Skip to content

Data pre-processing: Negating #1

@linwaytin

Description

@linwaytin

According to https://github.com/wm-bupt/oracle-mnist?tab=readme-ov-file#get-the-data, the original data were processed by

  1. Grayscaling
  2. Negating
  3. Rescaling
  4. Extending

I found Negating is a little tricky. This step is supposed to make the strokes white, and the background black. Some images have white background and black strokes, and thus negating is needed in this case. But how do we distinguish this kind of images from the others. In particular, some images contain a lot of white parts even the strokes are black.

My current idea is using the 4 corners to determine the category of the image, because it seems those black-stroke images always have clear white background, and white-stroke images usually contain at least one black corner.

Any idea about this issue? I'm also curious if negating is necessary. I'm thinking maybe the model can accommodate both kinds of images.

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