Merge branch 'main' into plotCBF

.gitattributes

@@ -1,2 +1,4 @@
 # Do not change line endings on test data, it will change the MD5
-/tests/data/*/** binary
+/tests/data/*/** -text
+# Test data generation files are fine though
+/tests/data/**/data.py text

.github/workflows/tests.yaml

@@ -59,9 +59,7 @@ jobs:
       if: ${{ runner.os == 'Windows' }}
     - name: Install conda dependencies (Windows)
       run: |
-        # PyTorch isn't compatible with setuptools 59.6+, pin for now until new PyTorch release
-        # https://github.com/pytorch/pytorch/pull/69904
-        conda install 'fiona>=1.5' h5py 'rasterio>=1.0.16' 'setuptools<59.6'
+        conda install 'fiona>=1.5' h5py 'rasterio>=1.0.16'
         conda list
         conda info
       if: ${{ runner.os == 'Windows' }}

.pre-commit-config.yaml

@@ -24,9 +24,9 @@ repos:
             additional_dependencies: ["toml"]
 
     -   repo: https://github.com/pre-commit/mirrors-mypy
-        rev: v0.931
+        rev: v0.940
         hooks:
         -   id: mypy
             args: [--strict, --ignore-missing-imports, --show-error-codes]
-            additional_dependencies: [torch>=1.7, torchmetrics>=0.7, pytorch-lightning>=1.3, pytest>=6, omegaconf>=2.1, kornia>=0.6, numpy>=1.22.0]
+            additional_dependencies: [torch>=1.11, torchmetrics>=0.7, pytorch-lightning>=1.3, pytest>=6, omegaconf>=2.1, kornia>=0.6, numpy>=1.22.0]
             exclude: (build|data|dist|logo|logs|output)/

.readthedocs.yaml

@@ -4,14 +4,18 @@
 # Required
 version: 2
 
+# Set the version of Python
+build:
+  os: ubuntu-20.04
+  tools:
+    python: "3.9"
+
 # Configuration of the Python environment to be used
 python:
    install:
    - requirements: docs/requirements.txt
    - method: pip
      path: .
-     extra_requirements:
-     - train
 
 # Configuration for Sphinx documentation
 sphinx:

benchmark.py

@@ -77,15 +77,16 @@ def set_up_parser() -> argparse.ArgumentParser:
         "--patch-size",
         default=224,
         type=int,
-        help="height/width of each patch",
-        metavar="SIZE",
+        help="height/width of each patch in pixels",
+        metavar="PIXELS",
     )
     parser.add_argument(
         "-s",
         "--stride",
         default=112,
         type=int,
-        help="sampling stride for GridGeoSampler",
+        help="sampling stride for GridGeoSampler in pixels",
+        metavar="PIXELS",
     )
     parser.add_argument(
         "-w",
@@ -139,15 +140,11 @@ def main(args: argparse.Namespace) -> None:
         length = args.num_batches * args.batch_size
         num_batches = args.num_batches
 
-    # Convert from pixel coords to CRS coords
-    size = args.patch_size * cdl.res
-    stride = args.stride * cdl.res
-
     samplers = [
-        RandomGeoSampler(landsat, size=size, length=length),
-        GridGeoSampler(landsat, size=size, stride=stride),
+        RandomGeoSampler(landsat, size=args.patch_size, length=length),
+        GridGeoSampler(landsat, size=args.patch_size, stride=args.stride),
         RandomBatchGeoSampler(
-            landsat, size=size, batch_size=args.batch_size, length=length
+            landsat, size=args.patch_size, batch_size=args.batch_size, length=length
         ),
     ]
 
@@ -211,17 +208,15 @@ def main(args: argparse.Namespace) -> None:
     # Benchmark model
     model = resnet34()
     # Change number of input channels to match Landsat
-    model.conv1 = nn.Conv2d(  # type: ignore[attr-defined]
+    model.conv1 = nn.Conv2d(
         len(bands), 64, kernel_size=7, stride=2, padding=3, bias=False
     )
 
-    criterion = nn.CrossEntropyLoss()  # type: ignore[attr-defined]
+    criterion = nn.CrossEntropyLoss()
     params = model.parameters()
     optimizer = optim.SGD(params, lr=0.0001)
 
