Mmdetection: compile.sh: ImportError: No module named torch.utils.cpp_extension

[stiv-yakovenko]

I have successfully installed miniconda and pytorch:

../miniconda3/bin/conda install pytorch torchvision cudatoolkit=10.0 -c pytorch

but when I run ./compile.sh from mmdetection folder, I get these compile errors:

Building roi align op...
Traceback (most recent call last):
  File "setup.py", line 2, in <module>
    from torch.utils.cpp_extension import BuildExtension, CUDAExtension
ImportError: No module named torch.utils.cpp_extension
Building roi pool op...
Traceback (most recent call last):
  File "setup.py", line 2, in <module>
    from torch.utils.cpp_extension import BuildExtension, CUDAExtension
ImportError: No module named torch.utils.cpp_extension
Building nms op...
Traceback (most recent call last):
  File "setup.py", line 5, in <module>
    from Cython.Build import cythonize
ImportError: No module named Cython.Build
Building dcn...
Traceback (most recent call last):
  File "setup.py", line 2, in <module>
    from torch.utils.cpp_extension import BuildExtension, CUDAExtension
ImportError: No module named torch.utils.cpp_extension
Building sigmoid focal loss op...
Traceback (most recent call last):
  File "setup.py", line 2, in <module>
    from torch.utils.cpp_extension import BuildExtension, CUDAExtension
ImportError: No module named torch.utils.cpp_extension
Building masked conv op...
Traceback (most recent call last):
  File "setup.py", line 2, in <module>
    from torch.utils.cpp_extension import BuildExtension, CUDAExtension
ImportError: No module named torch.utils.cpp_extension

[hellock]

You may check the pytorch version, which should be 1.1 or higher.

[stiv-yakovenko]

Solved. The root cause was that $PYTHON variable was pointing to python, not python3. On miniconda python has version 2.7

export PYTHON=/home/xrdp/miniconda3/bin/python3