Dataset Viewer
Auto-converted to Parquet Duplicate
Search is not available for this dataset
The dataset viewer is not available for this split.
Server error while post-processing the rows. Please report the issue.
Error code:   RowsPostProcessingError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

BoundlessNYC: compiled city

A simulator frame at W 125th St and Lenox Ave in Harlem, with its semantic segmentation and depth

This dataset holds the runtime content of BoundlessNYC, a real-time digital twin of Manhattan, the Bronx, Brooklyn and Queens compiled from public municipal records. It contains:

  • the compiled city as binary streaming tiles;
  • the vehicle, pedestrian and prop models, textures and skies the renderer loads;
  • the built WebGL2 client.

The simulation server of the desktop release serves this folder, as does the interactive demo. The source code, compiler and Python API are in the GitHub repository mkturkcan/boundless-nyc.

Version v0.1.0 (git tag; use --revision v0.1.0 for a release-matched download)
Coverage Manhattan, the Bronx, Brooklyn, Queens
Contents building prisms with per-building attributes, streets and markings, lane and sidewalk graphs, street furniture and trees
Size about 3 GB

Every building of the four boroughs shaded by height, with the tile grid and the streaming radii

Layout

Path Contents
Content/tiles/ t_<i>_<j>.bin near tiles, far_<i>_<j>.bin far-field tiles, manifest.json, bridges.json
Content/models/ glTF vehicles (fleet24/), pedestrian bodies, outfits and animation clips (peds24/), street props (carla/), street trees (trees25/)
Content/textures/ PBR ground and facade sets (KTX2 and JPEG), eight HDRI skies, city-scale sky-occlusion bake (cityao.*)
Content/index.html, Content/assets/ built client
Content/{data,luts,fonts,basis,settings}/ campus, elevated-rail and landmark geometry; colour LUTs; Inter; Basis transcoder; graphics defaults

Each near tile holds, for one 512 m square in a metric frame (x east, z south):

  • building prisms with per-building attributes;
  • road, kerb and sidewalk surfaces;
  • lane and crosswalk markings;
  • the lane and sidewalk graphs for traffic and pedestrians;
  • street-furniture placements and tree records.

Public records, the compiled city, the real-time render and its ground truth for one block of Harlem

The client decodes tiles in boundlessjs/src/world/tiledata.js. boundlessjs/tools/pipeline/compile.mjs produces them.

Usage

Desktop release (Windows x64). The release archive on GitHub includes this folder. To restore it in an extracted release:

hf download mehmetkeremturkcan/boundless-nyc --repo-type dataset --revision v0.1.0 --include "Content/*" --local-dir <release folder>

From source, the client needs the three binary banks (tiles, models, textures). The smaller banks are committed to the repository.

hf download mehmetkeremturkcan/boundless-nyc --repo-type dataset --revision v0.1.0 \
    --include "Content/tiles/*" --include "Content/models/*" --include "Content/textures/*" --local-dir .cache/hf

BUILDING.md in the repository explains how to place the files and how to recompile the tiles from the public sources.

Sources and licences

The tiles derive from NYC Open Data (Building Footprints, PLUTO, facade-inspection filings, Street Centerline, 2015 Street Tree Census, hydrants, parks, borough boundaries, bus shelters, LinkNYC kiosks, bicycle parking, bus lanes, the DoITT 3D Building Model), New York State open data (MTA subway entrances, NYSERDA distributed solar), OpenStreetMap, USGS 3DEP elevation, FHWA MUTCD outlines, green-roof footprints (Treglia et al., The Nature Conservancy, 2016) and Roofpedia (Wu and Biljecki, 2021).

Licences, in brief (LICENSE.md has the full table):

  • Compiled city (tiles, derived data): ODbL 1.0, © OpenStreetMap contributors. Rendered imagery may be used under any terms with that attribution.
  • Vehicle, prop and pedestrian models: CC BY 4.0, CARLA Simulator (carla.org). Motion clips (100STYLE) and bag models: CC BY 4.0.
  • Textures and skies: CC0 1.0 (Poly Haven, ambientCG). Font: SIL OFL 1.1 (Inter). Built client: MIT.

The dataset contains no Street View or other third-party photography.

Citation

@article{turkcan2024boundless,
  title   = {Boundless: Generating photorealistic synthetic data for object detection in urban streetscapes},
  author  = {Turkcan, Mehmet Kerem and Li, Yuyang and Zang, Chengbo and Ghaderi, Javad and Zussman, Gil and Kostic, Zoran},
  journal = {arXiv preprint arXiv:2409.03022},
  year    = {2024}
}
Downloads last month
1,525

Paper for mehmetkeremturkcan/boundless-nyc