Saved in:
Bibliographic Details
Main Authors: Tang, Jason, Law, Stephen
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.22103
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908990207688704
author Tang, Jason
Law, Stephen
author_facet Tang, Jason
Law, Stephen
contents Street-view perception models predict subjective attributes such as safety at scale, but remain correlational: they do not identify which localized visual changes would plausibly shift human judgement for a specific scene. We propose a lever-based interventional counterfactual framework that recasts scene-level explainability as a bounded search over structured counterfactual edits. Each lever specifies a semantic concept, spatial support, intervention direction, and constrained edit template. Candidate edits are generated through prompt-conditioned image editing and retained only if they satisfy validity checks for same-place preservation, locality, realism, and plausibility. In a pilot across 50 scenes from five cities, the framework reveals preliminary proxy-based directional patterns and a practical failure taxonomy under prompt-only editing, with Mobility Infrastructure and Physical Maintenance showing the largest auxiliary safety shifts. Human pairwise judgements remain the ground-truth endpoint for future validation.
format Preprint
id arxiv_https___arxiv_org_abs_2604_22103
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle How Many Visual Levers Drive Urban Perception? Interventional Counterfactuals via Multiple Localised Edits
Tang, Jason
Law, Stephen
Computers and Society
Computer Vision and Pattern Recognition
Street-view perception models predict subjective attributes such as safety at scale, but remain correlational: they do not identify which localized visual changes would plausibly shift human judgement for a specific scene. We propose a lever-based interventional counterfactual framework that recasts scene-level explainability as a bounded search over structured counterfactual edits. Each lever specifies a semantic concept, spatial support, intervention direction, and constrained edit template. Candidate edits are generated through prompt-conditioned image editing and retained only if they satisfy validity checks for same-place preservation, locality, realism, and plausibility. In a pilot across 50 scenes from five cities, the framework reveals preliminary proxy-based directional patterns and a practical failure taxonomy under prompt-only editing, with Mobility Infrastructure and Physical Maintenance showing the largest auxiliary safety shifts. Human pairwise judgements remain the ground-truth endpoint for future validation.
title How Many Visual Levers Drive Urban Perception? Interventional Counterfactuals via Multiple Localised Edits
topic Computers and Society
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2604.22103