Merge branch 'main' of https://github.com/computational-cell-analytics/synapse-net into 2d_imod_export

Author: SarahMuth

synapse_net/inference/inference.py

@@ -155,7 +155,7 @@ def compute_scale_from_voxel_size(
 #
 
 
-def _ribbon_AZ_postprocessing(predictions, vesicles, n_slices_exclude, n_ribbons):
+def _ribbon_AZ_postprocessing(predictions, vesicles, n_slices_exclude, n_ribbons, resolution, min_membrane_size):
     from synapse_net.inference.postprocessing import (
         segment_ribbon, segment_presynaptic_density, segment_membrane_distance_based,
     )
@@ -170,6 +170,7 @@ def _ribbon_AZ_postprocessing(predictions, vesicles, n_slices_exclude, n_ribbons
     ref_segmentation = PD if PD.sum() > 0 else ribbon
     membrane = segment_membrane_distance_based(
         predictions["membrane"], ref_segmentation, max_distance=500, n_slices_exclude=n_slices_exclude,
+        resolution=resolution, min_size=min_membrane_size,
     )
 
     segmentations = {"ribbon": ribbon, "PD": PD, "membrane": membrane}
@@ -182,6 +183,8 @@ def _segment_ribbon_AZ(image, model, tiling, scale, verbose, return_predictions=
     threshold = kwargs.pop("threshold", 0.5)
     n_slices_exclude = kwargs.pop("n_slices_exclude", 20)
     n_ribbons = kwargs.pop("n_slices_exclude", 1)
+    resolution = kwargs.pop("resolution", None)
+    min_membrane_size = kwargs.pop("min_membrane_size", 0)
 
     predictions = segment_ribbon_synapse_structures(
         image, model=model, tiling=tiling, scale=scale, verbose=verbose, threshold=threshold, **kwargs
@@ -197,7 +200,9 @@ def _segment_ribbon_AZ(image, model, tiling, scale, verbose, return_predictions=
     else:
         if verbose:
             print("Vesicle segmentation was passed, WILL run post-processing.")
-        segmentations = _ribbon_AZ_postprocessing(predictions, vesicles, n_slices_exclude, n_ribbons)
+        segmentations = _ribbon_AZ_postprocessing(
+            predictions, vesicles, n_slices_exclude, n_ribbons, resolution, min_membrane_size
+        )
 
     if return_predictions:
         return segmentations, predictions

synapse_net/inference/postprocessing/membranes.py

@@ -85,7 +85,19 @@ def segment_membrane_distance_based(
     n_slices_exclude: int,
     max_distance: float,
     resolution: Optional[float] = None,
+    min_size: int = 0,
 ):
+    """Derive boundary segmentation from boundary predictions by selecting the fragment closest to the ribbon.
+
+    Args:
+        boundary_prediction: Binary prediction for boundaries in the tomogram.
+        reference_segmentation: The reference segmentation, typically of the ribbon.
+        n_slices_exclude: The number of slices to exclude on the top / bottom
+            in order to avoid segmentation errors due to imaging artifacts in top and bottom.
+        max_distance: The maximal distance from the ribbon to consider.
+        resolution: The resolution / voxel size of the data.
+        min_size: The minimal size of a boundary fragment to be included.
+    """
     assert boundary_prediction.shape == reference_segmentation.shape
 
     original_shape = boundary_prediction.shape
@@ -95,6 +107,13 @@ def segment_membrane_distance_based(
     boundary_prediction = boundary_prediction[slice_mask]
     reference_segmentation = reference_segmentation[slice_mask]
 
+    if min_size > 0:
+        boundary_prediction = label(boundary_prediction, block_shape=(32, 256, 256))
+        ids, sizes = np.unique(boundary_prediction, return_counts=True)
+        ids, sizes = ids[1:], sizes[1:]
+        keep_ids = ids[sizes > min_size]
+        boundary_prediction = np.isin(boundary_prediction, keep_ids)
+
     # Get the unique objects in the reference segmentation.
     reference_ids = np.unique(reference_segmentation)
     assert reference_ids[0] == 0

synapse_net/inference/scalable_segmentation.py

@@ -79,6 +79,7 @@ def scalable_segmentation(
     prediction: Optional[ArrayLike] = None,
     verbose: bool = True,
     mask: Optional[ArrayLike] = None,
+    devices: Optional[List[str]] = None,
 ) -> None:
     """Run segmentation based on a prediction with foreground and boundary channel.
 
