Visual Diagnostics Research

PersenseAI

Explorations in visual diagnostics and perception-system reasoning.

Public-safe research concepts for understanding how Computer Vision systems behave across video, logs, tracks, and model signals.

Synthetic PersenseAI video diagnostic showing bounding boxes, occlusion, track paths, and an ID switch verdict.
ID switch after occlusion embedding spike · bbox jump · frame gap

Concept workflow

From observation noise to research-oriented diagnosis.

Concept flow from video and logs to features, anomaly signals, diagnostic reasoning, and analysis summary.
Input

Video + Logs + Tracks

Extract

Feature Table

Detect

Anomaly Signals

Reason

Diagnostic Reasoning

Output

Analysis Summary

Failure cards

Exploring common CV failure modes.

ID Switch

identity flips

ReID Mismatch

embedding conflict

Occlusion

lost visibility

Detection Drift

box decay

Camera Handoff

handover gap

Domain Shift

scene change

Signals

Feature signals become evidence.

feature_table.track_017
featurevaluerisk
frame_gap6
bbox_jump0.82
confidence_drop-0.41
embedding_distance0.91
motion_consistency0.28
occlusion_score0.77
anomaly_score0.93
why suspicious?

Suspicious track

embedding_distance+0.34
bbox_jump+0.27
frame_gap+0.18
occlusion_score+0.14
motion_consistency-0.07

Analysis summary

Research evidence for iterative CV investigation.

Analysis summary mockup with ID switch hypothesis, evidence, and feature contribution chart.
VISUAL ANALYSIS NOTEseverity: high
Likely hypothesisID switch after occlusion
Evidenceembedding spike + bbox jump + frame gap
Recommended next probethreshold / ReID gallery / occlusion interval

Demo slot

Future interactive diagnostics replay.

Placeholder for a short concept walkthrough: inspect → compare signals → reason → summarize.

PersenseAI

Lab notes for visual reasoning in perception systems.

hello@persenseai.com persenseai.com Public research concept site. More references coming soon.