Conda environment installation error about anyloc HOT 3 CLOSED

Hey @Mui33,

Are you trying to replicate our work or only use the AnyLoc method for estimating global descriptors?

Estimating Global Descriptors

I'd suggest you check out the demos in this case. It contains a script (and a Colab notebook) to do just the same.

In fact, you can now also use torch.hub for doing this. See this comment for more information. It has the following dependencies

dependencies = [
    'torch', 'torchvision',     # Basic PyTorch
    'einops',       # Functionalities (utilities)
    'fast_pytorch_kmeans', # KMeans clustering (in VLAD)
]

You can install them using conda, mamba, or even pip (however you choose). This is going to be our preferred way of using AnyLoc.

Replicating our work

Though the Getting Started README suggests that You can also use an existing conda environment, I suggest you use a separate environment for this purpose (to avoid any conflicts with already installed packages).

Let me know if I have answered your concerns.

On a side note (for future reference), our installer runs this

from anyloc.

Just saw that it downloads a separate cudatoolkit (11.8)

This doesn't seem interfere with the CUDA toolkit for PyTorch (for our versions at least)

The CUDA conflict causes no problems running scripts in AnyLoc on our end.

from anyloc.

Yes, I'm trying to replicate your work. Thank you for your validation and answers! @TheProjectsGuy

Since I had other reported errors during my replication, I've been worried that maybe a conda environment conflict was causing the error. Thank you for helping me to eliminate this concern.

Maybe you guys could modify the installation code for the faiss package. Because the code conda_raw_install -c pytorch faiss-gpu==1.7.2 does not install the package correctly.

from anyloc.