Fields of The World (FTW) is a large-scale benchmark dataset designed to advance machine learning models for instance segmentation of agricultural field boundaries. This dataset supports the need for accurate and scalable field boundary data, which is essential for global agricultural monitoring, land use assessments, and environmental studies.
This repository provides the codebase for working with the FTW dataset, including tools for data pre-processing, model training, and evaluation.
Note
The Fields of The World Command Line Inferface (FTW CLI), published under the name ftw-tools, currently lives in this ftw-baselines repository due to legacy reasons. We plan to migrate the FTW CLI and related tools into an ftw-tools repository soon. Until then, the latest and most complete version of the FTW CLI still lives in ftw-baselines.
- System setup
- Predicting field boundaries
- FTW Semantic Segmentation Baseline Model
- 1. Decide which model you want to use
- 2. FTW Inference all (using
ftw inference all) - 3. Download S2 image scene (using
ftw inference download) - 4. Run inference (using
ftw inference run) - 5. Filter predictions by land cover (using
ftw inference filter-by-lulc) - 6. Polygonize the output (using
ftw inference polygonize)
- Delineate Anything
- FTW Semantic Segmentation Baseline Model
- FTW Baseline Dataset
- CC-BY vs. the full model
- Experimentation
- Notes
- Upcoming features
- Contributing
- License
To ensure consistent behavior and compatibility, use a dedicated Python virtual environment to isolate the dependencies for the FTW CLI (ftw-tools).
First, install uv if you haven't already:
# On macOS and Linux:
curl -LsSf https://astral.sh/uv/install.sh | sh
# On Windows:
powershell -c "irm https://astral.sh/uv/install.ps1 | iex"
# Or via pip:
pip install uvCreate a virtual environment:
uv venvActivate your virtual environment:
# On macOS and Linux:
source .venv/bin/activate
# On Windows:
.venv\Scripts\activateInstall ftw-tools in development mode:
uv sync --all-extras --devUsing the FTW CLI:
# Use the FTW CLI directly (no prefix needed)
ftw --help
# Run any ftw command
ftw data download --countries=Rwanda
ftw model fit -c configs/example_config.yamlFor development work with testing and linting tools:
# Run tests
uv run pytest tests/
# Set up pre-commit hooks (only run this once)
uv run pre-commit install
# Run pre-commit hooks
uv run pre-commit run --all-filesTo install the optional delineate-anything feature:
uv sync --extra delineate-anythingTo install everything (all optional dependencies):
uv sync --all-extrasTo confirm the FTW CLI is properly installed:
# Check FTW CLI
ftw --help
# Check PyTorch
uv run python -c "import torch; print('PyTorch:', torch.__version__); print('CUDA available:', torch.cuda.is_available())"
# Check geospatial stack
uv run python -c "import rasterio, geopandas; print('Geospatial stack working')"
# Check FTW CLI import
uv run python -c "from ftw_tools.cli import ftw; print('FTW CLI ready')"You should see:
You should see:
Usage: ftw [OPTIONS] COMMAND [ARGS]...
Fields of The World (FTW) - Command Line Interface
Options:
--help Show this message and exit.
Commands:
data Downloading, unpacking, and preparing the FTW dataset.
inference Running inference on satellite images plus data prep.
model Training and testing FTW models.
The following commands show the steps for using the FTW CLI to obtain the FTW model and data, and then run an inference using that model on that data, and finally polygonizing that output. This example uses a pair of Sentinel-2 (S2) scenes over Austria.
Note: Make sure you have activated your Python virtual environment before running these commands (e.g.,
source venv/bin/activate).
In order to use ftw inference cli command you need to select one of the existing pre-trained models.
The pre-trained models with descriptions are in the releases portion of the repo, see here for more details.
The string representations of the models released are defined in models/model_registry.py and are:
- 2_Class_CCBY_v1
- 2_Class_FULL_v1
- 3_Class_CCBY_v1
- 3_Class_FULL_v1
- 3_Class_FULL_singleWindow_v2
- 3_Class_FULL_multiWindow_v2
Note: If you want more control ie provide specific Sentinel2 scenes to work with follow steps 3-6 to run each part of the inference pipeline sequentially. There is the option to run step 2 all which links together the distinct inference steps. If you decide to run step 2 you will get extracted field boundaries as polygons and don't need to proceed with steps 3-6.
