Cropping tool: add support for Neuroglancer Precomputed data

tomka · tomka · commit 6594af42fde0 · 2025-02-06T11:25:28.000+01:00
This update allows users to extract sub-volumes from Neuroglancer
Precomputed stacks using the Cropping Tool.

Also move the Neuroglancer Precomputed changelog text to the correct
position.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -34,6 +34,25 @@
 
 ### Features and enhancements
 
+Data sources:
+
+- A new tile source has been added: Neuroglancer Precomputed data, with ID 14.
+  This image block source works very similar to the N5 tile source. At the
+  moment only GZip and JPEG compression is supported. [Sharded](Sharded) datasets can be
+  loaded. In order for the voxel space coordinates to match between CATMAID and
+  Neuroglancer, if the Neuroglancer dataset defines a voxel offset, the
+  respective CATMAID stack needs to have its zoom-level zero voxel offset
+  defined in the stack meta data in the admin view, e.g. `{"voxelOffset":
+  [-3072, -3072, 0]}`.
+
+- Neuroglancer Precomputed URLs that start with `gs://` are rewritten on the fly
+  to use Google Cloud storage get API requess, improving CORS support.
+
+Cropping tool:
+
+- Neuroglancer Precomputed datasets (sharded and unsharded) are now supported
+  and can be used extract sub-volumes.
+
 Connector types:
 
 - A new connector has been added to link mitochondria to cells. It can be used
@@ -49,7 +68,6 @@ Neuron similarity widget:
   original NBLAST publication. The UI allows to adjust dot and distance breaks
   as well as an easy way to add an optional scale factor.
 
-
 Tracing data cache:
 
 - The management command `catmaid_update_cache_tables` can now remove existing
@@ -635,19 +653,7 @@ Vagrant:
 
 Data sources:
 
-- A new tile source has been added: Neuroglancer Precomputed data, with ID 14.
-  This image block source works very similar to the N5 tile source. At the
-  moment only GZip and JPEG compression is supported. Sharded datasets can be
-  loaded. In order for the voxel space coordinates to match between CATMAID and
-  Neuroglancer, if the Neuroglancer dataset defines a voxel offset, the
-  respective CATMAID stack needs to have its zoom-level zero voxel offset
-  defined in the stack meta data in the admin view, e.g. `{"voxelOffset":
-  [-3072, -3072, 0]}`.
-
-- Neuroglancer Precomputed URLs that start with `gs://` are rewritten on the fly
-  to use Google Cloud storage get API requess, improving CORS support.
-
-- A second new tile source is available: CloudVolume based tiles, generated by
+- A new tile source is available: CloudVolume based tiles, generated by
   the back-end. While slow for many users, it is a convenient way of make
   neuroglancer volumes available to CATMAID if they aren't supported by the
   Neuroglancer Precomputed source. It works better for a small set of users and
diff --git a/django/applications/catmaid/control/cropping.py b/django/applications/catmaid/control/cropping.py
@@ -7,6 +7,7 @@
 import os
 import os.path
 import math
+import numpy as np
 from PIL import Image as PILImage, TiffImagePlugin
 import requests
 from time import time
@@ -84,11 +85,10 @@ def __init__(self, user, project_id, stack_mirror_ids, x_min,
         self.single_channel = single_channel
         self.output_path = output_path
 
-    def get_tile_path(self, stack, mirror, tile_coords) -> str:
-        """ This method will be used when get_tile_path is called after the
-        job has been initialized.
-        """
+    def get_source_metadata(self, stack, mirror, tile_coords):
         tile_source = self.stack_tile_sources[stack.id]
+        if tile_source.is_block_source:
+            return tile_source.get_block_details(mirror, tile_coords, self.zoom_level)
         return tile_source.get_tile_url(mirror, tile_coords, self.zoom_level)
 
     def create_tiff_metadata(self, n_images):
@@ -181,8 +181,8 @@ class ImagePart:
     """ A part of a 2D image where height and width are not necessarily
     of the same size. Provides readout of the defined sub-area of the image.
     """
-    def __init__( self, path, x_min_src, x_max_src, y_min_src, y_max_src, x_dst, y_dst ):
-        self.path = path
+    def __init__( self, meta, x_min_src, x_max_src, y_min_src, y_max_src, x_dst, y_dst ):
+        self.meta = meta
         self.x_min_src = x_min_src
         self.x_max_src = x_max_src
         self.y_min_src = y_min_src
@@ -208,19 +208,39 @@ def __str__(self):
             f'({self.x_max_src}, {self.y_min_src})')
 
