search

SamsungLabs / ritm_interactive_segmentation

Reviving Iterative Training with Mask Guidance for Interactive Segmentation
MIT License
632 stars 125 forks source link

Clarify the dependency on python >= 3.6.1 in the README #6

Open drewm1980 opened 3 years ago

drewm1980 commented 3 years ago

I did an install in a fresh conda environment and hit a compatibility issue with pytorch:

python: symbol lookup error: /home/awagner/miniconda3/envs/ritm/lib/python3.6/site-packages/torch/lib/libtorch_python.so: undefined symbol: PySlice_Unpack

The pytorch version that got installed is 1.4.0 (which is stated in the README to be the correct version):

$ python Python 3.6.0 | packaged by conda-forge | (default, Feb 9 2017, 14:36:55) [GCC 4.8.2 20140120 (Red Hat 4.8.2-15)] on linux Type "help", "copyright", "credits" or "license" for more information.

import torch torch.file '/home/awagner/miniconda3/envs/ritm/lib/python3.6/site-packages/torch/init.py' torch.version '1.4.0'

PySlice_Unpack is a symbol in the python API, added in Python 3.6.1:

https://docs.python.org/3/c-api/slice.html#c.PySlice_Unpack

The REAME states that python 3.6 is required, but it seems you actually need at least 3.6.1.

I confirmed this by conda install python=3.6.1. The bindings no longer error out and the gui appears.