UrbanVGGT: Scalable Sidewalk Width Estimation from Street View Images

Kaizhen Tan; Fan Zhang

cs.CVMar 23, 2026

UrbanVGGT: Scalable Sidewalk Width Estimation from Street View Images

Authors: Kaizhen Tan, Fan Zhang

Organizations: Heinz College of Information Systems and Public Policy, Carnegie Mellon University, USA · Institute of Remote Sensing and Geographical Information System, Peking University, China

Abstract

Sidewalk width is an important indicator of pedestrian accessibility, comfort, and network quality, yet large-scale width data remain scarce in most cities. Existing approaches typically rely on costly field surveys, high-resolution overhead imagery, or simplified geometric assumptions that limit scalability or introduce systematic error. To address this gap, we present UrbanVGGT, a measurement pipeline for estimating metric sidewalk width from a single street-view image. The method combines semantic segmentation, feed-forward 3D reconstruction, adaptive ground-plane fitting, camera-height-based scale calibration, and directional width measurement on the recovered plane. On a ground-truth benchmark from Washington, D.C., UrbanVGGT achieves a mean absolute error of 0.252 m, with 95.5% of estimates within 0.50 m of the reference width. Ablation experiments show that metric scale calibration is the most critical component, and controlled comparisons with alternative geometry backbones support the effectiveness of the overall design. As a feasibility demonstration, we further apply the pipeline to three cities and generate SV-SideWidth, a prototype sidewalk-width dataset covering 527 OpenStreetMap street segments. The results indicate that street-view imagery can support scalable generation of candidate sidewalk-width attributes, while broader cross-city validation and local ground-truth auditing remain necessary before deployment as authoritative planning data.

Figures & tables

Figure 1: Sidewalk width data in OpenStreetMap. Grey lines denote all drivable streets; blue lines (if any) denote streets with a sidewalk-width tag. Both New York City (461 streets) and Nairobi (1958 streets) have zero sidewalk-width tags, highlighting the data gap that UrbanVGGT aims to fill.

Figure 2: UrbanVGGT pipeline overview. (a) Input street-view image. (b) Semantic segmentation with inner (yellow) and outer (red) boundary detection. (c) Midline overlap region used to pair boundary points. (d) VGGT-based 3D reconstruction. (e) Ground-plane fitting with semantic point cloud. (f) Width estimation and preliminary dataset construction.

Figure 3: Qualitative measurement examples on the D.C. dataset. Each panel shows the segmentation overlay with detected inner (yellow) and outer (red) boundaries. Predicted width: model estimate; ground-truth width: reference measurement.

Variant	MAE (m)	RMSE (m)	$<$ 0.25 m (%)	$<$ 0.50 m (%)
Full pipeline	0.252	0.293	49.9	95.5
$-$ Scale calibration	1.571	1.599	0.0	0.0
Pinhole only	1.096	1.203	4.5	11.0
Full image width	0.265	0.351	54.2	88.5

Table 1: Ablation study on the Washington, D.C. dataset. Best values are shown in bold.

Camera height (m)	MAE (m)	Bias (m)	$<$ 0.25 m (%)	$<$ 0.50 m (%)
2.00	0.384	$-$ 0.380	29.7	65.5
2.25	0.283	$-$ 0.217	50.6	82.4
2.50	0.252	$-$ 0.055	49.9	95.5
2.75	0.276	$+$ 0.104	50.0	85.0
3.00	0.349	$+$ 0.263	46.5	69.1

Table 2: Camera height sensitivity analysis.

Figure 4: Camera height sensitivity: MAE as a function of the assumed camera mounting height

h_{\mathrm{cam}}

Model	MAE (m)	$<$ 0.25 m (%)	$<$ 0.50 m (%)
DepthAnythingV2-m	1.712	0.0	2.7
ZoeDepth	1.067	10.1	20.6
DepthPro	0.768	10.2	23.6
Metric3DV2	1.361	5.5	11.5
UniDepthV2	0.289	54.7	83.3
MapAnything	0.463	17.8	60.5

Table 3: Category 1: metric depth models evaluated with native scale geometry. No camera-height calibration is applied.

Model	MAE (m)	$<$ 0.25 m (%)	$<$ 0.50 m (%)
DepthAnythingV2-r	0.628	18.0	38.0
DepthPro	0.718	12.2	31.7
DPT	0.701	15.2	33.7
Metric3DV2	0.417	39.9	67.8
UniDepthV2	0.436	23.9	66.7
ZoeDepth	0.583	21.0	44.0

Table 4: Category 2: monocular depth models evaluated with pinhole unprojection and camera-height scale calibration.

Model	MAE (m)	$<$ 0.25 m (%)	$<$ 0.50 m (%)
CUT3R	0.672	24.3	47.1
MapAnything	0.334	44.7	78.0
$\pi^{3}$	0.324	49.2	79.6
UrbanVGGT (ours)	0.252	49.9	95.5

Table 5: Category 3: single-image point-cloud reconstruction models evaluated with camera-height scale calibration.

