Merge pull request #921 from azavea/lf/unit-tests

Add rv2 unit tests and other fixes

docs/rv2/index.rst

@@ -105,20 +105,21 @@ and maintain.
             ])
 
         # Use the PyTorch backend for the SemanticSegmentation pipeline.
-        train_chip_sz = 300
+        chip_sz = 300
         backend = PyTorchSemanticSegmentationConfig(
             model=SemanticSegmentationModelConfig(backbone=Backbone.resnet50),
             solver=SolverConfig(lr=1e-4, num_epochs=1, batch_sz=2))
         chip_options = SemanticSegmentationChipOptions(
-            window_method=SemanticSegmentationWindowMethod.random_sample, chips_per_scene=10)
+            window_method=SemanticSegmentationWindowMethod.random_sample,
+            chips_per_scene=10)
 
         return SemanticSegmentationConfig(
             root_uri=root_uri,
             dataset=dataset,
             backend=backend,
-            train_chip_sz=train_chip_sz,
-            chip_options=chip_options,
-            debug=False)
+            train_chip_sz=chip_sz,
+            predict_chip_sz=chip_sz,
+            chip_options=chip_options)
 
 
 Raster Vision uses a ``unittest``-like method for executing pipelines. For instance, if the

docs/rv2/quickstart.rst

@@ -108,21 +108,21 @@ Create a Python file in the ``${RV_QUICKSTART_CODE_DIR}`` named ``tiny_spacenet.
             ])
 
         # Use the PyTorch backend for the SemanticSegmentation pipeline.
-        train_chip_sz = 300
+        chip_sz = 300
         backend = PyTorchSemanticSegmentationConfig(
             model=SemanticSegmentationModelConfig(backbone=Backbone.resnet50),
             solver=SolverConfig(lr=1e-4, num_epochs=1, batch_sz=2))
         chip_options = SemanticSegmentationChipOptions(
-            window_method=SemanticSegmentationWindowMethod.random_sample, chips_per_scene=10)
+            window_method=SemanticSegmentationWindowMethod.random_sample,
+            chips_per_scene=10)
 
         return SemanticSegmentationConfig(
             root_uri=root_uri,
             dataset=dataset,
             backend=backend,
-            train_chip_sz=train_chip_sz,
-            chip_options=chip_options,
-            debug=False)
-
+            train_chip_sz=chip_sz,
+            predict_chip_sz=chip_sz,
+            chip_options=chip_options)
 
 Running the pipeline
 ---------------------

rastervision2/core/analyzer/stats_analyzer_config.py

@@ -12,11 +12,11 @@ class StatsAnalyzerConfig(AnalyzerConfig):
         description=(
             'URI for output. If None and this is part of an RVPipeline, this is '
             'auto-generated.'))
-    sample_prob: float = Field(
+    sample_prob: Optional[float] = Field(
         0.1,
         description=(
-            'The probability of using a random window for computing statistics.'
-        ))
+            'The probability of using a random window for computing statistics. '
+            'If None, will use a sliding window.'))
 
     def update(self, pipeline=None):
         if pipeline is not None and self.output_uri is None:

rastervision2/core/data/label/semantic_segmentation_labels.py

@@ -44,30 +44,20 @@ def set_label_arr(self, window, label_arr):
     def get_label_arr(self, window):
         return self.window_to_label_arr[window.tuple_format()]
 
-    def filter_by_aoi(self, aoi_polygons):
-        """Get the label array for a window.
-
-        Args:
-            window: Box
-            extent: a Box representing the extent of the corresponding Scene
-
-        Returns:
-            np.ndarray of class_ids with zeros filled in outside the AOIs and
-            clipped to the clip_extent
-        """
+    def filter_by_aoi(self, aoi_polygons, null_class_id):
         new_labels = SemanticSegmentationLabels()
 
         for window in self.get_windows():
             window_geom = window.to_shapely()
             label_arr = self.get_label_arr(window)
 