This single CLI call handles the complete inference pipeline: Sentinel-2 scene selection, imagery download, model inference, and polygonization. Sentinel-2 data is selected based on the crop calendar harvest dates.
ftw inference all --help
Usage: ftw inference all [OPTIONS]
Run all inference commands from crop calendar scene selection,then download,
inference and polygonize.
Options:
-o, --out PATH Directory to save downloaded inference
imagery, and inference output to [required]
-m, --model str String representation of released model name. [required]
--year INTEGER RANGE Year to run model inference over
[2015<=x<=2025; required]
--bbox TEXT Bounding box to use for the download in the
format 'minx,miny,maxx,maxy'
-ccx, --cloud_cover_max INTEGER RANGE
Maximum percentage of cloud cover allowed in
the Sentinel-2 scene [default: 20;
0<=x<=100]
-b, --buffer_days INTEGER RANGE
Number of days to buffer the date for
querying to help balance decreasing cloud
cover and selecting a date near the crop
calendar indicated date. [default: 14;
x>=0]
-f, --overwrite Overwrites the outputs if they exist
-r, --resize_factor INTEGER RANGE
Resize factor to use for inference.
[default: 2; x>=1]
--gpu INTEGER GPU to use, zero-based index. Set to -1 to
use CPU. CPU is also always used if CUDA or
MPS is not available. [default: -1]
-ps, --patch_size INTEGER RANGE
Size of patch to use for inference. Defaults
to 1024 unless the image is < 1024x1024px
and a smaller value otherwise. [x>=128]
-bs, --batch_size INTEGER RANGE
Batch size. [default: 2; x>=1]
--num_workers INTEGER RANGE Number of workers to use for inference.
[default: 4; x>=1]
-p, --padding INTEGER RANGE Pixels to discard from each side of the
patch. Defaults to 64 unless the image is <
1024x1024px and a smaller value otherwise.
[x>=0]
-mps, --mps_mode Run inference in MPS mode (Apple GPUs).
--save_scores Save segmentation softmax scores (rescaled to [0,255])
instead of classes (argmax of scores)
-h, --stac_host [mspc|earthsearch]
The host to download the imagery from. mspc
= Microsoft Planetary Computer, earthsearch
= EarthSearch (Element84/AWS). [default:
mspc]
-s2, --s2_collection [old-baseline|c1]
Sentinel-2 collection to use with
EarthSearch only: 'old-baseline' =
sentinel-2-l2a, 'c1' = sentinel-2-c1-l2a
(default). Ignored when using MSPC.
[default: c1]
-v, --verbose Enable verbose output showing STAC calls,
scene details, and download URLs.
--help Show this message and exit.
Example usage:
ftw inference all \
--bbox=13.0,48.0,13.2,48.2 \
--year=2024 \
--out=/path/to/output \
--cloud_cover_max=20 \
--buffer_days=14 \
--model=3_Class_FULL_multiWindow_v2 \
--resize_factor=2 \
--overwriteThis will create the following files in the output directory:
inference_data.tif- The downloaded and stacked Sentinel-2 imageryinference_output.tif- The raw model inference outputpolygons.parquet- The final polygonized field boundaries
Steps 3-5 all use ftw inference. We provide the inference CLI commands to allow users to run models that have been pre-trained on FTW on any temporal pair of S2 images.
ftw inference --help
Usage: ftw inference [OPTIONS] COMMAND [ARGS]...
Inference-related commands.
Options:
--help Show this message and exit.
Commands:
download Download 2 Sentinel-2 scenes & stack them in a single file...
polygonize Polygonize the output from inference
run Run inference on the stacked satellite images
You need to concatenate the bands of two aligned Sentinel-2 scenes that show your area of interest in two seasons (e.g. planting and harvesting seasons) in the following order: B04_t1, BO3_t1, BO2_t1, B08_t1, B04_t2, BO3_t2, BO2_t2, B08_t2 (t1 and t2 represent two different points in time). The ftw inference download command does this automatically given two STAC items. The Microsoft Planetary Computer Explorer is a convenient tool for finding relevant scenes and their corresponding STAC items.
