Infrastructure estimates
Naive vs. coverage-reweighted infrastructure estimates
The headline output of this project: for each city, what fraction of the road network has a sidewalk, paved surface, street lighting, and so on — computed two ways, so the size of the coverage-bias correction is visible rather than assumed away.
Naive estimate
The simple average of model predictions over whatever imagery happens to exist for a city. This is what most street-view urban-informatics papers report by default — and it is exactly the estimate this project expects to be biased, because imagery availability is not random with respect to road type, neighborhood income, or density.
Reweighted estimate
An inverse-probability-weighted / post-stratified estimate that corrects for a road segment’s estimated probability of having usable imagery at all (see /bias for the propensity model). The gap between naive and reweighted numbers, per city, is itself the headline finding this project is built to produce.
Target attributes
Estimates will be computed for each of the five attributes in the annotation protocol.
- Sidewalk presence
- Road surface type (paved / unpaved)
- Street vegetation (ordinal, 3-level)
- Street lighting presence
- Immediate building frontage density
Results
Not yet measured
No infrastructure estimate exists yet, naive or reweighted, for any city or attribute. This requires, in order: Mapillary imagery coverage measurement (/coverage), a trained vision model (/vision), and a fitted propensity model (/bias). None of those exist yet, so no estimate — naive or reweighted — is reported here.
see docs/data_sources.md