Skip to content

Latest commit

 

History

History
81 lines (59 loc) · 3.32 KB

PASTIS.md

File metadata and controls

81 lines (59 loc) · 3.32 KB
Raw

Panoptic Agricultural Satellite TIme-Series (PASTIS)

Submission Details

Overview

PASTIS is a benchmark dataset for panoptic and semantic segmentation of agricultural parcels from satellite time series. It contains 2,433 patches within the French metropolitan territory with panoptic annotations (instance index + semantic label for each pixel). Each patch is a Sentinel-2 multispectral image time series of variable length.

Data

  • URL: https://zenodo.org/record/5012942
  • Documentation: https://github.com/VSainteuf/pastis-benchmark/blob/main/documentation/pastis-documentation.pdf
  • File Format: Numpy files
  • Geographic Coverage: France (~4000 km2 sampled from four regions)
  • Has Train/Test split?: Yes
  • Input data sensor(s): Sentinel-2
  • Input patch size: 128x128 pixels
  • Input data channels(s): The 10 channels correspond to the non-atmospheric spectral bands of the Sentinel-2 satellite, after atmospheric correction and re-sampling at a spatial resolution of 10 meters per pixel.
  • Input data timesteps: Variable length (30-60, all available Sentinel-2 acquisitions)
  • Label information:
    • Semantic segmentation (binary masks)
    • Instance segmentation (instance masks)
    • Panoptic segmentation (full-coverage instance masks)
    • Includes crop type or category labels: 18 crop types plus a background class
    • Ground truth polygons (vector data used to create label masks)
  • Projection (if georeferenced): N/A
  • License: MIT License
  • Data Creation Details: Garnot et al., ICCV 2021
  • Label Source: France Graphical Plot Register (RPG)

...

Metadata

Splits used for 5-fold cross validation

Dataset splits

Recommends using 5-fold cross validation rather than a single train/test split.

Type of split: 5-fold CV split uniform at random by patch

Subset # samples
Total 2,433
Train n/a
Validation n/a
Test n/a

SOTA performance

Model OA F1 IoU mAP
TSViT 83.4% 65.4%

Example

pastis

API (optional)

Library/package URL
Github repo https://github.com/VSainteuf/pastis-benchmark
TorchGeo https://torchgeo.readthedocs.io/en/stable/_modules/torchgeo/datasets/pastis.html

Example

https://github.com/VSainteuf/pastis-benchmark/blob/main/demo.ipynb