Figure 5: MAE comparison across all methods. Models are grouped by evaluation category: Category 1 (metric depth with native scale), Category 2 (monocular depth with pinhole unprojection and scale calibration), and Category 3 (single-image point-cloud reconstruction with scale calibration). All methods share the same segmentation, boundary extraction, plane fitting, and outlier filtering; only the 3D geometry backbone differs. The dashed red line indicates the UrbanVGGT MAE (0.252 m).

City	Neighbourhood	Characteristics
New York City	Midtown	Dense grid, rich Google Street View coverage
São Paulo	Jardins/Paulista	Dense district with relatively wide sidewalks
Nairobi	Central business district/Westlands	Sparser Google Street View coverage

Table 6: SV-SideWidth study areas.

City	Valid measurements	Street segments covered	Total street segments	Coverage rate (%)	Median width (m)
New York City	502	176	461	38.2	2.58
São Paulo	866	203	1539	13.2	2.64
Nairobi	563	148	1958	7.6	2.28
Total	1931	527	3958	N/A	N/A

Table 7: SV-SideWidth prototype dataset summary and OpenStreetMap coverage.

Figure 6: SV-SideWidth prototype dataset coverage maps for New York City (left), São Paulo (centre), and Nairobi (right). Grey lines indicate the OpenStreetMap street network; green lines indicate street segments with SV-SideWidth measurements.

Urban Planning

Urban planning uses geospatial data, machine learning, and optimization to analyze cities and guide transport, land use, housing, and climate decisions.

See all 48 papers

Can VLMs Reliably Assess Sidewalk Accessibility Attributes from Pedestrian-Level Imagery?
Sep 15, 2026Seung Jae Lieu, Diego Morra, Chiara Cadoni +3VLM Evaluation Urban Planning
Senseable City Lab Massachusetts Institute of Technology Cambridge, MA 02139
Safety-oriented sidewalk and road segmentation for smartphone-based assistive navigation
Jul 23, 2026Hakan Calim, Anamaria Dumitrescu, Adarsh Bhandary Panambur +2Visual Navigation Visually Impaired Navigation Assistance
Pattern Recognition Lab, Friedrich-Alexander-University Erlangen-Nuremberg, Erlangen, Germany · Department of Telecommunications, National University of Science and Technology “Politehnica” Bucharest, Romania
Seeing the City or Recognizing the Place? What Street-View Imagery Adds Beyond Existing Urban Data in VLM-Based Urban Sensing
Sep 2, 2026Kaizhen TanVLM Evaluation
New York University

Sep 15, 2026J/K move · Enter open · S save

Can VLMs Reliably Assess Sidewalk Accessibility Attributes from Pedestrian-Level Imagery?

Seung Jae Lieu, Diego Morra, Chiara Cadoni, Wonseop Song+2

Senseable City Lab Massachusetts Institute of Technology Cambridge, MA 02139

VLM Evaluation Urban Planning VLM Calibration

Open paper AskPDF arXiv

An important component of urban accessibility, particularly for wheelchair users and people with reduced mobility, is sidewalk compliance with measurable requirements. We test whether effective width, longitudinal slope, cross slope, and pavement condition can be assessed reliably from pedestrian-level imagery using vision-language models (VLMs). We present the first application of sampling-based conformal prediction (CP) for VLM-based accessibility assessment. We evaluate four VLMs on 514 sidewalk images from Seoul, South Korea, with field-measured ground truth. Conformal calibration attains the nominal 90% coverage for all models and attributes, but the calibrated regions differ in informativeness. Effective width yields the most informative estimates, with a mean interval half-width of about 1.0 m for the best model. Since every model overestimates width, asymmetric calibration shortens the intervals by up to 33% at unchanged coverage. Longitudinal slope is marginally informative, cross-slope intervals are too wide to resolve regulatory thresholds, and pavement-condition sets degenerate to all five grades (A-E) for three of the four models. Uncalibrated intervals from raw sampling dispersion cover only 17-47% of field-measured values at a nominal 90% level. Among the images with the most self-consistent responses, these intervals miss the field-measured value in up to 96% of cases. Response self-consistency is therefore not evidence of accuracy, and sampling dispersion cannot be interpreted as uncertainty until it has been calibrated against field-measured ground truth. No quantitative attribute reaches the precision required for general compliance assessment, but CP identifies from calibration data alone which attributes can support screening of segments far from the thresholds. We release the annotated pedestrian-level images and their corresponding field-measured attribute values.

Abstract

Urban Planning

Explore similar work

Can VLMs Reliably Assess Sidewalk Accessibility Attributes from Pedestrian-Level Imagery?

Safety-oriented sidewalk and road segmentation for smartphone-based assistive navigation

Seeing the City or Recognizing the Place? What Street-View Imagery Adds Beyond Existing Urban Data in VLM-Based Urban Sensing