Feature/segment properties (#24)

Adds ability to tag segmentations with a string label. This is across
segmentation masks and meshes


![image_720](https://github.com/user-attachments/assets/7b9d4216-395d-4df1-8354-110b2be60fae)

cryoet_data_portal_neuroglancer/models/json_generator.py

@@ -48,6 +48,33 @@ def __str__(self):
         return self.name.lower()
 
 
+@dataclass
+class SegmentPropertyJSONGenerator:
+    """Generates a JSON file for segmentation properties.
+
+    Supports a subset of the properties that can be set in Neuroglancer.
+    See https://github.com/google/neuroglancer/blob/3efc90465e702453916d2b03d472c16378848132/src/datasource/precomputed/segment_properties.md
+    """
+
+    ids: list[int]
+    labels: list[str]
+
+    def generate_json(self) -> dict:
+        return {
+            "inline": {
+                "ids": [str(val) for val in self.ids],
+                "properties": [
+                    {
+                        "values": self.labels,
+                        "type": "label",
+                        "id": "label",
+                    },
+                ],
+            },
+            "@type": "neuroglancer_segment_properties",
+        }
+
+
 @dataclass
 class RenderingJSONGenerator:
     """Generates a JSON file for Neuroglancer to read."""

cryoet_data_portal_neuroglancer/precompute/mesh.py

@@ -26,6 +26,7 @@
 from tqdm import tqdm
 
 from cryoet_data_portal_neuroglancer.igneous_patch import patched_create_octree_level_from_mesh, patched_process_mesh
+from cryoet_data_portal_neuroglancer.precompute.segmentation_properties import write_segment_properties
 from cryoet_data_portal_neuroglancer.utils import determine_mesh_shape_from_lods
 
 LOGGER = logging.getLogger(__name__)
@@ -253,6 +254,7 @@ def _generate_standalone_mesh_info(
     mesh_dir: str = "mesh",
     resolution: tuple[float, float, float] | float = (1.0, 1.0, 1.0),
     mesh_chunk_size: tuple[float, float, float] | float = (448, 448, 448),
+    has_segment_properties: bool = False,
 ):
     outfolder = Path(outfolder)
     outfolder.mkdir(exist_ok=True, parents=True)
@@ -262,32 +264,31 @@ def _generate_standalone_mesh_info(
 
     # offset = bbox.transform[:, 3][:3].tolist()
     info = outfolder / "info"
-    info.write_text(
-        json.dumps(
+    output = {
+        "@type": "neuroglancer_multiscale_volume",
+        "data_type": "uint32",
+        "mesh": "mesh",
+        "num_channels": 1,
+        "scales": [
             {
-                "@type": "neuroglancer_multiscale_volume",
-                "data_type": "uint32",
-                "mesh": "mesh",
-                "num_channels": 1,
-                "scales": [
-                    {
-                        "chunk_sizes": [[256, 256, 256]],  # information required by neuroglancer but not used
-                        "compressed_segmentation_block_size": [
-                            64,
-                            64,
-                            64,
-                        ],  # information required by neuroglancer but not used
-                        "encoding": "compressed_segmentation",
-                        "key": "data",
-                        "resolution": resolution_conv,
-                        "size": size,
-                    },
-                ],
-                "type": "segmentation",
+                "chunk_sizes": [[256, 256, 256]],  # information required by neuroglancer but not used
+                "compressed_segmentation_block_size": [
+                    64,
+                    64,
+                    64,
+                ],  # information required by neuroglancer but not used
+                "encoding": "compressed_segmentation",
+                "key": "data",
+                "resolution": resolution_conv,
+                "size": size,
             },
-            indent=2,
-        ),
-    )
+        ],
+        "type": "segmentation",
+    }
+
+    if has_segment_properties:
+        output["segment_properties"] = "segment_properties"
+    info.write_text(json.dumps(output, indent=2))
 
     mesh_info = outfolder / mesh_dir
     mesh_info.mkdir(exist_ok=True, parents=True)
@@ -452,6 +453,7 @@ def generate_mesh_from_lods(
     min_mesh_chunk_dim: int = 16,
     label: int = 1,
     bounding_box_size: tuple[float, float, float] | None = None,
+    string_label: str | None = None,
 ):
     """
     Generate a sharded mesh from a list of LODs
@@ -474,6 +476,8 @@ def generate_mesh_from_lods(
         This calculation is often not accurate, so it is recommended to
         provide the bounding box size. Or turn off the bounding box in the
         neuroglancer state.
+    string_label : str | None, optional
+        The string label to use, by default None - ie. not used
     """
     concatenated_lods: list[trimesh.Trimesh] = cast(list[trimesh.Trimesh], [lod.dump(concatenate=True) for lod in lods])
     num_lod = len(concatenated_lods)
@@ -499,8 +503,12 @@ def _compute_size(bbx):
         size=(size_x, size_y, size_z),
         resolution=1.0,
         mesh_chunk_size=actual_chunk_shape,
+        has_segment_properties=string_label is not None,
     )
 
