Semantic segmentation test sites that are all one category

[NoaMillsUSDA-ARS]

I ran into this issue accidentally. When constructing test sites, I accidentally configured one of the test sites to have equivalent polygons in as both the aoi file and the label mask file. This caused my code to error out. I fixed this issue so I have the correct label file, but I figured it should be possible to have test sites that are entirely covered by one category. I've tried a few times to create test sites that are completely covered by one category, and each time I've gotten runtime errors. I'm wondering how to do this correctly, or if this functionality is not supported?

When there are two separate vector files that store equivalent polygons used as the label mask and the aoi respectively, I get this error. I have ensured that both files are in the appropriate crs.

2023-08-04 15:20:53:rastervision.pytorch_learner.learner: INFO - Building datasets ...
Traceback (most recent call last):
  File "/opt/conda/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/opt/conda/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/opt/src/rastervision_pipeline/rastervision/pipeline/cli.py", line 251, in <module>
    _main()
  File "/opt/src/rastervision_pipeline/rastervision/pipeline/cli.py", line 247, in _main
    main()
  File "/opt/conda/lib/python3.10/site-packages/click/core.py", line 1130, in __call__
    return self.main(*args, **kwargs)
  File "/opt/conda/lib/python3.10/site-packages/click/core.py", line 1055, in main
    rv = self.invoke(ctx)
  File "/opt/conda/lib/python3.10/site-packages/click/core.py", line 1657, in invoke
    return _process_result(sub_ctx.command.invoke(sub_ctx))
  File "/opt/conda/lib/python3.10/site-packages/click/core.py", line 1404, in invoke
    return ctx.invoke(self.callback, **ctx.params)
  File "/opt/conda/lib/python3.10/site-packages/click/core.py", line 760, in invoke
    return __callback(*args, **kwargs)
  File "/opt/src/rastervision_pipeline/rastervision/pipeline/cli.py", line 236, in run_command
    _run_command(
  File "/opt/src/rastervision_pipeline/rastervision/pipeline/cli.py", line 218, in _run_command
    command_fn()
  File "/opt/src/rastervision_core/rastervision/core/rv_pipeline/rv_pipeline.py", line 154, in train
    backend.train(source_bundle_uri=self.config.source_bundle_uri)
  File "/opt/src/rastervision_pytorch_backend/rastervision/pytorch_backend/pytorch_learner_backend.py", line 119, in train
    learner = self.learner_cfg.build(self.tmp_dir, training=True)
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/semantic_segmentation_learner_config.py", line 215, in build
    return SemanticSegmentationLearner(
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/learner.py", line 197, in __init__
    self.setup_training(loss_def_path=loss_def_path)
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/learner.py", line 790, in setup_training
    self.setup_data()
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/learner.py", line 868, in setup_data
    train_ds, valid_ds, test_ds = self.build_datasets()
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/learner.py", line 880, in build_datasets
    train_ds, val_ds, test_ds = self.cfg.data.build(
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/learner_config.py", line 1318, in build
    train_ds, val_ds, test_ds = self.make_datasets(
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/learner_config.py", line 1280, in make_datasets
    train_scenes, val_scenes, test_scenes = self.build_scenes(tmp_dir)
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/learner_config.py", line 1247, in build_scenes
    val_scenes = [
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/learner_config.py", line 1248, in <listcomp>
    s.build(class_cfg, tmp_dir, use_transformers=True)
  File "/opt/src/rastervision_core/rastervision/core/data/scene_config.py", line 59, in build
    label_source = self.label_source.build(
  File "/opt/src/rastervision_core/rastervision/core/data/label_source/semantic_segmentation_label_source_config.py", line 32, in build
    rs = self.raster_source.build(class_config, crs_transformer, bbox)
  File "/opt/src/rastervision_core/rastervision/core/data/raster_source/rasterized_source_config.py", line 60, in build
    return RasterizedSource(
  File "/opt/src/rastervision_core/rastervision/core/data/raster_source/rasterized_source.py", line 88, in __init__
    self.df = self.vector_source.get_dataframe()
  File "/opt/src/rastervision_core/rastervision/core/data/vector_source/vector_source.py", line 123, in get_dataframe
    self._gdf = geojson_to_geodataframe(geojson)
  File "/opt/src/rastervision_core/rastervision/core/data/utils/geojson.py", line 362, in geojson_to_geodataframe
    df = gpd.GeoDataFrame.from_features(geojson)
  File "/opt/conda/lib/python3.10/site-packages/geopandas/geodataframe.py", line 648, in from_features
    return cls(rows, columns=columns, crs=crs)
  File "/opt/conda/lib/python3.10/site-packages/geopandas/geodataframe.py", line 176, in __init__
    self["geometry"] = _ensure_geometry(self["geometry"].values, crs)
  File "/opt/conda/lib/python3.10/site-packages/geopandas/geodataframe.py", line 1533, in __setitem__
    value = _ensure_geometry(value, crs=crs)
  File "/opt/conda/lib/python3.10/site-packages/geopandas/geodataframe.py", line 52, in _ensure_geometry
    data.crs = crs
  File "/opt/conda/lib/python3.10/site-packages/geopandas/array.py", line 359, in crs
    self._crs = None if not value else CRS.from_user_input(value)
  File "/opt/conda/lib/python3.10/site-packages/pandas/core/generic.py", line 1466, in __nonzero__
    raise ValueError(
ValueError: The truth value of a Series is ambiguous. Use a.empty, a.bool(), a.item(), a.any() or a.all().
make: *** [/opt/data/output_user_test/Makefile:6: 0] Error 1

When I use the same file for the aoi and the label mask, I get the following error:

2023-08-07 16:35:05:rastervision.pytorch_learner.learner: INFO - Building datasets ...
Traceback (most recent call last):
  File "/opt/conda/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/opt/conda/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/opt/src/rastervision_pipeline/rastervision/pipeline/cli.py", line 251, in <module>
    _main()
  File "/opt/src/rastervision_pipeline/rastervision/pipeline/cli.py", line 247, in _main
    main()
  File "/opt/conda/lib/python3.10/site-packages/click/core.py", line 1130, in __call__
    return self.main(*args, **kwargs)
  File "/opt/conda/lib/python3.10/site-packages/click/core.py", line 1055, in main
    rv = self.invoke(ctx)
  File "/opt/conda/lib/python3.10/site-packages/click/core.py", line 1657, in invoke
    return _process_result(sub_ctx.command.invoke(sub_ctx))
  File "/opt/conda/lib/python3.10/site-packages/click/core.py", line 1404, in invoke
    return ctx.invoke(self.callback, **ctx.params)
  File "/opt/conda/lib/python3.10/site-packages/click/core.py", line 760, in invoke
    return __callback(*args, **kwargs)
  File "/opt/src/rastervision_pipeline/rastervision/pipeline/cli.py", line 236, in run_command
    _run_command(
  File "/opt/src/rastervision_pipeline/rastervision/pipeline/cli.py", line 218, in _run_command
    command_fn()
  File "/opt/src/rastervision_core/rastervision/core/rv_pipeline/rv_pipeline.py", line 154, in train
    backend.train(source_bundle_uri=self.config.source_bundle_uri)
  File "/opt/src/rastervision_pytorch_backend/rastervision/pytorch_backend/pytorch_learner_backend.py", line 119, in train
    learner = self.learner_cfg.build(self.tmp_dir, training=True)
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/semantic_segmentation_learner_config.py", line 215, in build
    return SemanticSegmentationLearner(
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/learner.py", line 197, in __init__
    self.setup_training(loss_def_path=loss_def_path)
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/learner.py", line 790, in setup_training
    self.setup_data()
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/learner.py", line 868, in setup_data
    train_ds, valid_ds, test_ds = self.build_datasets()
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/learner.py", line 880, in build_datasets
    train_ds, val_ds, test_ds = self.cfg.data.build(
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/learner_config.py", line 1318, in build
    train_ds, val_ds, test_ds = self.make_datasets(
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/learner_config.py", line 1285, in make_datasets
    val_ds_list = [
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/learner_config.py", line 1286, in <listcomp>
    self.scene_to_dataset(s, val_tf, **kwargs) for s in val_scenes
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/semantic_segmentation_learner_config.py", line 132, in scene_to_dataset
    ds = SemanticSegmentationRandomWindowGeoDataset(
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/dataset/semantic_segmentation_dataset.py", line 174, in __init__
    super().__init__(
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/dataset/dataset.py", line 379, in __init__
    self.aoi_sampler = AoiSampler(self.aoi_polygons)
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/dataset/utils/aoi_sampler.py", line 27, in __init__
    self.triangulate(self.polygons)
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/dataset/utils/aoi_sampler.py", line 30, in triangulate
    triangulations = [self.triangulate_polygon(p) for p in polygons]
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/dataset/utils/aoi_sampler.py", line 30, in <listcomp>
    triangulations = [self.triangulate_polygon(p) for p in polygons]
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/dataset/utils/aoi_sampler.py", line 80, in triangulate_polygon
    hole_vertices, hole_edges = self.polygon_to_graph(hole)
  File "/opt/src/rastervision_pytorch_learner/rastervision/pytorch_learner/dataset/utils/aoi_sampler.py", line 122, in polygon_to_graph
    vertices = np.array(polygon.exterior.coords)
AttributeError: 'LinearRing' object has no attribute 'exterior'
make: *** [/opt/data/debug2/Makefile:6: 0] Error 1

[NoaMillsUSDA-ARS]

Additional info I forgot: I'm running rastervision 0.20 with docker tag dc23f1f.

[AdeelH]

I think the errors might be unrelated to the AOI and labels file being the same. The first error seems to suggest that there's something wrong with the crs field in the GeoJSON, while the second error seems to suggest that there is a non-polygon in the AOI GeoJSON.

Either way, I think RV should do a better job of handling these situations. Can you share a sample GeoJSON file that runs into these errors?

[NoaMillsUSDA-ARS]

debug_test_data.zip
Here's a file I created to be an exact copy of the aoi file. When I use this file as the label mask and an identical but renamed file as the aoi, I get the LinearRing error as above. I know there isn't an issue with the polygon because when I use the same polygon as just the aoi, and a different file for the label mask, I don't get any errors.

[AdeelH]

Thanks. I will look into this. In the meantime, one way to work around the LinearRing error is to pass efficient_aoi_sampling=False to GeoDataWindowConfig.

[AdeelH]

Okay, I think I have figured it out:

Your GeoJSON file has two "issues":

1. Each feature has a "crs" property which gets converted into a GeoDataframe column when passed to geopandas, which confuses it and results in the ValueError: The truth value of a Series is ambiguous error.
   • Fix: remove this property from your features. I cal also make it so that RV deletes any such "crs" keys.
2. The second issue is not present in the file you shared, but presumably in the file that generated the LinearRing error. Based on the stack trace, that GeoJSON has geometries that have "holes". This is not really a problem in itself, but you are running into the error because of a bug in RV. That part of RV is trying to use the pre-v2.0 shapely API (it should have been updated in Bump pygeos, shapely, and geopandas versions #1811 but was missed).
   • Fix: I'll update the hole-related code.

[AdeelH]

Should be fixed by #1856.