Skip to content

Input File Formats

Jump to bottom
Melissa R Dale edited this page Aug 17, 2020 · 1 revision

This application performs analysis of provided scores. Because analysis and this application's performance relies upon correctly interpreting the data provided, it is very important to understand the file formats the application can understand.

Things that are true, regardless of data format:

  • Subject IDs: The application does its best to identify what might be a subject id. IF IDs are strictly numeric, that makes detection more challenging. If Subject IDs are included as column or row headers, ensure they are not strictly numeric.

  • Test / Train detection: The application checks filenames for "test" and "train" to determine if the data is already split into training and testing samples. Please be sure to have train and test anywhere in the filename where appropriate.

Matrix Form

This score format contains subjects along the rows and columns such that genuine scores are along the diagonal, and the impostor scores are off diagonal.

Good things to know:

  • labels are not necessary, scores along the diagonal are genuine, off diagonal scores are impostor
  • Each modality should be its own file. That file name will be what the modality is referred to as in the application.

A pictorial example of this format for the NIST Biometric Score Dataset (Face x Finger) at NIST BSSR1, with subject IDs as row and column headers.

BSSR1 Score Format

Assumptions:

  • The order of subjects is consistent along the column and rows
  • There are the same number of rows and columns
  • IF there are row or column headers identifying the subject id, that id is not in the same format as the scores (i.e. subject ids should not be floats or pure integers)

Column Form

There are more formats available when using the Column Format. Because of the many options, there are more requirements to allow the application to properly assess the data.

Important things to know:

This format is based along the concept that there are modalities along the columns, with the last column containing the label(0 impostor, 1 genuine) OR the first 2 columns must be subject ids - in which case, genuine impostor labels may be determined by checking if the first two column ids are equal (genuine) or not (imposter).

  • Each column must have a header. This is how the application knows modality titles.
  • The label for this last column should be titled label or class (capitalization does not matter)
Clone this wiki locally