upgrade to vre 1.9.0

scripts/dronescapes_viewer/dronescapes_representations.py

@@ -1,19 +1,20 @@
 import sys
 from pathlib import Path
 from vre.representations import Representation
-from vre.representations.color import RGB, HSV
-from vre.representations.depth import DepthRepresentation
-from vre.representations.normals import NormalsRepresentation
-from vre.representations.edges import EdgesRepresentation
-from vre.representations.optical_flow import OpticalFlowRepresentation
-from vre.representations.semantic_segmentation import SemanticRepresentation
+from vre_repository.color.rgb import RGB
+from vre_repository.color.hsv import HSV
+from vre_repository.depth import DepthRepresentation
+from vre_repository.normals import NormalsRepresentation
+from vre_repository.edges import EdgesRepresentation
+# from vre_repository.optical_flow import OpticalFlowRepresentation
+from vre_repository.semantic_segmentation import SemanticRepresentation
 
 def get_gt_tasks() -> dict[str, Representation]:
     color_map = [[0, 255, 0], [0, 127, 0], [255, 255, 0], [255, 255, 255],
                 [255, 0, 0], [0, 0, 255], [0, 255, 255], [127, 127, 63]]
     classes_8 = ["land", "forest", "residential", "road", "little-objects", "water", "sky", "hill"]
     tasks = [
-        SemanticRepresentation("semantic_output", classes=classes_8, color_map=color_map, semantic_argmax_only=True),
+        SemanticRepresentation("semantic_output", classes=classes_8, color_map=color_map, disk_data_argmax=True),
         DepthRepresentation("depth_output", min_depth=0, max_depth=300),
         NormalsRepresentation("camera_normals_output"),
     ]

scripts/semantic_mapper/semantic_mapper.py

@@ -12,9 +12,10 @@ from vre.utils import (semantic_mapper, colorize_semantic_segmentation, DiskData
 from vre.logger import vre_logger as logger
 from vre.readers.multitask_dataset import MultiTaskDataset, MultiTaskItem
 from vre.representations import TaskMapper, NpIORepresentation, Representation, build_representations_from_cfg
-from vre.representations.depth import DepthRepresentation
-from vre.representations.normals import NormalsRepresentation
-from vre.representations.semantic_segmentation import SemanticRepresentation
+from vre_repository import get_vre_repository
+from vre_repository.depth import DepthRepresentation
+from vre_repository.normals import NormalsRepresentation
+from vre_repository.semantic_segmentation import SemanticRepresentation
 
 def plot_one(data: MultiTaskItem, title: str, order: list[str] | None,
              name_to_task: dict[str, Representation]) -> np.ndarray:
@@ -97,11 +98,11 @@ mapillary_color_map = [[165, 42, 42], [0, 192, 0], [196, 196, 196], [190, 153, 1
                        [0, 0, 70], [0, 0, 192], [32, 32, 32], [120, 10, 10]]
 
 m2f_coco = SemanticRepresentation("semantic_mask2former_coco_47429163_0", classes=coco_classes,
-                                  color_map=coco_color_map, semantic_argmax_only=True)
+                                  color_map=coco_color_map, disk_data_argmax=True)
 m2f_mapillary = SemanticRepresentation("semantic_mask2former_mapillary_49189528_0", classes=mapillary_classes,
-                                       color_map=mapillary_color_map, semantic_argmax_only=True)
+                                       color_map=mapillary_color_map, disk_data_argmax=True)
 m2f_r50_mapillary = SemanticRepresentation("semantic_mask2former_mapillary_49189528_1", classes=mapillary_classes,
-                                           color_map=mapillary_color_map, semantic_argmax_only=True)
+                                           color_map=mapillary_color_map, disk_data_argmax=True)
 marigold = DepthRepresentation("depth_marigold", min_depth=0, max_depth=1)
 normals_svd_marigold = NormalsRepresentation("normals_svd(depth_marigold)")
 
@@ -145,8 +146,7 @@ class SemanticMask2FormerMapillaryConvertedPaper(TaskMapper, NpIORepresentation)
 
     @overrides
     def merge_fn(self, dep_data: list[MemoryData]) -> MemoryData:
-        m2f_mapillary = self.dependencies[0].to_argmaxed_representation(dep_data[0])
-        m2f_mapillary_converted = semantic_mapper(m2f_mapillary, self.mapping, self.original_classes)
+        m2f_mapillary_converted = semantic_mapper(dep_data[0].argmax(-1), self.mapping, self.original_classes)
         return self.disk_to_memory_fmt(m2f_mapillary_converted)
 
     @overrides
@@ -204,8 +204,7 @@ class SemanticMask2FormerCOCOConverted(TaskMapper, NpIORepresentation):
 
     @overrides
     def merge_fn(self, dep_data: list[MemoryData]) -> MemoryData:
-        m2f_mapillary = self.dependencies[0].to_argmaxed_representation(dep_data[0])
-        m2f_mapillary_converted = semantic_mapper(m2f_mapillary, self.mapping, self.original_classes)
+        m2f_mapillary_converted = semantic_mapper(dep_data[0].argmax(-1), self.mapping, self.original_classes)
         res = self.disk_to_memory_fmt(m2f_mapillary_converted)
         return res
 
@@ -258,10 +257,7 @@ class BinaryMapper(TaskMapper, NpIORepresentation):
 