To select the timeframe for the two images (Window A and Window B), we looked at the crop calendar by USDA and found the approximate time for planting and harvesting. For example, if you open the crop calendar and select China, you will find that most of the crops are planted from Feb to May, and harvested from Aug to Nov. We then put these dates as filtering parameters in the Planetary Computer Explorer. Set the cloud threshold to 10% or less. Then select a clear observation that covers the full tile.
ftw inference download --help
Usage: ftw inference download [OPTIONS]
Download 2 Sentinel-2 scenes & stack them in a single file for inference.
Options:
--win_a TEXT URL to or Microsoft Planetary Computer ID of an Sentinel-2
L2A STAC item for the window A image [required]
--win_b TEXT URL to or Microsoft Planetary Computer ID of an Sentinel-2
L2A STAC item for the window B image [required]
-o, --out TEXT Filename to save results to [required]
-f, --overwrite Overwrites the outputs if they exist
--bbox TEXT Bounding box to use for the download in the format
'minx,miny,maxx,maxy'
--help Show this message and exit.
Run this line to download our S2 scenes of interest. This line specifies a bounding box (bbox) to download a smaller subset of the data, with --bbox 13.0,48.0,13.3,48.3. If you leave that off you'll get the full S2 scenes downloaded.
ftw inference download --win_a S2B_MSIL2A_20210617T100559_R022_T33UUP_20210624T063729 --win_b S2B_MSIL2A_20210925T101019_R022_T33UUP_20210926T121923 --out inference_imagery/austria_example.tif --bbox 13.0,48.0,13.3,48.3If you are looking to download data from the FTW Baseline Dataset, you would use ftw data download. You can see an example of this lower on this README in the FTW Baseline Dataset section.
ftw inference run is the command that will run a given model on overlapping patches of input imagery (i.e. the output of ftw inference download) and stitch the results together in GeoTIFF format.
ftw inference run --help
Usage: ftw inference run [OPTIONS] INPUT
Run inference on the stacked Sentinel-2 L2A satellite images specified via
INPUT.
Options:
-m, --model str String name of released model, valid model names are defined in `model_registry.py`. [required]
-o, --out PATH Output filename for the inference imagery.
Defaults to the name of the input file name
with 'inference.' prefix.
-r, --resize_factor INTEGER RANGE
Resize factor to use for inference.
[default: 2; x>=1]
--gpu INTEGER GPU to use, zero-based index. Set to -1 to
use CPU. CPU is also always used if CUDA or
MPS is not available. [default: -1]
-ps, --patch_size INTEGER RANGE
Size of patch to use for inference. Defaults
to 1024 unless the image is < 1024x1024px
and a smaller value otherwise. [x>=128]
-bs, --batch_size INTEGER RANGE
Batch size. [default: 2; x>=1]
--num_workers INTEGER RANGE Number of workers to use for inference.
[default: 4; x>=1]
-p, --padding INTEGER RANGE Pixels to discard from each side of the
patch. Defaults to 64 unless the image is <
1024x1024px and a smaller value otherwise.
[x>=0]
-f, --overwrite Overwrite outputs if they exist.
-mps, --mps_mode Run inference in MPS mode (Apple GPUs).
--save_scores Save segmentation softmax scores (rescaled to [0,255])
instead of classes (argmax of scores)
--help Show this message and exit.
Let's run inference on the entire downloaded scene.
ftw inference run inference_imagery/austria_example.tif --model 3_Class_FULL_FTW_Pretrained.ckpt --out austria_example_output_full.tif --gpu 0 --overwriteFTW models are known to make some errors where land parcels that are not cropland (for example, pasture) are segmented as fields. You can try to filter out these errors by filtering the predicted map using a land cover/land use map. The ftw inference filter-by-lulc command filters the GeoTIFF predictions raster to only include pixels that are cropland in the land cover map.
ftw inference filter-by-lulc --help
Usage: ftw inference filter-by-lulc [OPTIONS] INPUT
Filter the output raster in GeoTIFF format by LULC mask.
Options:
-o, --out TEXT Output filename for the (filtered) polygonized data.
Defaults to the name of the input file with parquet
extension. Available file extensions: .parquet
(GeoParquet, fiboa-compliant), .fgb (FlatGeoBuf),
.gpkg (GeoPackage), .geojson / .json / .ndjson
(GeoJSON)
-f, --overwrite Overwrite outputs if they exist.
