Hi,I can’t figure out how to load "the depth completion model trained on ScanNet",or more specific,the "data.pkl" about dense_depth_priors_nerf

[barbararoessle]

Hi, it was trained with pytorch version 1.9.0+cu111
Maybe the information here helps.

[Karbo123]

@barbararoessle
I had the same problem here. I have tried pytorch version like 1.8.0+cu111, 1.9.0+cu111, 1.10.2+cu111, 1.12.1+cu116, but unfortunately none of these worked. It is very likely that the checkpoint has been corrupted. By the way, I cannot extract the 20211027_092436.tar on Linux platform, and got the following messages when extracting:

tar: This does not look like a tar archive
tar: Skipping to next header
tar: Exiting with failure status due to previous errors

Strangely, I can extract the tar file on Win11. I am not sure what is wrong with the pretrained model downloaded here.

When I try to load the model weight from file, I always got the message like this, despite different pytorch versions:

_pickle.UnpicklingError: A load persistent id instruction was encountered,
but no persistent_load function was specified.

I load the checkpoint with something like: torch.load("./experiments_denseDepthPriors/20211027_092436/archive/data.pkl").

I have no idea on the way to solve it, I would truly appreciate if you can help me.

---

Update

I found a fix. It should be mentioned in the README that the checkpoint file must be renamed from "20211027_092436.tar" to "20211027_092436.pkl", and just simply load the "20211027_092436.pkl".