     def get_image(self):
-        # Open the image
-        try:
-            r = requests.get(self.path, allow_redirects=True, verify=verify_ssl, timeout=1)
-            if not r:
-                raise ValueError(f"Could not get {self.path}")
-            if r.status_code != 200:
-                raise ValueError(f"Unexpected status code ({r.status_code}) for {self.path}")
-            img_data = r.content
-            bytes_read = len(img_data)
-        except requests.exceptions.RequestException as e:
-            raise ImageRetrievalError(self.path, str(e))
-
-        image = PILImage.open(BytesIO(img_data))
+        if type(self.meta) == str:
+            # Open the image
+            try:
+                r = requests.get(self.meta, allow_redirects=True, verify=verify_ssl, timeout=1)
+                if not r:
+                    raise ValueError(f"Could not get {self.meta}")
+                if r.status_code != 200:
+                    raise ValueError(f"Unexpected status code ({r.status_code}) for {self.meta}")
+                img_data = r.content
+                bytes_read = len(img_data)
+            except requests.exceptions.RequestException as e:
+                raise ImageRetrievalError(self.meta, str(e))
+
+            image = PILImage.open(BytesIO(img_data))
+        elif type(self.meta) == dict:
+            mirror = self.meta['mirror']
+            tile_coord = self.meta['tile_coord']
+            x = tile_coord[0] * mirror.tile_width
+            y = tile_coord[1] * mirror.tile_height
+            z = self.meta['tile_coord'][2]
+
+            cutout = self.meta['dataset'][
+                x:(x + mirror.tile_width),
+                y:(y + mirror.tile_height),
+                z]
+
+            # FIXME: Don't just assume first channel (last dimension below)
+            image_data = np.transpose(cutout[:,:,0,0])
+            bytes_read = len(image_data)
+            image = PILImage.frombuffer('RGBA',
+                    (mirror.tile_width, mirror.tile_height), image_data, 'raw', 'L', 0, 1)
+        else:
+            raise ValueError(f'Unknown image meta data type: {self.meta}')
 
         src_width, src_height = image.size
 
@@ -374,6 +394,11 @@ class BB:
     width = 0
     height = 0
 
+    def __str__(self):
+        return f'Bounding Box: ({self.px_x_min}, {self.px_y_min}, {self.px_z_min}) - ' + \
+                f'({self.px_x_max}, {self.px_y_max}, {self.px_z_max}) ' + \
+                f'[width: {self.width}, height: {self.height}]'
+
 
 def extract_substack_no_rotation(job) -> List:
     """ Extracts a sub-stack as specified in the passed job without respecting
@@ -484,9 +509,9 @@ def extract_substack_no_rotation(job) -> List:
                         cur_px_y_max = bb.px_y_max - y * tile_height
                     # Create an image part definition
                     z = bb.px_z_min + nz
-                    path = job.get_tile_path(stack, mirror, (x, y, z))
+                    source_meta = job.get_source_metadata(stack, mirror, (x, y, z))
                     try:
-                        part = ImagePart(path, cur_px_x_min, cur_px_x_max,
+                        part = ImagePart(source_meta, cur_px_x_min, cur_px_x_max,
                                 cur_px_y_min, cur_px_y_max, x_dst, y_dst)
                         image_parts.append( part )
                     except Exception as e:
diff --git a/django/applications/catmaid/control/tile.py b/django/applications/catmaid/control/tile.py
@@ -214,7 +214,9 @@ def put_tile(request:HttpRequest, project_id=None, stack_id=None) -> HttpRespons
     return HttpResponse("Image pushed to HDF5.", content_type="plain/text")
 
 
-class TileSource(object):
+class TileSource():
+
+    is_block_source = False
 
     def get_canary_url(self, mirror) -> str:
         """Get the canary URL for this mirror.
@@ -287,10 +289,63 @@ def get_tile_url(self, mirror, tile_coords, zoom_level=0) -> str:
         return path
 
 
+class NeuroglancerTileSource(TileSource):
+    """ Uses cloud-volume to access neuroglancer data.
+    """
+
+    is_block_source = True
+    description = "Neuroglancer Precomputed image stack"
+    param_token = '%SCALE_DATASET%'
+
+    def __init__(self):
+        # Init cloud volume
+        self.dataset = None
+
+        super().__init__()
+
+    def normalize_url_path(self, url, path):
+        if not url[-1] == '/':
+            url = url + '/'
+
+        url = url + path
+
+        if url.startswith("gs://"):
+            components = url.split("/");
+            if len(components) < 4:
+                raise ValueError(f'Unsupported Google Cloud URL: {path}');
+            bucketName = components[2];
+            subPath = "%2F".join(components[3:])
+            return f'https://www.googleapis.com/storage/v1/b/{bucketName}/o/{subPath}?alt=media'
+
+        return url
+
+    def get_canary_url(self, mirror) -> str:
+        # Basic info URL
+        url = mirror.image_base[:mirror.image_base.index(self.param_token)]
+        return self.normalize_url_path(url, 'info')
+
+    def get_block_details(self, mirror, tile_coords, zoom_level=0) -> dict:
+        if not cloudvolume_available:
+            raise ValueError('Cloudvolume not available')
+        if not self.dataset:
+            url = mirror.image_base[:mirror.image_base.index(self.param_token)]
+            self.dataset = cloudvolume.CloudVolume(f"precomputed://{url}",
+                                                   mip=zoom_level,
+                                                   use_https=True,
+                                                   bounded=False)
+        return {
+            'dataset': self.dataset,
+            'tile_coord': tile_coords,
+            'zoom_level': zoom_level,
+            'mirror': mirror,
+        }
+
+
 tile_source_map = {
     TileSourceTypes.FILE_BASED_STACK: DefaultTileSource,
     TileSourceTypes.FILE_BASED_ZOOM_STACK: BackslashTileSource,
-    TileSourceTypes.DIRECTORY_BASED_STACK: LargeDataTileSource
+    TileSourceTypes.DIRECTORY_BASED_STACK: LargeDataTileSource,
+    TileSourceTypes.NEUROGLANCER_PRECOMPUTED: NeuroglancerTileSource,
 }
 
 def get_tile_source(type_id):
