search

facebookresearch / pytorch3d

PyTorch3D is FAIR's library of reusable components for deep learning with 3D data
https://pytorch3d.org/
Other
8.81k stars 1.32k forks source link

adds join_pointclouds_as_batch() in __init__.py for v0.7.2 #1589

Closed sinAshish closed 1 year ago

sinAshish commented 1 year ago

Those who are using the pytorch3d==0.7.2 with CUDA==11.3 and torch==1.12.1, can't import join_pointclouds_as_batch even though the function is defined in pytorch3d.structures.pointclouds. So instead of manually editing the code, it's better to update the released version v0.7.2, with this update.

So that others can install it using pip

bottler commented 1 year ago

I'm afraid I don't understand this. join_pointclouds_as_batch is exported from pytorch3d.structures in main and in the currently released version 0.7.4 of pytorch3d. It may not be in version 0.7.2, but we can't change version 0.7.2. This PR is asking to add something to the main branch which is already there.

Anyone should be able to import it from pytorch3d.structures.pointclouds.

sinAshish commented 1 year ago

Yes, I think I made a blunder while creating the PR, I should have pushed the commit to the tag v0.7.2, such that the released version could have been updated.

join_pointclouds_as_batch is exported from pytorch3d.structures in main and in the currently released version 0.7.4 of pytorch3d. It may not be in version 0.7.2

I understand, but those with 0.7.2, which is the only available version on some clusters (such as Compute Canada), or are compatible with certain specific versions of CUDA/torch, it is difficult to bump the pytorch3d to latest, without messing up the entire venv. @bottler

bottler commented 1 year ago

Can't you change the downstream code?

sinAshish commented 1 year ago

Can't you change the downstream code?

yeah I thought making my life easier with pip install, but yeah I can pip install from my own fork.