-            if not self.aoi_polygons:
+            if not aoi_polygons:
                 return self
             else:
                 # For each aoi_polygon, intersect with window, and put in window frame of
                 # reference.
                 window_aois = []
-                for aoi in self.aoi_polygons:
+                for aoi in aoi_polygons:
                     window_aoi = aoi.intersection(window_geom)
                     if not window_aoi.is_empty:
 
@@ -87,7 +77,7 @@ def transform_shape(x, y, z=None):
                         out_shape=label_arr.shape,
                         fill=1,
                         dtype=np.uint8)
-                    label_arr[mask.astype(np.bool)] = 0
+                    label_arr[mask.astype(np.bool)] = null_class_id
                     new_labels.set_label_arr(window, label_arr)
 
         return new_labels

rastervision2/core/data/label_source/object_detection_label_source.py

@@ -3,7 +3,7 @@
 
 
 class ObjectDetectionLabelSource(LabelSource):
-    def __init__(self, vector_source, extent):
+    def __init__(self, vector_source, extent=None):
         """Constructor.
 
         Args:

rastervision2/core/data/raster_source/rasterized_source_config.py

@@ -1,6 +1,6 @@
 from rastervision2.core.data.raster_source import (RasterizedSource)
 from rastervision2.core.data.vector_source import (VectorSourceConfig)
-from rastervision2.pipeline.config import register_config, Config, Field
+from rastervision2.pipeline.config import register_config, Config, Field, ConfigError
 
 
 @register_config('rasterizer')
@@ -29,3 +29,9 @@ def build(self, class_config, crs_transformer, extent):
 
         return RasterizedSource(vector_source, self.rasterizer_config, extent,
                                 crs_transformer)
+
+    def validate_config(self):
+        if self.vector_source.has_null_class_bufs():
+            raise ConfigError(
+                'Setting buffer to None for a class in the vector_source is '
+                'not allowed for RasterizedSourceConfig.')

rastervision2/core/evaluation/semantic_segmentation_evaluation.py

@@ -138,7 +138,7 @@ def get_geoms(x):
             gt_count = len(gt)
             class_name = 'vector-{}-{}'.format(
                 mode,
-                self.class_map.get_by_id(class_id).name)
+                self.class_config.names[class_id])
 
             evaluation_item = ClassEvaluationItem(precision, recall, f1,
                                                   count_error, gt_count,

rastervision2/core/evaluation/semantic_segmentation_evaluator.py

@@ -4,7 +4,7 @@
 from shapely.strtree import STRtree
 
 from rastervision2.core.data import ActivateMixin
-from rastervision2.core.data.vector_source import GeoJSONVectorSource
+from rastervision2.core.data.vector_source import GeoJSONVectorSourceConfig
 from rastervision2.core.evaluation import (ClassificationEvaluator,
                                            SemanticSegmentationEvaluation)
 
@@ -44,6 +44,7 @@ def create_evaluation(self):
     def process(self, scenes, tmp_dir):
         evaluation = self.create_evaluation()
         vect_evaluation = self.create_evaluation()
+        null_class_id = self.class_config.get_null_class_id()
 