@@ -100,6 +101,8 @@ def scalable_segmentation(
             If given, this can be a numpy array, a zarr array, or similar
             If not given will be stored in a temporary n5 array.
         verbose: Whether to print timing information.
+        devices: The devices for running prediction. If not given will use the GPU
+            if available, otherwise the CPU.
     """
     if mask is not None:
         raise NotImplementedError
@@ -133,5 +136,5 @@ def scalable_segmentation(
         seeds = f.create_dataset("seeds", shape=input_.shape, dtype="uint64", chunks=chunks)
 
         # Run prediction and segmentation.
-        get_prediction(input_, prediction=prediction, tiling=tiling, model=model, verbose=verbose)
+        get_prediction(input_, prediction=prediction, tiling=tiling, model=model, verbose=verbose, devices=devices)
         _run_segmentation(prediction, output, seeds, chunks, seed_threshold, min_size, verbose, original_shape)

synapse_net/inference/util.py

@@ -125,6 +125,7 @@ def get_prediction(
     channels_to_standardize: Optional[List[int]] = None,
     mask: Optional[ArrayLike] = None,
     prediction: Optional[ArrayLike] = None,
+    devices: Optional[List[str]] = None,
 ) -> ArrayLike:
     """Run prediction on a given volume.
 
@@ -143,6 +144,8 @@ def get_prediction(
             the foreground region of the mask.
         prediction: An array like object for writing the prediction.
             If not given, the prediction will be computed in moemory.
+        devices: The devices for running prediction. If not given will use the GPU
+            if available, otherwise the CPU.
 
     Returns:
         The predicted volume.
@@ -189,7 +192,7 @@ def get_prediction(
         # print(f"updated_tiling {updated_tiling}")
         prediction = get_prediction_torch_em(
             input_volume, updated_tiling, model_path, model, verbose, with_channels,
-            mask=mask, prediction=prediction,
+            mask=mask, prediction=prediction, devices=devices,
         )
 
     return prediction
@@ -204,6 +207,7 @@ def get_prediction_torch_em(
     with_channels: bool = False,
     mask: Optional[ArrayLike] = None,
     prediction: Optional[ArrayLike] = None,
+    devices: Optional[List[str]] = None,
 ) -> np.ndarray:
     """Run prediction using torch-em on a given volume.
 
@@ -218,6 +222,8 @@ def get_prediction_torch_em(
             the foreground region of the mask.
         prediction: An array like object for writing the prediction.
             If not given, the prediction will be computed in moemory.
+        devices: The devices for running prediction. If not given will use the GPU
+            if available, otherwise the CPU.
 
     Returns:
         The predicted volume.
@@ -227,14 +233,15 @@ def get_prediction_torch_em(
     halo = [tiling["halo"]["z"], tiling["halo"]["x"], tiling["halo"]["y"]]
 
     t0 = time.time()
-    device = "cuda" if torch.cuda.is_available() else "cpu"
+    if devices is None:
+        devices = ["cuda" if torch.cuda.is_available() else "cpu"]
 
     # Suppress warning when loading the model.
     with warnings.catch_warnings():
         warnings.simplefilter("ignore")
         if model is None:
             if os.path.isdir(model_path):  # Load the model from a torch_em checkpoint.
-                model = torch_em.util.load_model(checkpoint=model_path, device=device)
+                model = torch_em.util.load_model(checkpoint=model_path, device=devices[0])
             else:  # Load the model directly from a serialized pytorch model.
                 model = torch.load(model_path, weights_only=False)
 
@@ -253,7 +260,7 @@ def get_prediction_torch_em(
 
         preprocess = None if isinstance(input_volume, np.ndarray) else torch_em.transform.raw.standardize
         prediction = predict_with_halo(
-            input_volume, model, gpu_ids=[device],
+            input_volume, model, gpu_ids=devices,
             block_shape=block_shape, halo=halo,
             preprocess=preprocess, with_channels=with_channels, mask=mask,
             output=prediction,

synapse_net/tools/segmentation_widget.py

@@ -122,7 +122,10 @@ def load_model_widget(self):
 
         # Exclude the models that are only offered through the CLI and not in the plugin.
         model_list = set(_get_model_registry().urls.keys())
-        excluded_models = ["vesicles_2d_maus", "vesicles_3d_endbulb", "vesicles_3d_innerear"]
+        # These are the models exlcuded due to their specificity and to keep the menu simple.
+        # TODO: we should at some point update the logic here, to make it easier to support further models
+        # without cluttering the UI.
+        excluded_models = ["vesicles_2d_maus"]
         model_list = [name for name in model_list if name not in excluded_models]
 
         models = ["- choose -"] + model_list
@@ -187,7 +190,8 @@ def on_predict(self):
         # For these models we read out the 'Extra Segmentation' widget.
         if model_type == "ribbon":  # Currently only the ribbon model needs the extra seg.
             extra_seg = self._get_layer_selector_data(self.extra_seg_selector_name)
-            kwargs = {"extra_segmentation": extra_seg}
+            resolution = tuple(voxel_size[ax] for ax in "zyx")
+            kwargs = {"extra_segmentation": extra_seg, "resolution": resolution, "min_membrane_size": 50_000}
         elif model_type == "cristae":  # Cristae model expects 2 3D volumes
             kwargs = {
                 "extra_segmentation": self._get_layer_selector_data(self.extra_seg_selector_name),