[Bug]: Failed download weights for patchcore

[LorenzoF6]

Describe the bug

I'm working with anomalib v2 on a jupyter notebook and when i create a new patchcore model i think that the download failed. Infact when i try the fit the model don't create a symbolinc path with the weights because there aren't in the results (i change the default_root_dir but with the default the outcome is the same) folder

Dataset

N/A

Model

N/A

Steps to reproduce the behavior

eng_patchcore = Engine(default_root_dir = 'risultati')
md_patchcore = Patchcore(evaluator = ev_DISCHITOP)
eng_patchcore.fit(model = md_patchcore, datamodule = datamodule_DISCHITOP)

OS information

OS information:

• OS: [e.g. Ubuntu 20.04]
• Python version: [e.g. 3.10.0]
• Anomalib version: [e.g. 0.3.6]
• PyTorch version: [e.g. 1.9.0]
• CUDA/cuDNN version: [e.g. 11.1]
• GPU models and configuration: [e.g. 2x GeForce RTX 3090]
• Any other relevant information: [e.g. I'm using a custom dataset]

Expected behavior

correct download of the weights and placed in results

Screenshots

No response

Pip/GitHub

pip

What version/branch did you use?

main

Configuration YAML

N/A

Logs

this is when i create the model :
INFO:anomalib.models.components.base.anomalib_module:Initializing Patchcore model.
INFO:timm.models._builder:Loading pretrained weights from Hugging Face hub (timm/wide_resnet50_2.racm_in1k)
INFO:timm.models._hub:[timm/wide_resnet50_2.racm_in1k] Safe alternative available for 'pytorch_model.bin' (as 'model.safetensors'). Loading weights using safetensors.
INFO:timm.models._builder:Missing keys (fc.weight, fc.bias) discovered while loading pretrained weights. This is expected if model is being adapted.

This is the error when i try the fit:
OSError                                   Traceback (most recent call last)
Cell In[26], line 2
      1 # Fit del modello : Patchcore è zero shot perchè pre addestrato
----> 2 eng_patchcore.fit(model = md_patchcore, datamodule = datamodule_DISCHITOP)

File /workspace/anomalib/src/anomalib/engine/engine.py:404, in Engine.fit(self, model, train_dataloaders, val_dataloaders, datamodule, ckpt_path)
    401 if ckpt_path:
    402     ckpt_path = Path(ckpt_path).resolve()
--> 404 self._setup_workspace(
    405     model=model,
    406     train_dataloaders=train_dataloaders,
    407     val_dataloaders=val_dataloaders,
    408     datamodule=datamodule,
    409     versioned_dir=True,
    410 )
    411 self._setup_trainer(model)
    412 if model.learning_type in {LearningType.ZERO_SHOT, LearningType.FEW_SHOT}:
    413     # if the model is zero-shot or few-shot, we only need to run validate for normalization and thresholding

File /workspace/anomalib/src/anomalib/engine/engine.py:298, in Engine._setup_workspace(self, model, train_dataloaders, val_dataloaders, test_dataloaders, datamodule, dataset, versioned_dir)
    295 # - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - #
    296 # 2. Update the default root directory
    297 root_dir = Path(self._cache.args["default_root_dir"]) / model.name / dataset_name / category
--> 298 self._cache.args["default_root_dir"] = create_versioned_dir(root_dir) if versioned_dir else root_dir / "latest"

File /workspace/anomalib/src/anomalib/utils/path.py:102, in create_versioned_dir(root_dir)
    100 if latest_link_path.is_symlink() or latest_link_path.exists():
    101     latest_link_path.unlink()
--> 102 latest_link_path.symlink_to(new_version_dir, target_is_directory=True)
    104 return latest_link_path

File /opt/conda/lib/python3.11/pathlib.py:1198, in Path.symlink_to(self, target, target_is_directory)
   1196 if not hasattr(os, "symlink"):
   1197     raise NotImplementedError("os.symlink() not available on this system")
-> 1198 os.symlink(target, self, target_is_directory)

OSError: [Errno 5] Input/output error: '/workspace/risultati/Patchcore/DISCHITOP/v1' -> '/workspace/risultati/Patchcore/DISCHITOP/latest'

Code of Conduct

• I agree to follow this project's Code of Conduct

[alexriedel1]

Hi!
I guess that #2554 is a duplicate of this?

Unfortunately I cannot reproduce the error. Are you running your code on a server? Can you tell a bit more about on what machine you are running this and what exact version of anomalib you are using?

[LorenzoF6]

Hi, yep sorry i duplicated and i forgot instead to update the #2554.
I'm working with the the latest branch main of anomalib (v2). I've installed it for development with all dependencies
(pip install -e .[full])

I'm using a jupyter notebook which is located inside a docker (i have all the privileges in that) and the docker is mounted on a private storage system which belong to the company where i am currently doing the job.

In this the 'results' folder isn't populated with the weights of patchcore or padim but if i try colab the 'results' folder is instead populated (unfortunately i cannot use colab for my work due to the limited ram). I might think that the problem is some missing permission for access to download the weights and create the correct 'results' folder but i don't know which

[alexriedel1]

I guess it's a permission issue in your docker, you should try to make sure that you have write permissions in your docker.
Anomalib tries to create a symbolic link inside your docker and fails

[LorenzoF6]

Ok thanks, yes i'm the root of the docker and have persmission of writng in it. So i think that i've to update the permission of the anomalib forlder to write and modify folders in my docker.