chore: remove a last import of SHAPE_TYPE

myria3d/pctl/dataset/hdf5.py

@@ -12,7 +12,6 @@
 
 from myria3d.pctl.dataset.utils import (
     LAS_PATHS_BY_SPLIT_DICT_TYPE,
-    SHAPE_TYPE,
     SPLIT_TYPE,
     pre_filter_below_n_points,
     split_cloud_into_samples,
@@ -71,7 +70,9 @@ def __init__(
         self._samples_hdf5_paths = None
 
         if not las_paths_by_split_dict:
-            log.warning("No las_paths_by_split_dict given, pre-computed HDF5 dataset is therefore used.")
+            log.warning(
+                "No las_paths_by_split_dict given, pre-computed HDF5 dataset is therefore used."
+            )
             return
 
         # Add data for all LAS Files into a single hdf5 file.
@@ -151,7 +152,9 @@ def testdata(self):
 
     def _get_split_subset(self, split: SPLIT_TYPE):
         """Get a sub-dataset of a specific (train/val/test) split."""
-        indices = [idx for idx, p in enumerate(self.samples_hdf5_paths) if p.startswith(split)]
+        indices = [
+            idx for idx, p in enumerate(self.samples_hdf5_paths) if p.startswith(split)
+        ]
         return torch.utils.data.Subset(self, indices)
 
     @property
@@ -164,7 +167,10 @@ def samples_hdf5_paths(self):
         # Load as variable if already indexed in hdf5 file. Need to decode b-string.
         with h5py.File(self.hdf5_file_path, "r") as hdf5_file:
             if "samples_hdf5_paths" in hdf5_file:
-                self._samples_hdf5_paths = [sample_path.decode("utf-8") for sample_path in hdf5_file["samples_hdf5_paths"]]
+                self._samples_hdf5_paths = [
+                    sample_path.decode("utf-8")
+                    for sample_path in hdf5_file["samples_hdf5_paths"]
+                ]
                 return self._samples_hdf5_paths
 
         # Otherwise, index samples, and add the index as an attribute to the HDF5 file.
@@ -175,7 +181,9 @@ def samples_hdf5_paths(self):
                     continue
                 for basename in hdf5_file[split].keys():
                     for sample_number in hdf5_file[split][basename].keys():
-                        self._samples_hdf5_paths.append(osp.join(split, basename, sample_number))
+                        self._samples_hdf5_paths.append(
+                            osp.join(split, basename, sample_number)
+                        )
 
         with h5py.File(self.hdf5_file_path, "a") as hdf5_file:
             # special type to avoid silent string truncation in hdf5 datasets.
@@ -198,7 +206,6 @@ def create_hdf5(
     subtile_overlap_train: Number = 0,
     points_pre_transform: Callable = lidar_hd_pre_transform,
 ):
-
     """Create a HDF5 dataset file from las.
 
     Args:
@@ -221,20 +228,24 @@ def create_hdf5(
             if split not in f:
                 f.create_group(split)
         for las_path in tqdm(las_paths, desc=f"Preparing {split} set..."):
-
             basename = os.path.basename(las_path)
 
             # Delete dataset for incomplete LAS entry, to start from scratch.
             # Useful in case data preparation was interrupted.
             with h5py.File(hdf5_file_path, "a") as hdf5_file:
-                if basename in hdf5_file[split] and "is_complete" not in hdf5_file[split][basename].attrs:
+                if (
+                    basename in hdf5_file[split]
+                    and "is_complete" not in hdf5_file[split][basename].attrs
+                ):
                     del hdf5_file[split][basename]
                     # Parse and add subtiles to split group.
             with h5py.File(hdf5_file_path, "a") as hdf5_file:
                 if basename in hdf5_file[split]:
                     continue
 
-                subtile_overlap = subtile_overlap_train if split == "train" else 0  # No overlap at eval time.
+                subtile_overlap = (
+                    subtile_overlap_train if split == "train" else 0
+                )  # No overlap at eval time.
                 for sample_number, (sample_idx, sample_points) in enumerate(
                     split_cloud_into_samples(
                         las_path,
@@ -249,15 +260,19 @@ def create_hdf5(
                     if pre_filter is not None and pre_filter(data):
                         # e.g. pre_filter spots situations where num_nodes is too small.
                         continue
-                    hdf5_path = os.path.join(split, basename, str(sample_number).zfill(5))
+                    hdf5_path = os.path.join(
+                        split, basename, str(sample_number).zfill(5)
+                    )
                     hd5f_path_x = os.path.join(hdf5_path, "x")
                     hdf5_file.create_dataset(
                         hd5f_path_x,
                         data.x.shape,
                         dtype="f",
                         data=data.x,
                     )
-                    hdf5_file[hd5f_path_x].attrs["x_features_names"] = copy.deepcopy(data.x_features_names)
+                    hdf5_file[hd5f_path_x].attrs["x_features_names"] = copy.deepcopy(
+                        data.x_features_names
+                    )
                     hdf5_file.create_dataset(
                         os.path.join(hdf5_path, "pos"),
                         data.pos.shape,