Skip to content

LineupIdentifier.md

Latest commit

 

History

History
50 lines (35 loc) · 2.45 KB

LineupIdentifier.md

File metadata and controls

50 lines (35 loc) · 2.45 KB
Raw

LineupIdentifier

1. What is it?

LineupIdentifier

2. Usage

To initialize a LineupIdentifier, you should provide: Threshold, targetLineup (optional), and target (optional). These are the parameters that govern how the decision is made.

  • threshold: the minimum distance required to select a lineup member. If the minimum distance is above this threshold, the identifier will return "rejectId".
  • targetLineup: "targetPresent" | "targetAbsent" | None. This indicates whether the lineup is target-present or target-absent. If None, the identifier will try to infer the context from the roles.
  • target: the explicit name of the suspect (guilty or innocent). This is optional and can be used to override the default behavior.

Now we can apply those settings to make decisions based on similarity matrices.

Example similarity matrix:

member_1 member_2 member_3 member_4 member_5 member_6
probe_face_1 0.2 0.5 0.8 0.3 0.8 0.3
probe_face_2 0.6 0.4 0.9 0.1 0.9 0.1

  • typically lower distance = more similar

  • Rows: probe face(s) from the current frame

  • Columns: lineup members (filenames/base names)

  • a role map derived from LineupLoader: {"member_name": "guilty_suspect" | "innocent_suspect" | "filler"}

  • Output: a compact decision summary (min distance, response type, confidence), optionally for target-present (TP) and target-absent (TA) simultaneously.

3. Strategies

  • decide_macro(sim_df, roles): Use when both guilty and innocent exist and lineup condition (TA or TP) is not set. Returns the minimum distance among fillers, TA and TP selections, TA and TP response types, and TA and TP confidences.
  • decide_tp_or_ta(sim_df, roles): Use when lineup condition is specified. Returns selection and confidence for that condition.
  • decide_for_pipeline(sim_df, lineup_loader): Auto‑select strategy for the pipeline based on what exists in the loader and specify in settings.
  • decide(sim_df): General interface when you only pass a matrix, not with a lineuploader-loaded lineup. In this general case, both targetLineup and target should be initialized in the constructor.

Return schemas: the dict shape for each strategy; how filler_min is computed. Integration: used inside pipeline’s _decide_summary.