metadata

dataset_info:
  features:
    - name: images
      dtype: image
    - name: metadata
      struct:
        - name: corners
          sequence:
            sequence: float64
        - name: tile_coords
          sequence: int64
        - name: tile_metadata
          struct:
            - name: bbox
              sequence: float64
            - name: crs
              dtype: string
            - name: edge_in_meters
              dtype: int64
            - name: resolution
              dtype: int64
        - name: transformed_trees
          list:
            - name: latitude
              dtype: float64
            - name: longitude
              dtype: float64
            - name: radius
              dtype: int64
            - name: x
              dtype: int64
            - name: 'y'
              dtype: int64
        - name: trees
          list:
            - name: area
              dtype: int64
            - name: e
              dtype: float64
            - name: height
              dtype: float64
            - name: latitude
              dtype: float64
            - name: longitude
              dtype: float64
            - name: volume
              dtype: float64
    - name: results
      dtype: image
  splits:
    - name: train
      num_bytes: 10822236808.976
      num_examples: 16848
  download_size: 11417810326
  dataset_size: 10822236808.976
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
tags:
  - geography
  - trees
  - tree
  - satellite
  - spacial
  - geospatial
  - city
  - wroclaw
size_categories:
  - 10K<n<100K
pretty_name: Satellite Trees of Wroclaw city 2022

Dataset Card for Satellite Trees Wroclaw 2022

The Satellite Trees Wroclaw 2022 dataset contains high-resolution satellite imagery and metadata of trees in Wroclaw, Poland, collected in 2022. The dataset is organized into three main directories: images, metadata, and results.

images/: Contains orthophotomaps of different regions in Wroclaw.
metadata/: Contains JSON files with metadata for each tile, including information about the trees in the corresponding satellite images.
results/: Contains examples of results with trees painted on each image. (using the center of tree and radius derived from area)

This dataset can be used for various tasks such as tree detection, classification, and other geospatial tasks!

Dataset Details

Metadata Description

Each JSON file in the metadata/ directory contains information about the trees in the corresponding satellite image tile. Structure of json object:

tile_coords: list of x, y, z of tile from specified orthophotomaps.
tile_metadata:
- bbox: bounding box of a tile "bbox = left,bottom,right,top"
- resolution: resulution in pixels of image
- crs: coordinate reference system
- edge_in_meters: how many meters does the edge of tile has.
corners: list of corners of a tile in order: left-bottom, left-top, right-top, right-bottom. [Longitute, Latitude]
trees: list of tree details from specified source
- height: Height of the tree.
- e: Eccentricity of the tree. (not confirmed!)
- volume: Volume of the tree. m^3
- area: Area covered by the tree. m^2
- latitude: Latitude coordinate of the tree.
- longitude: Longitude coordinate of the tree.
transformed_trees: list of trees after transformation to image space in pixels with radius calculated from area.
- latitude: Latitude coordinate of the tree.
- longitude: Longitude coordinate of the tree.
- x: X-coordinate in the image space.
- y: Y-coordinate in the image space.
- radius: Radius of the tree in pixels, calculated from the area.

Metadata Example

{
    "tile_coords": [
        143378,
        87608,
        18
    ],
    "tile_metadata": {
        "bbox": [
            16.89971923828125,
            51.13024583390033,
            16.901092529296875,
            51.131107637580136
        ],
        "resolution": 1024,
        "crs": "CRS:84",
        "edge_in_meters": 96
    },
    "corners": [
        [
            16.89971923828125,
            51.13024583390033
        ],
        [
            16.901092529296875,
            51.13024583390033
        ],
        [
            16.901092529296875,
            51.131107637580136
        ],
        [
            16.89971923828125,
            51.131107637580136
        ]
    ],
    "trees": [
        {
            "height": 8.05,
            "e": 1.2,
            "volume": 239.54,
            "area": 27,
            "latitude": 51.13105191475769,
            "longitude": 16.89974462238265
        },
        {
            "height": 9.49,
            "e": 1.27,
            "volume": 311.35,
            "area": 62,
            "latitude": 51.13101159452683,
            "longitude": 16.899798270669734
        },
        ...
    ],
    "transformed_trees": [
        {
            "latitude": 51.13105191475769,
            "longitude": 16.89974462238265,
            "x": 18,
            "y": 66,
            "radius": 31
        },
        {
            "latitude": 51.13101159452683,
            "longitude": 16.899798270669734,
            "x": 58,
            "y": 114,
            "radius": 47
        },
        ...
    ]
}

Of course you can extract more info about trees and place them into the image by iterating through the original trees list and modifying the transformed one.

Dataset Sources

Repository: Code Repository

Dataset Creation

Dataset was generated by iterating the maximum possible zoom of tile for chosen orthophotomaps (zoom: 18) in x and y directions. We downloaded each tile as an image with 1024x1024 resolution. We calculated the lat long coordinates for futher calculations. After having corners of tile we could get trees details from public api. We had to make some transformations to be able to draw trees on the images.

Annotation process

We believe that ground truth annotations are legit. (More inside github repository) GIS Wroclaw Trees data

Authors

Filip Strózik
Dawid Wolkiewicz
Izabela Majchrowska