Purpose: It is still troublesome to configure the NeuralRecon environment recently. When the above problems occur when using conda to install and configure the environment, write them down for future reference.
When configuring NeuralRecon, you need to install a module torchsparse, and its github library is as follows (search for torchsparse):
Installing it asks to install, then the problem occurs as:
ImportError: No module named 'backports'
but if pip install backports doesn't work. Because I can't find this module, look at the FAQ.
My computer is the cuda version of nvcc:
The cuda version corresponding to torch is:
11.3 and 11.4, I thought it was the problem. troubled for a long time. Don't want to reinstall. cuda version of nvcc. Torch does not yet have a cuda version of 11.4.
Therefore, I downloaded the source code git clone to the local at that time, and found the problem. Its requirement.txt is .
It needs backport.cached_property
. Therefore, pip install backport.cached_property is required
. If it still doesn’t work, I found one on the Internet
[External link image transfer failed, the source site may have an anti-leeching mechanism, it is recommended to save the image and upload it directly (img-AgVqakyV- 1666703282161)(http 
so that you can use pip setup.y install to compile and install torchsparse locally.