Fastflow lightning CLI method config: No action for destination key OR 'Namespace' object has no attribute 'category'

[ghost]

Describe the bug

I use a custom dataset to train Fastflow with the new Lightning CLI Method. If I define 'category' in the config there is an Error:

anomalib: error: Parser key "data": Problem with given class_path "anomalib.data.Folder":
 - 'Configuration check failed :: No action for destination key "category" to check its value.'

When I comment 'category' out in the config, there is an other AttributeError:

File "/home/USERNAME/masterthesis/anomalib/anomalib/utils/cli/cli.py", line 142, in __set_default_root_dir
    category = config.data.init_args.category if config.data.init_args.keys() else ""
AttributeError: 'Namespace' object has no attribute 'category'

To Reproduce

Steps to reproduce the behavior:

• Create config for Fastflow with custom Dataset:

# This is the config file for the new Anomalib CLI.
# To use this, run the following command:
# anomalib fit --config ./config/model/fastflow.yaml
ckpt_path: null
seed_everything: null
data:
  class_path: anomalib.data.Folder
  init_args:
    root: ./synthetic_data/anomalib_lego_dataset
    category: lego # not used in anomalib.data.Folder
    normal_dir: good
    abnormal_dir: abnormal
    task: segmentation
    mask_dir: ./synthetic_data/anomalib_lego_dataset/mask
    extensions: ['.jpg']
    normal_test_dir: null
    split_ratio: 0.2
    seed: 0
    image_size: [256,256] # options: [256, 256, 448, 384] - for each supported backbone
    train_batch_size: 16
    test_batch_size: 16
    num_workers: 8
    transform_config_train: null
    transform_config_val: null
    create_validation_set: true
    #tiling:
      #apply: false
      #tile_size: null
      #stride: null
      #remove_border_count: 0
      #use_random_tiling: False
      #random_tile_count: 16

model:
  class_path: anomalib.models.Fastflow
  init_args:
    input_size: [256, 256]
    backbone: wide_resnet50_2 # options: [resnet18, wide_resnet50_2, cait_m48_448, deit_base_distilled_patch16_384]
    pre_trained: true
    flow_steps: 8 # options: [8, 8, 20, 20] - for each supported backbone
    conv3x3_only: False # options: [True, False, False, False] - for each supported backbone
    hidden_ratio: 1.0 # options: [1.0, 1.0, 0.16, 0.16] - for each supported backbone

...

• Call anomalib fit --config anomalib_model_config/fastflow_lightning.yaml
  Output is:

anomalib: error: Parser key "data": Problem with given class_path "anomalib.data.Folder":
 - 'Configuration check failed :: No action for destination key "category" to check its value.'

• Modify Config File by commenting 'category' out and call anomalib fit --config anomalib_model_config/fastflow_lightning.yaml
  Output is:

File "/home/USERNAME/masterthesis/anomalib/anomalib/utils/cli/cli.py", line 142, in __set_default_root_dir
    category = config.data.init_args.category if config.data.init_args.keys() else ""
AttributeError: 'Namespace' object has no attribute 'category'

Expected behavior

• Training should work by leaving out 'category'

Hardware and Software Configuration

• OS: [Ubuntu 20.04]
• NVIDIA Driver Version [515.65.01]
• CUDA Version [11.7]
• CUDNN Version [V10.1.243]

[samet-akcay]

Thanks for reporting this @dominikgraf. New LightningCLI is still beta, and we expect this sort of issues. We'll try to fix it asap

[ghost]

Hey again, the bug is still present in the current version, so I took a second look.

The training works after I made a small change to anomalib/anomalib/utils/cli/cli.py and my config.
I removed 'category' from the config and removed the keys check in line 146 in cli.py:

Changed from:
146 category = config.data.init_args.category if config.data.init_args.keys() else ""
to:
146 category = ""

[samet-akcay]

@dominikgraf, you are right. The error is because category is specified in your config. The new CLI is much stricter when it comes to passing the argument. If they are not part of the class args, it immediately throws error.

In this case, category is not one of the class arguments of the FolderDataset, which is why you encountered this issue.

You don't even need to change the code. If you remove category from your config file, it should work.

[ghost]

Thanks for your answer, @samet-akcay . I tried removing / commenting out 'category' as described in the lower part of my report:

• Modify config file by removing / commenting out 'category' and call anomalib fit --config anomalib_model_config/fastflow_lightning.yaml again
  Output is:

File "/home/USERNAME/masterthesis/anomalib/anomalib/utils/cli/cli.py", line 142, in __set_default_root_dir
    category = config.data.init_args.category if config.data.init_args.keys() else ""
AttributeError: 'Namespace' object has no attribute 'category'

To me it seems that, the check of 'keys()' is skipped for some reason.
Removing 'category' from my config did not solve the Error.