-    device = torch.device(  # type: ignore[attr-defined]
-        "cuda" if torch.cuda.is_available() else "cpu", args.device
-    )
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu", args.device)
     model = model.to(device)
 
     tic = time.time()
@@ -230,7 +225,7 @@ def main(args: argparse.Namespace) -> None:
         num_total_patches += args.batch_size
         x = torch.rand(args.batch_size, len(bands), args.patch_size, args.patch_size)
         # y = torch.randint(0, 256, (args.batch_size, args.patch_size, args.patch_size))
-        y = torch.randint(0, 256, (args.batch_size,))  # type: ignore[attr-defined]
+        y = torch.randint(0, 256, (args.batch_size,))
         x = x.to(device)
         y = y.to(device)
 

docs/api/datasets.rst

@@ -12,6 +12,16 @@ Geospatial Datasets
 
 :class:`GeoDataset` is designed for datasets that contain geospatial information, like latitude, longitude, coordinate system, and projection. Datasets containing this kind of information can be combined using :class:`IntersectionDataset` and :class:`UnionDataset`.
 
+Aboveground Live Woody Biomass Density
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+.. autoclass:: AbovegroundLiveWoodyBiomassDensity
+
+Aster Global Digital Evaluation Model
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+.. autoclass:: AsterGDEM
+
 Canadian Building Footprints
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
@@ -32,6 +42,11 @@ Chesapeake Bay High-Resolution Land Cover Project
 .. autoclass:: ChesapeakeWV
 .. autoclass:: ChesapeakeCVPR
 
+CMS Global Mangrove Canopy Dataset
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+
+.. autoclass:: CMSGlobalMangroveCanopy
+
 Cropland Data Layer (CDL)
 ^^^^^^^^^^^^^^^^^^^^^^^^^
 
@@ -47,6 +62,16 @@ Esri2020
 
 .. autoclass:: Esri2020
 
+EU-DEM
+^^^^^^
+
+.. autoclass:: EUDEM
+
+GlobBiomass
+^^^^^^^^^^^
+
+.. autoclass:: GlobBiomass
+
 Landsat
 ^^^^^^^
 
@@ -67,6 +92,11 @@ National Agriculture Imagery Program (NAIP)
 
 .. autoclass:: NAIP
 
+Open Buildings
+^^^^^^^^^^^^^^
+
+.. autoclass:: OpenBuildings
+
 Sentinel
 ^^^^^^^^
 
@@ -217,6 +247,11 @@ UC Merced
 
 .. autoclass:: UCMerced
 
+USAVars
+^^^^^^^
+
+.. autoclass:: USAVars
+
 Vaihingen
 ^^^^^^^^^
 

docs/api/samplers.rst

@@ -16,10 +16,12 @@ Samplers are used to index a dataset, retrieving a single query at a time. For :
    from torchgeo.samplers import RandomGeoSampler
 
    dataset = Landsat(...)
-   sampler = RandomGeoSampler(dataset, size=1000, length=100)
+   sampler = RandomGeoSampler(dataset, size=256, length=10000)
    dataloader = DataLoader(dataset, sampler=sampler)
 
 
+This data loader will return 256x256 px images, and has an epoch length of 10,000.
+
 Random Geo Sampler
 ^^^^^^^^^^^^^^^^^^
 
@@ -43,10 +45,12 @@ When working with large tile-based datasets, randomly sampling patches from each
    from torchgeo.samplers import RandomBatchGeoSampler
 
    dataset = Landsat(...)
-   sampler = RandomBatchGeoSampler(dataset, size=1000, batch_size=10, length=100)
+   sampler = RandomBatchGeoSampler(dataset, size=256, batch_size=128, length=10000)
    dataloader = DataLoader(dataset, batch_sampler=sampler)
 
 
+This data loader will return 256x256 px images, and has a batch size of 128 and an epoch length of 10,000.
+
 Random Batch Geo Sampler
 ^^^^^^^^^^^^^^^^^^^^^^^^
 
@@ -66,3 +70,24 @@ Batch Geo Sampler
 ^^^^^^^^^^^^^^^^^
 
 .. autoclass:: BatchGeoSampler
+
+Units
+-----
+
+By default, the ``size`` parameter specifies the size of the image in *pixel* units. If you would instead like to specify the size in *CRS* units, you can change the ``units`` parameter like so:
+
+.. code-block:: python
+
+   from torch.utils.data import DataLoader
+
+   from torchgeo.datasets import Landsat
+   from torchgeo.samplers import RandomGeoSampler, Units
+
+   dataset = Landsat(...)
+   sampler = RandomGeoSampler(dataset, size=256 * 30, length=10000, units=Units.CRS)
+   dataloader = DataLoader(dataset, sampler=sampler)
+
+
+Assuming that each pixel in the CRS is 30 m, this data loader will return 256x256 px images, and has an epoch length of 10,000.
+
+.. autoclass:: Units

