LA_Windowview / README.md
Zongrong's picture
Add files using upload-large-folder tool
6e2b3dd verified
metadata
license: cc-by-4.0
task_categories:
  - image-to-3d
  - object-detection
tags:
  - building
  - urban
  - los-angeles
  - 3d-tiles
  - window-view
  - cesium
  - street-view
  - architecture
size_categories:
  - 10K<n<100K
language:
  - en
pretty_name: LA Windowview
dataset_info:
  features:
    - name: osm_id
      dtype: string
    - name: building_type
      dtype: string
    - name: image
      dtype: image
  splits:
    - name: train
      num_examples: 37569

LA Windowview

A synthetic multi-view image dataset of 1,058 mid- and high-rise buildings in Los Angeles, CA. Each building is captured from multiple viewpoints simulating window-level perspectives across different facades and floor heights, rendered from Google Photorealistic 3D Tiles via Cesium.

Dataset Summary

Item Value
Buildings 1,058
Mid-rise 1,051
High-rise 7
Total images 37,569
Image format JPEG
Building source OpenStreetMap
3D rendering Google Photorealistic 3D Tiles (Cesium)
Region Los Angeles, CA, USA

Dataset Structure

LA_Windowview/
├── LA_Building_folder/        # 1,058 GeoJSON files (one per building)
│   ├── {osm_id}.geojson
│   └── ...
└── screenshots/               # 37,569 JPEG images
    ├── {osm_id}_building_{building_id}_facade_{facade_id}_floor_{floor_idx}_view_{view_id}.jpg
    └── ...

GeoJSON Schema

Each .geojson file is a FeatureCollection with building footprint geometry and viewpoint metadata:

{
  "type": "FeatureCollection",
  "features": [{
    "type": "Feature",
    "geometry": { "type": "Polygon", "coordinates": [...] },
    "properties": {
      "source": "osm",
      "id": "1013455134",
      "building_type": "mid-rise",
      "osm_type": "commercial",
      "name": "...",
      "centroid_lon": -118.305,
      "centroid_lat": 34.090,
      "viewpoint_count": 70,
      "processed_building_id": 17702,
      "viewpoints": [
        {
          "view_id": "building_17702_facade_0_floor_0_view_0",
          "building_id": 17702,
          "facade_id": 0,
          "floor_idx": 0,
          "floor_name": "base",
          "lon": -118.30571,
          "lat": 34.09067,
          "alt_m": 100.0,
          "heading": 0.385,
          "pitch": 0.0,
          "roll": 0.0,
          "fov_deg": 60.0
        }
      ]
    }
  }]
}

Image Naming Convention

{osm_id}_building_{building_id}_facade_{facade_id}_floor_{floor_idx}_view_{view_id}.jpg
Field Description
osm_id OpenStreetMap building ID
building_id Internal building index
facade_id Facade index (0-based, one per building side)
floor_idx Floor level index of the viewpoint
view_id View index within the facade/floor combination

Data Collection

Viewpoints are programmatically placed around each building at multiple facade directions and floor heights. Screenshots are rendered using CesiumJS with Google Photorealistic 3D Tiles, capturing the building appearance as seen from a neighboring window perspective.

Camera parameters per viewpoint:

  • Altitude: 100 m above ground
  • FOV: 60°
  • Pitch / Roll: 0° (level horizon)
  • Heading: oriented toward building centroid

Usage

import json, os
from PIL import Image

# Load building metadata
with open("LA_Building_folder/1013455134.geojson") as f:
    building = json.load(f)

props = building["features"][0]["properties"]
print(props["building_type"], props["viewpoint_count"])

# Load corresponding images
osm_id = props["id"]
images = [f for f in os.listdir("screenshots") if f.startswith(osm_id)]
img = Image.open(f"screenshots/{images[0]}")
img.show()

Citation

If you use this dataset, please cite:

@dataset{la_windowview_2026,
  author    = {Zongrong},
  title     = {LA Windowview: Multi-View Window-Level Imagery of Los Angeles Buildings},
  year      = {2026},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/Zongrong/LA_Windowview}
}

License

Creative Commons Attribution 4.0 International (CC BY 4.0)