe can be found in the path
    4. build the cpp and cu extensions by running python setup.py build_ext,and
  7. install this module,python setup.py install
  8. (hack)Now there should be an new folder build created in the root dir. For some reason the align folder under $(STR-TDSL_RootDir)/maskrcnn_benchmark/modeling/one_stage_head is not copied. Therefore on my computer i have to do the following manual copying to $(AnacondaRootDir)\envs\pytorch-1.0\Lib\site-packages\maskrcnn_benchmark-0.1-py3.7-win-amd64.egg\maskrcnn_benchmark\modeling\one_stage_head