Skip to content

Gaze Leadership

What It Is

Gaze Leadership identifies which participant's gaze shifts are followed by others in a scene. When a participant is the first to look at an object that subsequently becomes a joint attention target, they receive a leadership credit. This metric quantifies initiative-taking in group visual attention.

Research Context

Gaze leadership is relevant to studies of group dynamics, social hierarchy, and collaborative decision-making. In leadership research, the person who consistently directs others' attention is often the de facto leader, even in leaderless groups. Measuring who initiates shared gaze targets provides a behavioural proxy for influence and social dominance.

How MindSight Detects It

MindSight supports two detection modes: object-based (default) and tip-convergence (optional).

Object-Based Leadership (default)

  1. Track first-looker: When an object class gains its first viewer (nobody was looking at it in the previous frame), record (face_idx, frame_no) as the first-look entry.
  2. Award on joint attention: When a joint attention event fires on that object, award a leadership credit to the recorded first-looker.
  3. Clear record: The first-look entry is removed after awarding to prevent credit accumulation on sustained joint attention.

Tip-Convergence Leadership (optional)

  1. Buffer each participant's gaze tip positions over the last tip_lag frames.
  2. When a new convergence cluster appears (not sustained from the previous frame), find who arrived near the centroid earliest by scanning their tip buffer.
  3. Award a leadership credit to the earliest arriver.
flowchart TD
    A[New frame arrives] --> B{Any object with\nnew first viewer?}
    B -- Yes --> C[Record first-looker\nface_idx, frame_no]
    B -- No --> D{Joint attention\nevent fires?}
    C --> D
    D -- Yes --> E[Award leadership credit\nto first-looker]
    E --> F[Clear first-look record\nfor that object class]
    D -- No --> G[Continue to next frame]
    F --> G

Parameters

Flag Type Default Description
--gaze-leader bool False Enable gaze leadership tracking.
--gaze-leader-tips bool False Also detect leadership via gaze-tip convergence (requires --gaze-tips).
--gaze-leader-tip-lag int 15 Lookback frames for tip-arrival priority.

Output

CSV -- Section gaze_leadership with columns: category, participant, object, frames_active (credit count), total_frames, value_pct.

Dashboard -- Panel titled "GAZE LEADERSHIP" ranking participants by credit count (e.g., P0: 5 events).

Console -- Prints a dictionary of leadership counts per participant, e.g., Gaze leadership counts: {'P0': 5, 'P1': 2}.

Time-series -- gaze_leadership_max: step chart of the maximum leadership credit across all participants over time.

Example

python MindSight.py --source video.mp4 --gaze-leader --joint-attention
  • Joint Attention -- Triggers leadership credits; gaze leadership has no effect without joint attention events.
  • Gaze Following -- Provides a complementary leader-follower view: gaze following tracks who follows whom, while gaze leadership tracks who initiates shared targets.

Source: Phenomena/Default/gaze_leadership.py