--collection_name TEXT Name of the LULC collection to use. Available
collections: io-lulc-annual-v02 (default) and esa-
worldcover
--save_lulc_tif Save the LULC mask as a GeoTIFF.
--help Show this message and exit.
You can then use the ftw inference polygonize command to convert the output of the inference into a vector format (defaults to GeoParquet/fiboa, with GeoPackage, FlatGeobuf and GeoJSON as other options).
ftw inference polygonize --help
Usage: ftw inference polygonize [OPTIONS] INPUT
Polygonize the output from inference for the raster image given via INPUT.
Results are in the CRS of the given raster image.
Options:
-o, --out PATH Output filename for the polygonized data.
Defaults to the name of the input file with
'.parquet' file extension. Available file
extensions: .parquet (GeoParquet, fiboa-
compliant), .fgb (FlatGeoBuf), .gpkg
(GeoPackage), .geojson / .json / .ndjson
(GeoJSON)
-s, --simplify FLOAT RANGE Simplification factor to use when
polygonizing in the unit of the CRS, e.g.
meters for Sentinel-2 imagery in UTM. Set to
0 to disable simplification. [default: 15;
x>=0.0]
-sn, --min_size FLOAT RANGE Minimum area size in square meters to
include in the output. Set to 0 to disable.
[default: 500; x>=0.0]
-sx, --max_size FLOAT RANGE Maximum area size in square meters to
include in the output. Disabled by default.
[x>=0.0]
-f, --overwrite Overwrite output if it exists.
--close_interiors Remove the interiors holes in the polygons.
-st, --stride INTEGER RANGE Stride size (in pixels) for cutting tif into
smaller tiles for polygonizing. Helps avoid
OOM errors. [default: 2048; x>=0]
--softmax_threshold FLOAT RANGE
Threshold on softmax scores for class
predictions. Note: To use this option, you
must pass a tif of scores (using
`--save_scores` option from `ftw inference
run`). [0<=x<=1]
-ma, --merge_adjacent FLOAT RANGE
Threshold for merging adjacent polygons.
Threshold is the percent of a polygon's
perimeter touching another polygon.
[0.0<=x<=1.0]
-ed, --erode_dilate FLOAT RANGE
Distance (in CRS units, e.g., meters) for a
morphological opening (erode then dilate)
applied to each polygon to shave spurs and
remove thin slivers. Set 0 to disable. A
good starting value is 0.5–1x the raster
pixel size. [default: 0; x>=0.0]
-de, --dilate_erode FLOAT RANGE
Distance (in CRS units, e.g., meters) for a
morphological closing (dilate then erode)
applied to each polygon to seal hairline
gaps, fill pinholes, and connect near-
touching parts without net growth. Set 0 to
disable. A good starting value is 0.5–1x the
raster pixel size. [default: 0; x>=0.0]
-edr, --erode_dilate_raster INTEGER RANGE
Number of iterations for a morphological
opening (erode then dilate) applied to
raster mask before polygonization. Set to 0
to disable. [default: 0; x>=0]
-der, --dilate_erode_raster INTEGER RANGE
Number of iterations for a morphological
closing (dilate then erode) applied to
raster mask before polygonization. Set to 0
to disable. [default: 0; x>=0]
-tb, --thin_boundaries Thin boundaries before polygonization using
Zhang-Suen thinning algorithm.
--help Show this message and exit.
Simplification factor is measured in the units of the coordinate reference system (CRS), and for Sentinel-2 this is meters, so a simplification factor of 15 or 20 is usually sufficient (and recommended, or the vector file will be as large as the raster file).
ftw inference polygonize austria_example_output_full.tif --simplify 20This results in a fiboa-compliant file named austria_example_output_full.parquet. You can then view this file in QGIS to see something similar to the following image of the sample prediction output. The polygons in red are the predicted fields.
And that's it! In 4 lines of code, you obtained an FTW model, downloaded S2 data, ran model inference on that data, and polygonized the output to have a final parquet product.
