PathwayBench: A Benchmark for Extracting Routable Pedestrian Path Network Graphs

This repository contains the PathwayBench dataset and benchmark for extracting routable pedestrian pathway graphs. The dataset includes aerial images, road graphs, road rasters, and ground truth data for multiple cities.

• Metadata: PathwayBench Croissant metadata
• License: ODbL

Installation

# create and activate the conda environment
conda create -n pathwaybench python=3.8
conda activate pathwaybench

# install the necessary packages with `requirements.txt`:
pip install -r requirements.txt

This code has been tested with Python 3.8 on Ubuntu 20.04.

Datasets

Each set of samples in the PathwayBench dataset includes five co-registered features. The filename of each set of samples and the corresponding features are listed below:

Filename	Feature Type
xxxx_aerial.png	The aerial satellite imagery.
xxxx_road.geojson	The street (road) graph.
xxxx_road.png	The rasterized street map (with additional features).
xxxx_gt_graph.geojson	The human-validated pedestrian pathway graph.
xxxx_gt_mask.png	The rasterized human-validated pedestrian pathway graph to support semantic segmentation tasks.
xxxx_gt_color.png	The color-coded version of xxxx_gt_mask.png for visualization purposes.

Below are the links to the dataset that are currently supported by PathwayBench

City	Data
Seattle, WA	Link to dataset
Washington, D.C.	Link to dataset
Portland, OR	Link to dataset
Bellevue, WA	[Will be released soon]
Quito, Ecuador	[Will be released soon]
Sao Paulo, Brazil	[Will be released soon]
Santiago, Chile	[Will be released soon]
Valparaiso, Chile	[Will be released soon]

Benchmark

PathwayBench provides utilities for evaluating graphs by the extent to which their structural characteristics align with ground truth, as described below.

Partition test area: This step partitions the entire test area into Tessellating Intersection Polygons (TIP). Each TIP is created by assigning a point location to a road intersection, then computing the associated Voronoi polygons to tessellate the entire test area. Ground Truth GeoJSON is provided for each of the support city in PathwayBench dataset.

python scripts/tessellate_area.py <Ground Truth GeoJSON>

Compute statistics per TIP: This step computes the statistics (edge-retrieval F1 score, betweenness centrality, number of connected components, TraversabilitySimilarity) for each TIP in the test area. The statistics computed in this step are useful for analyzing local graph routability. Prediction GeoJSON is the prediction graph GeoJSON, TIP GeoJSON is the area partition generated in the previous step.

python scripts/compute_stats.py <Prediction GeoJSON> <Ground Truth GeoJSON> <TIP GeoJSON>

Summarize statistics: This step aggregates the computed statistics from the previous step and provides summarizing statistics for the entire test area. The statistics computed in this step are useful for analyzing global graph routability. Prediction Stats GeoJSON and GT Stats GeoJSON are generated in the previous step. \end{enumerate}

python scripts/summarize_stats.py <Prediction Stats GeoJSON> <GT Stats GeoJSON>