How to pass a custom config to overwrite the configs to train a model

[shadeekam]

Describe the bug

I am trying to train a model using anomalib train where i have my own configurations for training and model cli is giving errors saying my configurations are wrong even though i used the configs from --print-config

Dataset

N/A

Model

PADiM

Steps to reproduce the behavior

1. install anomalib
2. create custom configs for training , model , dataset
3. run following command anomalib train -c ../configs/train_all.yaml --model ../configs/model.yaml

OS information

OS information:

• OS: [macOs 14.1]
• Python version: [3.11]
• Anomalib version: [source checked out ]
• I'm using a custom dataset
• Mac M3 apple silicon processor

Expected behavior

should be able to train the model

Screenshots

Pip/GitHub

GitHub

What version/branch did you use?

main

Configuration YAML

model:
  class_path: anomalib.models.Padim
  init_args:
    layers:
      - layer1
      - layer2
      - layer3
    input_size: null
    backbone: resnet18
    pre_trained: true
    n_features: null

Logs

error: Parser key "model":
  Not a valid subclass of AnomalyModule. Got value: {'model': {'name': 'padim', 'init_args': {'layers': ['layer1', 'layer2', 'layer3'], 'input_size': None, 'backbone': 'resnet18', 'pre_trained': True, 'n_features': None}}}
  Subclass types expect one of:
  - a class path (str)
  - a dict with class_path entry
  - a dict without class_path but with init_args entry (class path given previously)

Code of Conduct

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

[blueclowd]

Same here. Wondering if there is any runnable sample configuration on anomalib v1.x

[shadeekam]

Hi @blueclowd i was not able to run through terminal. But I trained a model from notebook, where i happen to key in all the params.

from anomalib.engine import Engine
from anomalib.utils.normalization import NormalizationMethod

engine = Engine(
    normalization=NormalizationMethod.MIN_MAX,
    threshold="F1AdaptiveThreshold",
    task=TaskType.SEGMENTATION,
    image_metrics=["AUROC"],
    accelerator="auto",
    check_val_every_n_epoch=1,
    limit_train_batches=1.0,
    limit_val_batches=1.0,
    limit_test_batches=1.0,
    precision=32,
    log_every_n_steps=50,
    devices=1,
    max_epochs=1,
    num_sanity_val_steps=0,
    val_check_interval=1.0,
)

engine.fit(model=model, datamodule=datamodule)

As I experienced certain configs formats are not yet updated in the updated source code

[blueclowd]

Hi @blueclowd i was not able to run through terminal. But I trained a model from notebook, where i happen to key in all the params.

from anomalib.engine import Engine from anomalib.utils.normalization import NormalizationMethod

engine = Engine(
    normalization=NormalizationMethod.MIN_MAX,
    threshold="F1AdaptiveThreshold",
    task=TaskType.SEGMENTATION,
    image_metrics=["AUROC"],
    accelerator="auto",
    check_val_every_n_epoch=1,
    limit_train_batches=1.0,
    limit_val_batches=1.0,
    limit_test_batches=1.0,
    precision=32,
    log_every_n_steps=50,
    devices=1,
    max_epochs=1,
    num_sanity_val_steps=0,
    val_check_interval=1.0,
)

engine.fit(model=model, datamodule=datamodule)

As I experienced certain configs formats are not yet updated in the updated source code

Thanks for sharing the hard work. Will try it out as well.

[Vbajeneza]

Describe the bug

I am trying to train a model using anomalib train where i have my own configurations for training and model cli is giving errors saying my configurations are wrong even though i used the configs from --print-config

Dataset

N/A

Model

PADiM

Steps to reproduce the behavior

1. install anomalib
2. create custom configs for training , model , dataset
3. run following command anomalib train -c ../configs/train_all.yaml --model ../configs/model.yaml

OS information

OS information:

• OS: [macOs 14.1]
• Python version: [3.11]
• Anomalib version: [source checked out ]
• I'm using a custom dataset
• Mac M3 apple silicon processor

Expected behavior

should be able to train the model

Screenshots

### Pip/GitHub GitHub

What version/branch did you use?

main

Configuration YAML

model:
  class_path: anomalib.models.Padim
  init_args:
    layers:
      - layer1
      - layer2
      - layer3
    input_size: null
    backbone: resnet18
    pre_trained: true
    n_features: null

Logs

error: Parser key "model":
  Not a valid subclass of AnomalyModule. Got value: {'model': {'name': 'padim', 'init_args': {'layers': ['layer1', 'layer2', 'layer3'], 'input_size': None, 'backbone': 'resnet18', 'pre_trained': True, 'n_features': None}}}
  Subclass types expect one of:
  - a class path (str)
  - a dict with class_path entry
  - a dict without class_path but with init_args entry (class path given previously)

Code of Conduct

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

Same issue here. Did you manage to resolve it?

[samet-akcay]

To pass a custom model or any component, you don't need to specify model section in the config file. These sections are only needed when a full config file is used. --model flag in the CLI already specifies that the yaml file will be for the model. So, to fix this you need to do the following:

# ~~model~~: This is not needed when passed with --model flag from the CLI
class_path: anomalib.models.Padim
init_args:
  layers:
    - layer1
    - layer2
  input_size: null
  backbone: resnet18
  pre_trained: true
  n_features: null