Skip to content

[Task]: How to perform training without validation  #1590

Closed

Description

@Abhijeet99999

Thank you !~

Activity

  1. willyfh commented on Jan 8, 2024

    @willyfh
    Contributor

    Hello @Abhijeet99999 , by setting the val_split_mode to none, you will prevent Anomalib from validating your model during training. However, even if you don't have any anomaly mask, it is recommended to set val_split_mode to synthetic. This will generate synthetic anomalous images for validation using Perlin noise. By having the validation process, an anomaly score threshold can be adaptively selected, and then used during inference. Similarly, you can also skip the testing phase by setting the test_split_mode to none.

  2. Abhijeet99999 commented on Jan 9, 2024

    @Abhijeet99999
    Author

    By Changing code flow, its possible as well. Thank you very much, brother!~

  3. changed the title [-][Task]: How to perform training without validation ? [/-] [+][Task]: Good Explanation[/+] on Jan 9, 2024
  4. changed the title [-][Task]: Good Explanation[/-] [+][Task]: How to perform training without validation [/+] on Jan 9, 2024
  5. Abhijeet99999 commented on Jan 10, 2024

    @Abhijeet99999
    Author

    This will generate synthetic anomalous images for validation using Perlin noise. Where you define perline noise function in the code ?

  6. samet-akcay commented on Jan 10, 2024

    @samet-akcay
    Contributor

    When you set the val_split_mode to synthetic, the code checks it and generates the synthetic anomalies here
    https://github.com/openvinotoolkit/anomalib/blob/470bb860d8478298fdf3f10a530184f8cdc9c938/src/anomalib/data/base/datamodule.py#L156-L159

  7. locked and limited conversation to collaborators on Jan 10, 2024
  8. converted this issue into a discussion #1609 on Jan 10, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions