ClarusC64/false_absence_detection_v01

ClarusC64/falseabsencedetection_v01

Dataset summary

This dataset tests whether models treat invisible entities as gone or still present. In each sequence, a target leaves the camera view. Some exits are real. Some are false absences with clear evidence that the target remains in the container.

Goal

• —check if models infer continued presence from indirect cues
• —avoid treating every disappearance as an exit
• —keep spatial grounding under occlusion and clutter

Key signals

• —absence_tag: present, stillpresent, leftscene
• —evidence_type: shadow, reflection, audio cue, motion trail, interaction trace, door state, physical constraint, none
• —false_absence_target: true when the model should infer “still present”
• —trap_flag: true when the sample is designed to lure models into assuming exit

Columns

• —sample_id – unique id per frame sample
• —split – train, valid, eval
• —modality – video
• —scene_type – indoorroom, factoryline, corridor, sports_pitch, warehouse
• —sequence_id – id for a temporal sequence
• —frame_index – index within the sequence
• —time_gap – not used here (fixed gaps inside raw video)
• —container_id – id of the main container
• —container_bounds – "xmin ymin xmax ymax"
• —boundary_type – hard, soft, porous
• —zone_id – region id inside the container
• —zone_type – sofa, door, conveyor, chute, passage, turnleft, exitdoor, leftflank, stands, robotlane, intersection, dock_exit
• —target_entity_id – tracked entity such as cat01, crate05, cart02, ball07, robot_11
• —target_visibility – visible, partial, not_visible
• —absence_tag – present, stillpresent, leftscene
• —evidence_type – none, shadow, motiontrail, interactiontrace, audiocue, doorstate, reflection, crowdreaction, physicalconstraint
• —false_absence_target – true if the model should infer continued presence from evidence
• —trap_flag – true if the scene is designed as a false-absence trap
• —label_type – baseline, falseabsencecase, true_exit
• —drift_risk – low, medium, high
• —comment – short human note

Example loading code

from datasets import load_dataset

ds = load_dataset("ClarusC64/false_absence_detection_v01")

row = ds["train"][1]
print(row["target_entity_id"], row["target_visibility"],
      row["absence_tag"], row["evidence_type"], row["false_absence_target"])