GeoBench PASTIS

GeoBench PASTIS#

Intro#

PASTIS (Panoptic Agricultural Satellite Time Series)(Garnot et al. 2021) is a comprehensive multi-temporal satellite dataset for agricultural monitoring and crop type classification. The dataset combines Sentinel-1 SAR and Sentinel-2 optical time series with crop type annotations, enabling both semantic and instance segmentation of agricultural parcels. We use the PASTIS-R dataset version.

Dataset Characteristics#

  • Modalities:

    • Sentinel-1 SAR imagery (ascending and descending orbits)

    • Sentinel-2 optical imagery (multi-spectral)

  • Spatial Resolution:

    • S1: 10m ground sample distance

    • S2: 10m ground sample distance

  • Temporal Resolution: Multi-temporal time series throughout growing seasons

  • Spectral Bands:

    • S2: 10 bands (B02, B03, B04, B05, B06, B07, B08, B8A, B11, B12)

    • S1 Ascending: VV, VH polarizations + VV/VH ratio (3 bands)

    • S1 Descending: VV, VH polarizations + VV/VH ratio (3 bands)

  • Image Dimensions: 128x128 pixels per patch (1.28km x 1.28km)

  • Labels: Agricultural crop type classification and instance segmentation

    • 18 crop classes + background class

    • Both semantic segmentation and instance segmentation masks

    • Parcel-level annotations

  • Geographic Distribution: France (comprehensive agricultural coverage)

  • Temporal Coverage: Full growing seasons (2018-2019)

  • Task: Crop type classification and agricultural parcel segmentation

GeoBenchV2 Processing Pipeline#

Preprocessing Steps#

  1. Split Generation:

    • We use the prescribed splits from the PASTIS-R dataset

  2. Dataset Subsampling:

    • The final version consists of

      • 1,200 training time-series samples

      • 482 validation time-series samples

      • 496 test time-series samples

References#

  1. Garnot, V.S.F., Landrieu, L. and Chehata, N., 2022. Multi-modal temporal attention models for crop mapping from satellite time series. ISPRS Journal of Photogrammetry and Remote Sensing, 187, pp.294-305.

[9]:

import os
from pathlib import Path

import torch

from geobench_v2.datamodules import GeoBenchPASTISDataModule
from geobench_v2.datasets import GeoBenchPASTIS
from geobench_v2.datasets.normalization import SatMAENormalizer, ZScoreNormalizer
from geobench_v2.datasets.visualization_util import (
    compute_batch_histograms,
    plot_batch_histograms,
    plot_channel_histograms,
    visualize_segmentation_target_statistics,
)

%load_ext autoreload
%autoreload 2

The autoreload extension is already loaded. To reload it, use:
  %reload_ext autoreload

[10]:

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

STATS_SATMAE_PATH = os.path.join(DATA_ROOT, "pastis_stats_satmae.json")
STATS_CLIP_RESCALE_PATH = os.path.join(DATA_ROOT, "pastis_stats_clip_rescale.json")

[11]:

band_order = GeoBenchPASTIS.band_default_order

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

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

[12]:

sample_dist_fig = datamodule.visualize_geospatial_distribution()
../_images/dataset_notebooks_pastis_4_0.png

Raw Image Statistics#

Computed over the training dataset.

[13]:

fig = plot_channel_histograms(STATS_SATMAE_PATH)
../_images/dataset_notebooks_pastis_6_0.png
../_images/dataset_notebooks_pastis_6_1.png
../_images/dataset_notebooks_pastis_6_2.png

Target Statistics#

[14]:

figs = visualize_segmentation_target_statistics(STATS_SATMAE_PATH, "FLAIR2")
../_images/dataset_notebooks_pastis_8_0.png

Raw Batch Statistics#

[15]:

# 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)"
)
raw_figs

[15]:

[<Figure size 1200x500 with 1 Axes>,
 <Figure size 1200x500 with 1 Axes>,
 <Figure size 1200x500 with 1 Axes>]
../_images/dataset_notebooks_pastis_10_1.png
../_images/dataset_notebooks_pastis_10_2.png
../_images/dataset_notebooks_pastis_10_3.png

Effect of Normalization Schemes#

[16]:

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

Effect of SatMAE Normalization scheme on batch#

[17]:

satmae_normalized_batch = satmae_normalizer(raw_batch)
satmae_batch_stats = compute_batch_histograms(satmae_normalized_batch, n_bins=100)
sat_mae_norm_fig = plot_batch_histograms(
    satmae_batch_stats, band_order, title_suffix=" (SatMAE Normalized Data)"
)
sat_mae_norm_fig

[17]:

[<Figure size 1200x500 with 1 Axes>,
 <Figure size 1200x500 with 1 Axes>,
 <Figure size 1200x500 with 1 Axes>]
../_images/dataset_notebooks_pastis_14_1.png
../_images/dataset_notebooks_pastis_14_2.png
../_images/dataset_notebooks_pastis_14_3.png

Effect of ClipZ Normalization scheme on batch#

[18]:

clip_z_normalized_batch = clip_z_normalizer(raw_batch)
clip_z_batch_stats = compute_batch_histograms(clip_z_normalized_batch, n_bins=100)
clip_z_norm_fig = plot_batch_histograms(
    clip_z_batch_stats, band_order, title_suffix=" (Clip Z-Score Normalized Data)"
)
clip_z_norm_fig

[18]:

[<Figure size 1200x500 with 1 Axes>,
 <Figure size 1200x500 with 1 Axes>,
 <Figure size 1200x500 with 1 Axes>]
../_images/dataset_notebooks_pastis_16_1.png
../_images/dataset_notebooks_pastis_16_2.png
../_images/dataset_notebooks_pastis_16_3.png

Visualize Batch Data#

[19]:

fig, batch = datamodule.visualize_batch()
../_images/dataset_notebooks_pastis_18_0.png 
[ ]:
Contents