     @overrides
     def merge_fn(self, dep_data: list[MemoryData]) -> MemoryData:
-        dep_data_argmaxed = []
-        for dep, data in zip(self.dependencies, dep_data):
-            assert isinstance(dep, SemanticRepresentation), type(dep)
-            dep_data_argmaxed.append(dep.to_argmaxed_representation(data))
+        dep_data_argmaxed = [data.argmax(-1) for data in dep_data]
         dep_data_converted = [semantic_mapper(x, mapping, oc)
                               for x, mapping, oc in zip(dep_data_argmaxed, self.mapping, self.original_classes)]
 
@@ -304,13 +300,9 @@ class SemanticMedian(TaskMapper, NpIORepresentation):
     def merge_fn(self, dep_data: list[MemoryData]) -> MemoryData:
         return MemoryData(np.eye(self.n_classes)[sum(dep_data).argmax(-1)].astype(np.uint8))
 
-    @overrides
-    def memory_to_disk_fmt(self, memory_data: MemoryData) -> DiskData:
-        return memory_data.argmax(-1).astype(np.uint8)
-
     @overrides
     def make_images(self, data: ReprOut) -> np.ndarray:
-        data_output = data.output.argmax(-1)# if np.issubdtype(data.output.dtype, np.floating) else data.output
+        data_output = data.output.argmax(-1) if np.issubdtype(data.output.dtype, np.floating) else data.output
         return colorize_semantic_segmentation(data_output, self.classes, self.color_map)
 
 class SafeLandingAreas(BinaryMapper, NpIORepresentation):
@@ -355,12 +347,9 @@ class SafeLandingAreas(BinaryMapper, NpIORepresentation):
         v1, v2, v3 = normals.transpose(2, 0, 1)
         where_safe = (v2 > 0.8) * ((v1 + v3) < 1.2) * (depth <= 0.9)
         if self.include_semantics:
-            mapi1 = self.dependencies[2].to_argmaxed_representation(dep_data[2])
-            coco = self.dependencies[3].to_argmaxed_representation(dep_data[3])
-            mapi2 = self.dependencies[4].to_argmaxed_representation(dep_data[4])
-            conv1 = np.isin(mapi1, self.safe_mapillary_ix).astype(int)
-            conv2 = np.isin(coco, self.safe_coco_ix).astype(int)
-            conv3 = np.isin(mapi2, self.safe_mapillary_ix).astype(int)
+            conv1 = np.isin(dep_data[2].argmax(-1), self.safe_mapillary_ix).astype(int)
+            conv2 = np.isin(dep_data[3].argmax(-1), self.safe_coco_ix).astype(int)
+            conv3 = np.isin(dep_data[4].argmax(-1), self.safe_mapillary_ix).astype(int)
             sema_safe = (conv1 + conv2 + conv3) >= 2
             where_safe = sema_safe * where_safe
         return self.disk_to_memory_fmt(where_safe)
@@ -500,22 +489,20 @@ def get_new_semantic_mapped_tasks(tasks_subset: list[str] | None = None) -> dict
     return {t.name: t for t in available_tasks if t.name in tasks_subset}
 
 if __name__ == "__main__":
-    cfg_path = Path.cwd() / "../../vre_dronescapes/cfg.yaml"
-    data_path = Path.cwd() / "../../vre_dronescapes/norway_210821_DJI_0015_full/"
+    cfg_path = Path.cwd() / "cfg.yaml"
+    data_path = Path.cwd() / "data"
     vre_dir = data_path
 
     task_names = ["rgb", "depth_marigold", "normals_svd(depth_marigold)",
-                  "semantic_mask2former_coco_47429163_0", "semantic_mask2former_mapillary_49189528_0",
-                  "semantic_mask2former_mapillary_49189528_1"]
+                  "semantic_mask2former_coco_47429163_0", "semantic_mask2former_mapillary_49189528_0"]
     order = ["rgb", "semantic_mask2former_mapillary_49189528_0", "semantic_mask2former_coco_47429163_0",
              "depth_marigold", "normals_svd(depth_marigold)"]
 
-    task_types = {r.name: r for r in build_representations_from_cfg(cfg_path) if r.name in task_names}
-    statistics = np.load(Path.cwd() / "../../data/train_set/.task_statistics.npz", allow_pickle=True)["arr_0"].item()
-    # breakpoint()
+    repr_types = get_vre_repository()
+    task_types = {r.name: r for r in build_representations_from_cfg(cfg_path, repr_types) if r.name in task_names}
     reader = MultiTaskDataset(vre_dir, task_names=task_names, task_types=task_types,
-                              handle_missing_data="fill_nan", normalization="min_max",
-                              cache_task_stats=True, batch_size_stats=100, statistics=statistics)
+                              handle_missing_data="fill_nan", normalization=None,
+                              cache_task_stats=True, batch_size_stats=100)
     orig_task_names = list(reader.task_types.keys())
 
     new_tasks = get_new_semantic_mapped_tasks()

vre_dronescapes/cfg.yaml

@@ -12,7 +12,7 @@ default_compute_parameters:
 
 representations:
   rgb:
-    type: default/rgb
+    type: color/rgb
     dependencies: []
     parameters: {}
 
@@ -30,7 +30,7 @@ representations:
     dependencies: []
     parameters:
       model_id: "47429163_0"
-      semantic_argmax_only: True
+      disk_data_argmax: True
     compute_parameters:
       batch_size: 1
 
@@ -39,7 +39,7 @@ representations:
     dependencies: []
     parameters:
       model_id: "49189528_0"
-      semantic_argmax_only: True
+      disk_data_argmax: True
     compute_parameters:
       batch_size: 1
 
@@ -48,7 +48,7 @@ representations:
     dependencies: []
     parameters:
       model_id: "49189528_1"
-      semantic_argmax_only: True
+      disk_data_argmax: True
     compute_parameters:
       batch_size: 1