GeoBench SpaceNet7

Intro

SpaceNet7 (Van Etten et al. 2021) is a multi-temporal building footprint tracking dataset using time-series Planet Dove satellite imagery. The dataset provides monthly time series of high-resolution optical imagery with building footprint annotations, for building detection and tracking over time.

Dataset Characteristics

• Modalities:

  • Planet satellite imagery

• Spatial Resolution: 4m (Planet PlanetScope)

• Spectral Bands:

  • RGB: 3 channels (Red, Green, Blue)

• Image Dimensions: 512x512 pixels per patch

• Labels:

  • Binary segmentation (building vs. non-building)

  • Monthly building status tracking

• Geographic Distribution: Global coverage (100+ cities)

GeoBenchV2 Processing Pipeline

Preprocessing Steps

1. Patch Generation

   • The original 1024x1024px tiles are separated into 4 512x512px patches

2. Split Generation:

   • train/val/test splits were generated by separating the different areas of interests (AOI) so that the sets are disjoint given the location

References

1. Van Etten, A., Hogan, D., Martinez-Manso, J., Shermeyer, J., Weir, N., & Lewis, R. (2021). The multi-temporal urban development SpaceNet dataset. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 6398-6407..

import os
from pathlib import Path

import torch

from geobench_v2.datamodules import GeoBenchSpaceNet7DataModule
from geobench_v2.datasets import GeoBenchSpaceNet7
from geobench_v2.datasets.normalization import SatMAENormalizer, ZScoreNormalizer
from geobench_v2.datasets.visualization_util import (
    compare_normalization_methods,
    compute_batch_histograms,
    plot_batch_histograms,
    plot_channel_histograms,
    visualize_segmentation_target_statistics,
)

%load_ext autoreload
%autoreload 2

DATA_ROOT = Path("../../data/spacenet7")

STATS_SATMAE_PATH = os.path.join(DATA_ROOT, "spacenet7_stats_satmae.json")
STATS_CLIP_RESCALE_PATH = os.path.join(DATA_ROOT, "spacenet7_stats_clip_rescale.json")

band_order = GeoBenchSpaceNet7.band_default_order

datamodule = GeoBenchSpaceNet7DataModule(
    img_size=256,
    batch_size=4,
    num_workers=4,
    root=DATA_ROOT,
    band_order=band_order,
    data_normalizer=torch.nn.Identity(),  # we do custom normalization in the tutorial
)
datamodule.setup("fit")

Using provided pre-initialized normalizer instance: Identity
Using provided pre-initialized normalizer instance: Identity

sample_dist_fig = datamodule.visualize_geospatial_distribution(s=20)

Dataset Statistics

Image Statistics

fig = plot_channel_histograms(STATS_SATMAE_PATH)

Target Statistics

figs = visualize_segmentation_target_statistics(STATS_SATMAE_PATH, "SpaceNet7")

Raw Batch Image Statistics

Statistics of a single batch before any normalization

# Get a batch of data from the dataloader
train_dataloader = datamodule.train_dataloader()
raw_batch = next(iter(train_dataloader))

raw_batch_stats = compute_batch_histograms(raw_batch, n_bins=100)

raw_figs = plot_batch_histograms(
    raw_batch_stats, band_order, title_suffix=" (Raw Data)"
)

Effect of different Normalization Schemes

zscore_normalizer = ZScoreNormalizer(STATS_CLIP_RESCALE_PATH, band_order)
satmae_normalizer = SatMAENormalizer(STATS_SATMAE_PATH, band_order)

norm_fig, normalized_batches = compare_normalization_methods(
    raw_batch, [zscore_normalizer, satmae_normalizer], datamodule
)

Visualize Batch

fig, batch = datamodule.visualize_batch()