Delineate Anything is a pretrained instance segmentation which can detect and segment out individual field boundaries directly to polygons without an intermediate predictions raster. It's trained on the FBIS-22M which is a large-scale, multi-resolution dataset comprising 672,909 high-resolution satellite image patches (0.25 m – 10 m) and 22,926,427 instance masks of individual fields. The model comes in two variants: DelineateAnything and DelineateAnything-S. DelineateAnything is the full model and DelineateAnything-S is a smaller model that is faster to run (see table below for details). If you use this model in your research, please cite the Delineate Anything paper.
| Method | [email protected] | [email protected]:0.95 | Latency (ms) | Size |
|---|---|---|---|---|
| Delineate Anything-S | 0.632 | 0.383 | 16.8 | 17.6 MB |
| Delineate Anything | 0.720 | 0.477 | 25.0 | 125 MB |
You can run Delineate Anything on a single scene using the ftw inference instance-segmentation-all command or optionally on an existing local file using ftw inference run-instance-segmentation. See below for examples.
Note that inference uses patching with overlap which will result in duplicate polygons in the overlapping regions. Postprocessing is used to merge polygons via IoU and containment thresholds which are defined by the --overlap_iou_threshold and --overlap_contain_threshold parameters. For large scenes with many polygons or using a low confidence threshold, this can become computationally slow.
Example usage:
ftw inference instance-segmentation-all \
S2B_MSIL2A_20210617T100559_R022_T33UUP_20210624T063729 \
--bbox=13.0,48.0,13.2,48.2 \
--out_dir=instance-segmentation-output \
--gpu=0 \
--model=DelineateAnything \
--resize_factor=2 \
--patch_size=256 \
--max_detections=100 \
--iou_threshold=0.3 \
--conf_threshold=0.05 \
--simplify=2 \
--min_size=500 \
--close_interiors \
--overlap_iou_threshold=0.2 \
--overlap_contain_threshold=0.8 \
--overwriteUsage:
ftw inference instance-segmentation-all --help
Usage: ftw inference instance-segmentation-all [OPTIONS] INPUT
Run all inference instance segmentation commands from download and
inference.
Options:
--bbox TEXT Bounding box to use for the download in the
format 'minx,miny,maxx,maxy'
-o, --out_dir TEXT Directory to save downloaded inference
imagery, and inference output to [required]
-h, --stac_host [mspc|earthsearch]
The host to download the imagery from. mspc
= Microsoft Planetary Computer, earthsearch
= EarthSearch (Element84/AWS). [default:
mspc]
-m, --model [DelineateAnything|DelineateAnything-S]
The model to use for inference. [default:
DelineateAnything]
--gpu INTEGER RANGE GPU ID to use. If not provided, CPU will be
used by default. [x>=0]
-r, --resize_factor INTEGER RANGE
Resize factor to use for inference.
[default: 2; x>=1]
-ps, --patch_size INTEGER RANGE
Size of patch to use for inference.
[x>=128]
-bs, --batch_size INTEGER RANGE
Batch size. [default: 4; x>=1]
--num_workers INTEGER RANGE Number of workers to use for inference.
[default: 4; x>=1]
--max_detections INTEGER RANGE Maximum number of detections to keep per
patch. [default: 100; x>=1]
-iou, --iou_threshold FLOAT RANGE
IoU threshold for matching predictions to
ground truths [default: 0.1; 0.0<=x<=1.0]
-ct, --conf_threshold FLOAT RANGE
Confidence threshold for keeping detections.
[default: 0.1; 0.0<=x<=1.0]
-p, --padding INTEGER RANGE Pixels to discard from each side of the
patch. [x>=0]
-f, --overwrite Overwrites the outputs if they exist
-mps, --mps_mode Run inference in MPS mode (Apple GPUs).
-s, --simplify FLOAT RANGE Simplification factor to use when
polygonizing in the unit of the CRS, e.g.
meters for Sentinel-2 imagery in UTM. Set to
0 to disable simplification. [default: 2;
x>=0.0]
-sn, --min_size FLOAT RANGE Minimum area size in square meters to
include in the output. Set to 0 to disable.
[default: 500; x>=0.0]
-sx, --max_size FLOAT RANGE Maximum area size in square meters to
include in the output. Disabled by default.
[default: 100000; x>=0.0]
--close_interiors Remove the interiors holes in the polygons.
[default: True]
-oit, --overlap_iou_threshold FLOAT RANGE
Overlap IoU threshold for merging polygons.
[default: 0.2; 0.0<=x<=1.0]
-cot, --overlap_contain_threshold FLOAT RANGE
Overlap containment threshold for merging polygons.
