🐞 cli does not allow custom input size

[ashwinvaidya17]

Describe the bug

With the pre-processor refactor, CLI does not allow setting the image size easily. If one really wants to modify it, they need to recreate the entire pre-processor

anomalib train --model Padim \
  --data MVTecAD \
  --data.category transistor \
  --model.pre_processor anomalib.pre_processing.PreProcessor \
  --model.pre_processor.transform torchvision.transforms.v2.Compose \
  --model.pre_processor.transform.transforms '[torchvision.transforms.v2.Resize([512,512]),torchvision.transforms.v2.Normalize([0.485,0.456,0.406],[0.229,0.224,0.225])]'

Or more practically, use a YAML config file:

model:
  class_path: anomalib.models.Padim
  init_args:
    backbone: resnet18
    pre_processor:
      class_path: anomalib.pre_processing.PreProcessor
      init_args:
        transform:
          class_path: torchvision.transforms.v2.Compose
          init_args:
            transforms:
              - class_path: torchvision.transforms.v2.Resize
                init_args:
                  size: [512, 512]
                  antialias: true
              - class_path: torchvision.transforms.v2.Normalize
                init_args:
                  mean: [0.485, 0.456, 0.406]
                  std: [0.229, 0.224, 0.225]
data:
  class_path: anomalib.data.MVTecAD
  init_args:
    category: transistor

Then run:
anomalib train --config config.yaml

Dataset

N/A

Model

N/A

Steps to reproduce the behavior

N/A

OS information

OS information:

• Affects all platforms

Expected behavior

Should allow a simple way to update input size

Screenshots

No response

Pip/GitHub

GitHub

What version/branch did you use?

main

Configuration YAML

n/a

Logs

n/a

Code of Conduct

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

[github-actions]

Thanks for submitting this issue! It has been added to our triage queue. A maintainer will review it shortly.

[AbhayKumarDas]

Hi @ashwinvaidya17 ! I’ve worked on this issue and opened a PR (#3482) with a fix.

The core problem was that jsonargparse only exposes parameters defined in the model’s init. Since image_size was only present in the base class (AnomalibModule) and not in individual models like Padim, STFPM, etc., the CLI rejected --model.image_size.

To fix this properly (without breaking CLI design), I followed the existing architecture pattern used for parameters like pre_processor and evaluator:

• Added image_size to all model init signatures
• Forwarded it to AnomalibModule
• Handled resizing centrally in the base module by updating the pre_processor transform

This keeps the solution aligned with current design and avoids adding CLI-level hacks.

I’ve attached a short PDF with:

1. Bug reproduction (CLI failure)
2. Implementation approach (architecture reasoning)
3. Working outputs (CLI + Python validation across models)

Validation Screenshots.pdf

Would really appreciate if you could take a look at the PR and the PDF for detailed context. Thanks!