         for scene in scenes:
             log.info('Computing evaluation for scene {}...'.format(scene.id))
@@ -56,8 +57,9 @@ def process(self, scenes, tmp_dir):
                 if scene.aoi_polygons:
                     # Filter labels based on AOI.
                     ground_truth = ground_truth.filter_by_aoi(
-                        scene.aoi_polygons)
-                    predictions = predictions.filter_by_aoi(scene.aoi_polygons)
+                        scene.aoi_polygons, null_class_id)
+                    predictions = predictions.filter_by_aoi(
+                        scene.aoi_polygons, null_class_id)
                 scene_evaluation = self.create_evaluation()
                 scene_evaluation.compute(ground_truth, predictions)
                 evaluation.merge(scene_evaluation, scene_id=scene.id)
@@ -69,10 +71,11 @@ def process(self, scenes, tmp_dir):
                 )
                 for vo in label_store.vector_output:
                     pred_geojson_uri = vo.uri
-                    mode = vo.mode
+                    mode = vo.get_mode()
                     class_id = vo.class_id
-                    pred_geojson_source = GeoJSONVectorSource(
-                        pred_geojson_uri,
+                    pred_geojson_source = GeoJSONVectorSourceConfig(
+                        uri=pred_geojson_uri, default_class_id=None).build(
+                        self.class_config,
                         scene.raster_source.get_crs_transformer())
                     pred_geojson = pred_geojson_source.get_geojson()
 

rastervision2/core/rv_pipeline/rv_pipeline_config.py

@@ -34,11 +34,10 @@ class RVPipelineConfig(PipelineConfig):
         ('Analyzers to run during analyzer command. A StatsAnalyzer will be added '
          'automatically if any scenes have a RasterTransformer.'))
 
-    debug: bool = Field(False, description='If True, use debug mode.')
     train_chip_sz: int = Field(
-        200, description='Size of training chips in pixels.')
+        300, description='Size of training chips in pixels.')
     predict_chip_sz: int = Field(
-        200, description='Size of predictions chips in pixels.')
+        300, description='Size of predictions chips in pixels.')
     predict_batch_sz: int = Field(
         8, description='Batch size to use during prediction.')
 

rastervision2/examples/chip_classification.py

@@ -78,6 +78,7 @@ def make_scene(scene_info):
             label_source=label_source,
             aoi_uris=[aoi_uri])
 
+    chip_sz = 200
     train_scenes = [make_scene(info) for info in train_scene_info]
     val_scenes = [make_scene(info) for info in val_scene_info]
     dataset = DatasetConfig(
@@ -92,12 +93,13 @@ def make_scene(scene_info):
         model=model,
         solver=solver,
         log_tensorboard=log_tensorboard,
-        run_tensorboard=run_tensorboard)
+        run_tensorboard=run_tensorboard,
+        test_mode=test)
 
     config = ChipClassificationConfig(
         root_uri=root_uri,
         dataset=dataset,
         backend=backend,
-        train_chip_sz=200,
-        debug=debug)
+        train_chip_sz=chip_sz,
+        predict_chip_sz=chip_sz)
     return config

rastervision2/examples/object_detection.py

@@ -61,6 +61,7 @@ def make_scene(id):
             id=id, raster_source=raster_source, label_source=label_source)
 
     class_config = ClassConfig(names=['vehicle'])
+    chip_sz = 300
     dataset = DatasetConfig(
         class_config=class_config,
         train_scenes=[make_scene(id) for id in train_ids],
@@ -78,13 +79,14 @@ def make_scene(id):
             batch_sz=16,
             one_cycle=True),
         log_tensorboard=True,
-        run_tensorboard=False)
+        run_tensorboard=False,
+        test_mode=test)
 
     return ObjectDetectionConfig(
         root_uri=root_uri,
         dataset=dataset,
         backend=backend,
-        train_chip_sz=300,
+        train_chip_sz=chip_sz,
+        predict_chip_sz=chip_sz,
         chip_options=chip_options,
-        predict_options=predict_options,
-        debug=test)
+        predict_options=predict_options)

rastervision2/examples/semantic_segmentation.py

@@ -94,9 +94,9 @@ def make_scene(id):
         class_config=class_config,
         train_scenes=[make_scene(id) for id in train_ids],
         validation_scenes=[make_scene(id) for id in val_ids])
-    train_chip_sz = 300
+    chip_sz = 300
     chip_options = SemanticSegmentationChipOptions(
-        window_method=SemanticSegmentationWindowMethod.sliding, stride=300)
+        window_method=SemanticSegmentationWindowMethod.sliding, stride=chip_sz)
 
