datamint 2.1.0__py3-none-any.whl → 2.1.2__py3-none-any.whl

datamint/api/base_api.py

@@ -14,6 +14,7 @@ from io import BytesIO
 import gzip
 import contextlib
 import asyncio
+from medimgkit.format_detection import GZIP_MIME_TYPES
 
 logger = logging.getLogger(__name__)
 
@@ -422,7 +423,7 @@ class BaseApi:
                 if file_path is not None:
                     return nib.load(file_path)
                 raise e
-        elif mimetype == 'application/gzip':
+        elif mimetype in GZIP_MIME_TYPES:
             # let's hope it's a .nii.gz
             with gzip.open(content_io, 'rb') as f:
                 return nib.Nifti1Image.from_stream(f)

datamint/dataset/base_dataset.py

@@ -360,11 +360,13 @@ class DatamintBaseDataset:
     @property
     def segmentation_labels_set(self) -> list[str]:
         """Returns the set of segmentation labels in the dataset."""
-        return self.frame_lsets['segmentation']
+        a = set(self.frame_lsets['segmentation'])
+        b = set(self.image_lsets['segmentation'])
+        return list(a.union(b))
 
     def _get_annotations_internal(
         self,
-        annotations: list[Annotation],
+        annotations: Sequence[Annotation],
         type: Literal['label', 'category', 'segmentation', 'all'] = 'all',
         scope: Literal['frame', 'image', 'all'] = 'all'
     ) -> list[Annotation]:
@@ -441,10 +443,8 @@ class DatamintBaseDataset:
 
     def get_resources_ids(self) -> list[str]:
         """Get list of resource IDs."""
-        return [
-            self.__getitem_internal(i, only_load_metainfo=True)['metainfo']['id']
-            for i in self.subset_indices
-        ]
+        return [self._get_image_metainfo(i, bypass_subset_indices=True)['metainfo']['id']
+                for i in self.subset_indices]
 
     def _get_labels_set(self, framed: bool) -> tuple[dict, dict[str, dict[str, int]]]:
         """Returns the set of labels and mappings to integers.
@@ -992,7 +992,6 @@ class DatamintBaseDataset:
             return Path(resource['file'])
         else:
             # ext = guess_extension(resource['mimetype'])
-            # _LOGGER.debug(f"Guessed extension for resource {resource['id']}|{resource['mimetype']}: {ext}")
             # if ext is None:
             #     _LOGGER.warning(f"Could not guess extension for resource {resource['id']}.")
             #     ext = ''

datamint/dataset/dataset.py

@@ -1,5 +1,5 @@
 from .base_dataset import DatamintBaseDataset
-from typing import List, Optional, Callable, Any, Dict, Literal
+from typing import List, Optional, Callable, Any, Dict, Literal, Sequence
 import torch
 from torch import Tensor
 import os
@@ -8,6 +8,7 @@ import logging
 from PIL import Image
 import albumentations
 from datamint.entities.annotation import Annotation
+from medimgkit.readers import read_array_normalized
 
 _LOGGER = logging.getLogger(__name__)
 
@@ -117,7 +118,9 @@ class DatamintDataset(DatamintBaseDataset):
         if semantic_seg_merge_strategy is not None and not return_as_semantic_segmentation:
             raise ValueError("semantic_seg_merge_strategy can only be used if return_as_semantic_segmentation is True")
 
-    def _load_segmentations(self, annotations: list[Annotation], img_shape) -> tuple[dict[str, list], dict[str, list]]:
+    def _load_segmentations(self,
+                            annotations: list[Annotation],
+                            img_shape) -> tuple[dict[str, list], dict[str, list]]:
         """
         Load segmentations from annotations.
 