[default: 0.5; 0.0<=x<=1.0]
patch. [default: 100; x>=1]
-iou, --iou_threshold FLOAT RANGE
IoU threshold for matching predictions to
ground truths [default: 0.3; 0.0<=x<=1.0]
-ct, --conf_threshold FLOAT RANGE
Confidence threshold for keeping detections.
[default: 0.05; 0.0<=x<=1.0]
-p, --padding INTEGER RANGE Pixels to discard from each side of the
patch. [x>=0]
-f, --overwrite Overwrites the outputs if they exist
-mps, --mps_mode Run inference in MPS mode (Apple GPUs).
-s, --simplify FLOAT RANGE Simplification factor to use when
polygonizing in the unit of the CRS, e.g.
meters for Sentinel-2 imagery in UTM. Set to
0 to disable simplification. [default: 2;
x>=0.0]
-sn, --min_size FLOAT RANGE Minimum area size in square meters to
include in the output. Set to 0 to disable.
[default: 500; x>=0.0]
-sx, --max_size FLOAT RANGE Maximum area size in square meters to
include in the output. Disabled by default.
[default: 100000; x>=0.0]
--close_interiors Remove the interiors holes in the polygons.
[default: True]
-oit, --overlap_iou_threshold FLOAT RANGE
Overlap IoU threshold for merging polygons.
[default: 0.2; 0.0<=x<=1.0]
-cot, --overlap_contain_threshold FLOAT RANGE
Overlap containment threshold for merging
polygons. [default: 0.8; 0.0<=x<=1.0]
--help Show this message and exit.
Download and unpack the FTW Baseline Dataset using the FTW CLI.
This will create a ftw folder under the given folder after unpacking.
ftw data download --help
Usage: ftw data download [OPTIONS]
Download and unpack the FTW dataset.
Options:
-o, --out TEXT Folder where the files will be downloaded to. Defaults
to './data'.
-f, --clean_download If set, the script will delete the root folder before
downloading.
--countries TEXT Comma-separated list of countries to download. If
'all' (default) is passed, downloads all available
countries.
--no-unpack If set, the script will NOT unpack the downloaded
files.
--help Show this message and exit.
If you had --no-unpack enabled during download, you can manually unpack the downloaded files using the unpack command.
This will create a ftw folder under the given folder after unpacking.
Usage: ftw data unpack [OPTIONS] [INPUT]
Unpack the downloaded FTW dataset. Specify the folder where the data is
located via INPUT. Defaults to './data'.
Options:
--help Show this message and exit.
To download and unpack the complete FTW Baseline Dataset, use following command:
ftw data downloadTo download and unpack the specific country or set of countries, use following command:
ftw data download --countries belgium,kenya,vietnamNote: Make sure to avoid adding any space in between the list of comma seperated countries.
Explore visualize_dataset.ipynb to know more about the dataset.
Consider using CC-BY FTW Trained Checkpoints from the release file for Commercial Purpose. For Non-Commercial Purpose and Academic purpose, you can use the FULL FTW Trained Checkpoints (See the graph below for perfrmance comparison).
We have also made FTW model checkpoints available that are pretrained only on CC-BY (or equivalent open licenses) datasets. You can download these checkpoints using the following command:
-
3 Class
wget https://github.com/fieldsoftheworld/ftw-baselines/releases/download/v1/3_Class_CCBY_FTW_Pretrained.ckpt
-
2 Class
https://github.com/fieldsoftheworld/ftw-baselines/releases/download/v1/2_Class_CCBY_FTW_Pretrained.ckpt
For details on the experimentation process, see Experimentation section.
If you see any warnings in this format:
/home/byteboogie/miniforge3/envs/ftw/lib/python3.12/site-packages/kornia/feature/lightglue.py:44: FutureWarning: `torch.cuda.amp.custom_fwd(args...)` is deprecated. Please use `torch.amp.custom_fwd(args..., device_type='cuda')` instead.
@torch.cuda.amp.custom_fwd(cast_inputs=torch.float32)This is due to outdated libraries that rely on an older version of pytorch.
Rest assured ftw won't face any issue in experimentation and dataset exploration.
Check out the Issues Section to see what we are working on and to suggest desired features.
We welcome contributions! Please fork the repository, make your changes, and submit a pull request. For any issues, feel free to open an issue ticket.
This codebase is released under the MIT License. See the LICENSE file for details.