Research dataset · APE

The APE dataset, sample imagery, annotations, and validation references.

APE (Annotations for Pedestrian Environment) was created to train and evaluate automated pedestrian-path understanding. It is a research dataset, not a direct substitute for a current, routable OS-CONNECT release.

[IMAGE NEEDED: Annotated APE sample showing aerial image, road raster and human-verified pedestrian feature mask; include research dataset citation. File: tdei-hero-ape.png]

What the researchers wanted to solve.

Pedestrian paths were missing or poorly connected in typical transportation maps. Aerial imagery and mapped street information were used to learn to identify sidewalks, crossings and corner bulbs.

Aligned APE inputs: aerial image, road-map image, annotated semantic mask and color visualization; learning produces candidate paths which require connectivity reconstruction and evaluation
APE research data: input imagery and aligned reference annotations for pedestrian feature extraction.
ComponentExample filename from original pageWhat it provides
Aerial imageryxxxx_sat_train.pngTop-down input imagery.
Raster street mapxxxx_road_train.pngAligned road-map representation.
Human-validated maskxxxx_gt_mask.pngSegmentation label/reference for sidewalks, crossings, corner bulbs.
Color visualizationxxxx_gt_color.pngHuman-readable rendering of annotation classes.

Extent, example and benchmark

The existing page reports approximately 2,700 km² of selected areas across Los Angeles, Bellevue, Seattle, Quito, São Paulo, Santiago and Greater Valparaíso. Geographic coverage does not imply complete coverage of each city.

Publication

APE paper

Yuxiang Zhang, Nicholas-J Bolten, Sachin Mehta and Anat Caspi. Describes a multi-input segmentation workflow, human annotation issues, and connected-network inference.

Open detailed page
Check access

Los Angeles sample folder

Access the APE sample data: [CONFIRM: download permission and public folder link].

Open detailed page
Evaluation

PathwayBench

Graph-level resources for co-registered road and pedestrian pathways in GeoJSON, referenced by the original APE page.

Open detailed page

Using APE for research and evaluation

Training masks, ground-truth validation data, inferred graph outputs and current operational datasets answer different questions.

QuestionUse thisCheck first
Can my model identify pedestrian features in images?APE imagery and labeled masks.Image and annotation rights, city splits, resolution, representation.
Can the model produce a connected routing network?PathwayBench or evaluated network graphs.Connectivity and error definitions beyond pixel overlap.
What are typical human annotation errors?APE paper, Fig. 1 and Tables 1a–c.Map disconnected sidewalk-to-crossing links and displaced curb nodes.
Can I deploy this map for trip planning today?Current TDEI/OS-CONNECT network release.Schema, QA/QC, vintage, destination context and rights.
Research dataset downloads

[CONFIRM: APE dataset public file list, validation split and download permissions.]