+    if string_label is not None:
+        write_segment_properties(outfolder, [label], [string_label])
+
     tq = LocalTaskQueue(progress=False)
     tasks = _create_sharded_multires_mesh_tasks_from_glb(
         f"precomputed://file://{outfolder}",
@@ -521,6 +529,7 @@ def generate_multiresolution_mesh(
     min_mesh_chunk_dim: int = 16,
     bounding_box_size: tuple[float, float, float] | None = None,
     label: int = 1,
+    string_label: str | None = None,
 ):
     """
     Generate a standalone sharded multiresolution mesh from a glb file or scene
@@ -549,6 +558,7 @@ def generate_multiresolution_mesh(
         min_mesh_chunk_dim=min_mesh_chunk_dim,
         bounding_box_size=bounding_box_size,
         label_dict={label: glb},
+        string_labels={label: string_label} if string_label is not None else None,
     )
 
 
@@ -558,6 +568,7 @@ def generate_multilabel_multiresolution_mesh(
     max_lod: int = 2,
     min_mesh_chunk_dim: int = 16,
     bounding_box_size: tuple[float, float, float] | None = None,
+    string_labels: dict[int, str] | None = None,
 ):
     """
     Generate standalone sharded multiresolution meshes from a mapping of labels to glb files or scenes.
@@ -626,8 +637,14 @@ def _compute_size():
         size=(size_x, size_y, size_z),
         resolution=1.0,
         mesh_chunk_size=actual_chunk_size,
+        has_segment_properties=string_labels is not None,
     )
 
+    if string_labels is not None:
+        ids = list(string_labels.keys())
+        string_labels_list = [string_labels[i] for i in ids]
+        write_segment_properties(outfolder, ids, string_labels_list)
+
     tq = LocalTaskQueue()
     tasks = _create_sharded_multires_mesh_tasks_from_glb(
         f"precomputed://file://{outfolder}",
@@ -648,6 +665,7 @@ def generate_multiresolution_mesh_from_segmentation(
     mesh_shape: tuple[int, int, int] | np.ndarray,
     min_mesh_chunk_dim: int = 16,
     max_simplification_error: int = 10,
+    labels_dict: dict[int, str] | None = None,
 ) -> None:
     """Generates the meshes for a segmentation stored as a precomputed Neuroglancer format.
 
@@ -668,6 +686,9 @@ def generate_multiresolution_mesh_from_segmentation(
         The maximal simplification error allowed for the mesh generation from the
         segmentation. This is used to simplify the mesh and reduce the number of
         vertices and faces in the mesh. The error is in the same unit as the data.
+    labels_dict: dict[int, str] | None
+        A dictionary of labels to string labels. This is used to generate the segment properties
+        for the segmentation. If None, no segment properties are generated.
     """
     tq = LocalTaskQueue()
 
@@ -709,6 +730,14 @@ def generate_multiresolution_mesh_from_segmentation(
     tq.insert(tasks)
     tq.execute()
 
+    if labels_dict is not None:
+        LOGGER.debug("Writing segment properties")
+        write_segment_properties(
+            Path(mesh_directory).parent,
+            list(labels_dict.keys()),
+            list(labels_dict.values()),
+        )
+
 
 def generate_single_resolution_mesh(
     dask_data: da.Array,

cryoet_data_portal_neuroglancer/precompute/segmentation_mask.py

@@ -16,6 +16,7 @@
     clean_mesh_folder,
     generate_multiresolution_mesh_from_segmentation,
 )
+from cryoet_data_portal_neuroglancer.precompute.segmentation_properties import write_segment_properties
 from cryoet_data_portal_neuroglancer.utils import (
     determine_size_of_non_zero_bounding_box,
     get_grid_size_from_block_shape,
@@ -245,6 +246,7 @@ def _create_metadata(
     data_directory: str,
     resolution: tuple[float, float, float] = (1.0, 1.0, 1.0),
     mesh_directory: str | None = None,
+    has_segment_properties: bool = False,
 ) -> dict[str, Any]:
     """Create the metadata for the segmentation"""
     metadata = {
@@ -263,6 +265,8 @@ def _create_metadata(
         ],
         "type": "segmentation",
     }
+    if has_segment_properties:
+        metadata["segment_properties"] = "segment_properties"
     if mesh_directory:
         metadata["mesh"] = mesh_directory
     return metadata
@@ -307,6 +311,7 @@ def encode_segmentation(
     min_mesh_chunk_dim: int = 16,
     fast_bounding_box: bool = False,
     max_simplification_error_in_voxels: int = 2,
+    labels_dict: dict[int, str] | None = None,
 ) -> None:
     """Convert the given OME-Zarr file to neuroglancer segmentation format with the given block size
 