@@ -152,19 +155,27 @@ class DatamintDataset(DatamintBaseDataset):
 
             segfilepath = ann.file  # png file
             segfilepath = os.path.join(self.dataset_dir, segfilepath)
-            # FIXME: avoid enforcing resizing the mask
-            seg = (Image.open(segfilepath)
-                   .convert('L')
-                   .resize((w, h), Image.Resampling.NEAREST)
-                   )
-            seg = np.array(seg)
+            seg = read_array_normalized(segfilepath)  # (frames, C, H, W)
+            if seg.shape[1] != 1:
+                raise ValueError(f"Segmentation file must have 1 channel, got {seg.shape} in {segfilepath}")
+            seg = seg[:, 0, :, :]  # (frames, H, W)
+            
+            # # FIXME: avoid enforcing resizing the mask
+            # seg = (Image.open(segfilepath)
+            #        .convert('L')
+            #        .resize((w, h), Image.Resampling.NEAREST)
+            #        )
+            # seg = np.array(seg)
 
             seg = torch.from_numpy(seg)
             seg = seg == 255   # binary mask
             # map the segmentation label to the code
-            seg_code = self.frame_lcodes['segmentation'][ann.name]
             if self.return_frame_by_frame:
                 frame_index = 0
+                if seg.shape[0] != 1:
+                    raise NotImplementedError(
+                        "Volume segmentations are not supported yet when return_frame_by_frame is True")
+                seg = seg[0:1]  # (#frames, H, W) -> (1, H, W)
             else:
                 frame_index = ann.index
 
@@ -174,12 +185,25 @@ class DatamintDataset(DatamintBaseDataset):
             author_segs = segmentations[author]
             author_labels = seg_labels[author]
 
-            if author_segs[frame_index] is None:
-                author_segs[frame_index] = []
-                author_labels[frame_index] = []
-
-            author_segs[frame_index].append(seg)
-            author_labels[frame_index].append(seg_code)
+            if frame_index is not None and ann.scope == 'frame':
+                seg_code = self.frame_lcodes['segmentation'][ann.name]
+                if author_segs[frame_index] is None:
+                    author_segs[frame_index] = []
+                    author_labels[frame_index] = []
+                s = seg[0] if seg.shape[0] == 1 else seg[frame_index]
+                author_segs[frame_index].append(s)
+                author_labels[frame_index].append(seg_code)
+            elif frame_index is None and ann.scope == 'image':
+                seg_code = self.image_lcodes['segmentation'][ann.name]
+                # apply to all frames
+                for i in range(nframes):
+                    if author_segs[i] is None:
+                        author_segs[i] = []
+                        author_labels[i] = []
+                    author_segs[i].append(seg[i])
+                    author_labels[i].append(seg_code)
+            else:
+                raise ValueError(f"Invalid segmentation annotation: {ann}")
 
         # convert to tensor
         for author in segmentations.keys():
@@ -196,8 +220,8 @@ class DatamintDataset(DatamintBaseDataset):
         return segmentations, seg_labels
 
     def _instanceseg2semanticseg(self,
-                                 segmentations: List[Tensor],
-                                 seg_labels: List[Tensor]) -> Tensor:
+                                 segmentations: Sequence[Tensor],
+                                 seg_labels: Sequence[Tensor]) -> Tensor:
         """
         Convert instance segmentation to semantic segmentation.
 
@@ -208,25 +232,26 @@ class DatamintDataset(DatamintBaseDataset):
         Returns:
             Tensor: tensor of shape (n, num_labels, H, W), where `n` is the number of frames.
         """
-        if segmentations is not None:
-            if len(segmentations) != len(seg_labels):
-                raise ValueError("segmentations and seg_labels must have the same length")
-
-            h, w = segmentations[0].shape[1:]
-            new_shape = (len(segmentations),
-                         len(self.segmentation_labels_set)+1,  # +1 for background
-                         h, w)
-            new_segmentations = torch.zeros(new_shape, dtype=torch.uint8)
-            # for each frame
-            for i in range(len(segmentations)):
-                # for each instance
-                for j in range(len(segmentations[i])):
-                    new_segmentations[i, seg_labels[i][j]] += segmentations[i][j]
-            new_segmentations = new_segmentations > 0
-            # pixels that are not in any segmentation are labeled as background
-            new_segmentations[:, 0] = new_segmentations.sum(dim=1) == 0
-            segmentations = new_segmentations.float()
-        return segmentations
+        if segmentations is None:
+            return None
+
+        if len(segmentations) != len(seg_labels):
+            raise ValueError("segmentations and seg_labels must have the same length")
+
+        h, w = segmentations[0].shape[1:]
+        new_shape = (len(segmentations),
+                     len(self.segmentation_labels_set)+1,  # +1 for background
+                     h, w)
+        new_segmentations = torch.zeros(new_shape, dtype=torch.uint8)
+        # for each frame
+        for i in range(len(segmentations)):
+            # for each instance
+            for j in range(len(segmentations[i])):
+                new_segmentations[i, seg_labels[i][j]] += segmentations[i][j]
+        new_segmentations = new_segmentations > 0
+        # pixels that are not in any segmentation are labeled as background
+        new_segmentations[:, 0] = new_segmentations.sum(dim=1) == 0
+        return new_segmentations.float()
 