docs/conf.py

@@ -38,6 +38,7 @@
 extensions = [
     "sphinx.ext.autodoc",
     "sphinx.ext.intersphinx",
+    "sphinx.ext.mathjax",
     "sphinx.ext.napoleon",
     "sphinx.ext.todo",
     "sphinx.ext.viewcode",

docs/tutorials/trainers.ipynb

@@ -492,13 +492,13 @@
     "        csv_reader = csv.DictReader(f, delimiter=',')\n",
     "        for i, row in enumerate(csv_reader):\n",
     "            try:\n",
-    "                train_rmse.append(float(row[\"train_rmse\"]))\n",
+    "                train_rmse.append(float(row[\"train_RMSE\"]))\n",
     "                train_steps.append(i)\n",
     "            except ValueError: # Ignore rows where train RMSE is empty\n",
     "                pass\n",
     "\n",
     "            try:\n",
-    "                val_rmse.append(float(row[\"val_rmse\"]))\n",
+    "                val_rmse.append(float(row[\"val_RMSE\"]))\n",
     "                val_steps.append(i)\n",
     "            except ValueError: # Ignore rows where val RMSE is empty\n",
     "                pass"

evaluate.py

@@ -12,7 +12,7 @@
 
 import pytorch_lightning as pl
 import torch
-from torchmetrics import Accuracy, JaccardIndex, Metric, MetricCollection
+from torchmetrics import Accuracy, JaccardIndex, MetricCollection
 
 from torchgeo.trainers import ClassificationTask, SemanticSegmentationTask
 from train import TASK_TO_MODULES_MAPPING
@@ -85,19 +85,20 @@ def set_up_parser() -> argparse.ArgumentParser:
 def run_eval_loop(
     model: pl.LightningModule,
     dataloader: Any,
-    device: torch.device,  # type: ignore[name-defined]
-    metrics: Metric,
+    device: torch.device,
+    metrics: MetricCollection,
 ) -> Any:
     """Runs a standard test loop over a dataloader and records metrics.
 
     Args:
         model: the model used for inference
         dataloader: the dataloader to get samples from
         device: the device to put data on
-        metrics: a torchmetrics compatible Metric to score the output from the model
+        metrics: a torchmetrics compatible metric collection to score the output
+            from the model
 
     Returns:
-        the result of ``metric.compute()``
+        the result of ``metrics.compute()``
     """
     for batch in dataloader:
         x = batch["image"].to(device)
@@ -158,7 +159,7 @@ def main(args: argparse.Namespace) -> None:
             "loss": model.hparams["loss"],
         }
     elif issubclass(TASK, SemanticSegmentationTask):
-        val_row: Dict[str, Union[str, float]] = {  # type: ignore[no-redef]
+        val_row = {
             "split": "val",
             "segmentation_model": model.hparams["segmentation_model"],
             "encoder_name": model.hparams["encoder_name"],
@@ -167,7 +168,7 @@ def main(args: argparse.Namespace) -> None:
             "loss": model.hparams["loss"],
         }
 
-        test_row: Dict[str, Union[str, float]] = {  # type: ignore[no-redef]
+        test_row = {
             "split": "test",
             "segmentation_model": model.hparams["segmentation_model"],
             "encoder_name": model.hparams["encoder_name"],
@@ -179,7 +180,7 @@ def main(args: argparse.Namespace) -> None:
         raise ValueError(f"{TASK} is not supported")
 
     # Compute metrics
-    device = torch.device("cuda:%d" % (args.gpu))  # type: ignore[attr-defined]
+    device = torch.device("cuda:%d" % (args.gpu))
     model = model.to(device)
 
     if args.task == "etci2021":  # Custom metric setup for testing ETCI2021

tests/data/agb_live_woody_density/Aboveground_Live_Woody_Biomass_Density.geojson

@@ -0,0 +1 @@
+{"type": "FeatureCollection", "name": "Aboveground_Live_Woody_Biomass_Density", "crs": {"type": "name", "properties": {"name": "urn:ogc:def:crs:OGC:1.3:CRS84"}}, "features": [{"type": "Feature", "properties": {"tile_id": "00N_000E", "download": "tests/data/agb_live_woody_density/00N_000E.tif", "ObjectId": 1, "Shape__Area": 1245542622548.87, "Shape__Length": 4464169.76558139}, "geometry": {"type": "Polygon", "coordinates": [[[0.0, 0.0], [10.0, 0.0], [10.0, -10.0], [0.0, -10.0], [0.0, 0.0]]]}}]}