     backend = PyTorchSemanticSegmentationConfig(
         model=SemanticSegmentationModelConfig(backbone=Backbone.resnet50),
@@ -107,12 +107,13 @@ def make_scene(id):
             batch_sz=8,
             one_cycle=True),
         log_tensorboard=True,
-        run_tensorboard=False)
+        run_tensorboard=False,
+        test_mode=test)
 
     return SemanticSegmentationConfig(
         root_uri=root_uri,
         dataset=dataset,
         backend=backend,
-        train_chip_sz=train_chip_sz,
-        chip_options=chip_options,
-        debug=test)
+        train_chip_sz=chip_sz,
+        predict_chip_sz=chip_sz,
+        chip_options=chip_options)

rastervision2/examples/tiny_spacenet.py

@@ -58,7 +58,7 @@ def make_scene(scene_id, image_uri, label_uri):
         ])
 
     # Use the PyTorch backend for the SemanticSegmentation pipeline.
-    train_chip_sz = 300
+    chip_sz = 300
     backend = PyTorchSemanticSegmentationConfig(
         model=SemanticSegmentationModelConfig(backbone=Backbone.resnet50),
         solver=SolverConfig(lr=1e-4, num_epochs=1, batch_sz=2))
@@ -70,6 +70,6 @@ def make_scene(scene_id, image_uri, label_uri):
         root_uri=root_uri,
         dataset=dataset,
         backend=backend,
-        train_chip_sz=train_chip_sz,
-        chip_options=chip_options,
-        debug=False)
+        train_chip_sz=chip_sz,
+        predict_chip_sz=chip_sz,
+        chip_options=chip_options)

rastervision2/pytorch_backend/pytorch_chip_classification_config.py

@@ -25,7 +25,7 @@ def get_learner_config(self, pipeline):
             data=data,
             model=self.model,
             solver=self.solver,
-            test_mode=pipeline.debug,
+            test_mode=self.test_mode,
             output_uri=pipeline.train_uri,
             log_tensorboard=self.log_tensorboard,
             run_tensorboard=self.run_tensorboard)

rastervision2/pytorch_backend/pytorch_learner_backend_config.py

@@ -21,6 +21,12 @@ class PyTorchLearnerBackendConfig(BackendConfig):
         description=(
             'Names of albumentations augmentors to use for training batches. '
             'Choices include: ' + str(augmentor_list)))
+    test_mode: bool = Field(
+        False,
+        description=
+        ('This field is passed along to the LearnerConfig which is returned by '
+         'get_learner_config(). For more info, see the docs for'
+         'pytorch_learner.learner_config.LearnerConfig.test_mode.'))
 
     def get_bundle_filenames(self):
         return ['model-bundle.zip']

rastervision2/pytorch_backend/pytorch_object_detection_config.py

@@ -24,7 +24,7 @@ def get_learner_config(self, pipeline):
             data=data,
             model=self.model,
             solver=self.solver,
-            test_mode=pipeline.debug,
+            test_mode=self.test_mode,
             output_uri=pipeline.train_uri,
             log_tensorboard=self.log_tensorboard,
             run_tensorboard=self.run_tensorboard)

rastervision2/pytorch_backend/pytorch_semantic_segmentation_config.py

@@ -24,7 +24,7 @@ def get_learner_config(self, pipeline):
             data=data,
             model=self.model,
             solver=self.solver,
-            test_mode=pipeline.debug,
+            test_mode=self.test_mode,
             output_uri=pipeline.train_uri,
             log_tensorboard=self.log_tensorboard,
             run_tensorboard=self.run_tensorboard)

tests_v2/__init__.py

@@ -0,0 +1,6 @@
+import os
+
+
+def data_file_path(rel_path):
+    data_dir = os.path.join(os.path.dirname(__file__), 'data_files')
+    return os.path.join(data_dir, rel_path)