@@ -364,6 +369,10 @@ def encode_segmentation(
         simplified. This parameter sets the maximum error in voxels for
         the simplification, by default 2.
         For large meshes, this can be increased to reduce memory consumption.
+    labels_dict : dict[int, str] | None, optional
+        A dictionary mapping the integer labels in the segmentation to
+        human-readable names, by default None
+        This is useful for generating a legend in the viewer
     """
     LOGGER.info("Converting %s to neuroglancer compressed segmentation format", filename)
     output_path = Path(output_path)
@@ -400,8 +409,15 @@ def encode_segmentation(
         data_directory,
         resolution,
         mesh_directory=mesh_directory if include_mesh else None,
+        has_segment_properties=labels_dict is not None,
     )
     write_metadata(metadata, output_path)
+    if labels_dict is not None:
+        write_segment_properties(
+            output_path,
+            list(labels_dict.keys()),
+            list(labels_dict.values()),
+        )
 
     if include_mesh:
         LOGGER.info("Converting %s to neuroglancer mesh format", filename)

cryoet_data_portal_neuroglancer/precompute/segmentation_properties.py

@@ -0,0 +1,12 @@
+import json
+from pathlib import Path
+
+from cryoet_data_portal_neuroglancer.models.json_generator import SegmentPropertyJSONGenerator
+
+
+def write_segment_properties(base_folder: str | Path, ids: list[int], labels: list[str]):
+    segment_generator = SegmentPropertyJSONGenerator(ids=ids, labels=labels)
+    segment_properties = segment_generator.generate_json()
+    segment_properties_path = Path(base_folder) / "segment_properties" / "info"
+    segment_properties_path.parent.mkdir(exist_ok=True, parents=True)
+    segment_properties_path.write_text(json.dumps(segment_properties, indent=2))

manual_tests/many_mesh_to_precompute.py

@@ -68,7 +68,25 @@ def main(output_folder="test_glb", port=1030, num_lods=2):
     here = Path(__file__).parent
     output_folder = here / output_folder
     mesh_dict = generate_basic_trimeshes()
-    generate_multilabel_multiresolution_mesh(mesh_dict, output_folder, num_lods - 1, 16, [500, 500, 500])
+    string_labels = [
+        "cube",
+        "torus",
+        "sphere",
+        "cylinder",
+        "cone",
+        "capsule",
+        "annulus",
+        "icosahedron",
+    ]
+    string_labels = dict(zip(mesh_dict.keys(), string_labels, strict=False))
+    generate_multilabel_multiresolution_mesh(
+        mesh_dict,
+        output_folder,
+        num_lods - 1,
+        16,
+        [500, 500, 500],
+        string_labels,
+    )
 
     print(
         f"Go to https://neuroglancer-demo.appspot.com/#!%7B%22dimensions%22:%7B%22x%22:%5B1e-9%2C%22m%22%5D%2C%22y%22:%5B1e-9%2C%22m%22%5D%2C%22z%22:%5B1e-9%2C%22m%22%5D%7D%2C%22position%22:%5B10.890692710876465%2C-2.4535868167877197%2C50.0444221496582%5D%2C%22crossSectionScale%22:1%2C%22projectionOrientation%22:%5B-0.9983203411102295%2C-0.04334384202957153%2C0.009803229942917824%2C-0.037171632051467896%5D%2C%22projectionScale%22:642.7371815420057%2C%22layers%22:%5B%7B%22type%22:%22segmentation%22%2C%22source%22:%7B%22url%22:%22precomputed://http://localhost:{port}%22%2C%22subsources%22:%7B%22default%22:true%2C%22mesh%22:true%7D%2C%22enableDefaultSubsources%22:false%7D%2C%22toolBindings%22:%7B%22B%22:%22selectSegments%22%7D%2C%22tab%22:%22rendering%22%2C%22hoverHighlight%22:false%2C%22ignoreNullVisibleSet%22:false%2C%22meshRenderScale%22:475.1228767657264%2C%22crossSectionRenderScale%22:2%2C%22segments%22:%5B%221%22%2C%222%22%2C%223%22%2C%224%22%2C%225%22%2C%226%22%2C%227%22%2C%228%22%5D%2C%22segmentQuery%22:%228%22%2C%22name%22:%22localhost:{port}%22%7D%5D%2C%22showAxisLines%22:false%2C%22showScaleBar%22:false%2C%22showSlices%22:false%2C%22selectedLayer%22:%7B%22size%22:605%2C%22visible%22:true%2C%22layer%22:%22localhost:{port}%22%7D%2C%22layout%22:%223d%22%7D to view",

manual_tests/segmentation_from_api.py

@@ -32,9 +32,11 @@ def make_precomputed_segmentation():
         zarr_path,
         output_path,
         (resolution, resolution, resolution),
-        include_mesh=False,
+        include_mesh=True,
         convert_non_zero_to=1,
-        delete_existing=False,
+        delete_existing=True,
+        fast_bounding_box=True,
+        labels_dict={1: "membrane"},
     )
 
 
@@ -49,6 +51,7 @@ def make_multi_res_mesh():
         max_lod=2,
         mesh_shape=mesh_shape,
         max_simplification_error=max_simplification_error,
+        labels_dict={1: "membrane"},
     )
     clean_mesh_folder(output_path, "mesh")
 
@@ -77,7 +80,7 @@ def serve_files():
 def main():
     grab_annotation()
     make_precomputed_segmentation()
-    make_multi_res_mesh()
+    # make_multi_res_mesh()
     # make_single_res_mesh()
     serve_files()