     def apply_semantic_seg_merge_strategy(self, segmentations: dict[str, Tensor],
                                           nframes: int,
@@ -338,7 +363,7 @@ class DatamintDataset(DatamintBaseDataset):
                 if isinstance(labels, Tensor):
                     # single tensor for the author
                     seg_names[author] = [code_to_name[code.item()-1] for code in labels]
-                elif isinstance(labels, list):
+                elif isinstance(labels, Sequence):
                     # list of frame tensors
                     seg_names[author] = [[code_to_name[code.item()-1] for code in frame_labels]
                                          for frame_labels in labels]
@@ -477,7 +502,7 @@ class DatamintDataset(DatamintBaseDataset):
         return new_item
 
     def _convert_labels_annotations(self,
-                                    annotations: list[Annotation],
+                                    annotations: Sequence[Annotation],
                                     num_frames: int | None = None) -> dict[str, torch.Tensor]:
         """
         Converts the annotations, of the same type and scope, to tensor of shape (num_frames, num_labels)

{datamint-2.1.0.dist-info → datamint-2.1.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: datamint
-Version: 2.1.0
+Version: 2.1.2
 Summary: A library for interacting with the Datamint API, designed for efficient data management, processing and Deep Learning workflows.
 Requires-Python: >=3.10
 Classifier: Programming Language :: Python :: 3
@@ -21,7 +21,7 @@ Requires-Dist: humanize (>=4.0.0,<5.0.0)
 Requires-Dist: lazy-loader (>=0.3.0)
 Requires-Dist: lightning
 Requires-Dist: matplotlib
-Requires-Dist: medimgkit (>=0.6.2)
+Requires-Dist: medimgkit (>=0.6.4)
 Requires-Dist: nest-asyncio (>=1.0.0,<2.0.0)
 Requires-Dist: nibabel (>=4.0.0)
 Requires-Dist: numpy

{datamint-2.1.0.dist-info → datamint-2.1.2.dist-info}/RECORD

@@ -1,6 +1,6 @@
 datamint/__init__.py,sha256=ucsnxrYClh6pdy7psRJXWam_9rjAQB4NXzvy7xLovmo,824
 datamint/api/__init__.py,sha256=7QYkmDBXbKh8-zchV7k6Lpolaw6h-IK6ezfXROIWh2A,43
-datamint/api/base_api.py,sha256=MIq1sQA4mD9_SWxAEDjxtxm3Q-tj6kZ05KRnNoLPM7E,16576
+datamint/api/base_api.py,sha256=Iu9oJEZ8YlIF5xcbH_M0Lkb7t9ZDNFLzjJp9bPDHW1c,16628
 datamint/api/client.py,sha256=1XTZUlbAISe0jwug1rrANgWJToXxYeXx8_HD-ZWJurU,3354
 datamint/api/dto/__init__.py,sha256=KOSNl1axDDE5eBt68MmsgkyE0Ds_1DDzWUg73iyoWvc,281
 datamint/api/endpoints/__init__.py,sha256=S46nVAlXgGe8wNcBEhW8ffGJjGNAmhhRTDTsvG9fWBE,402
@@ -24,8 +24,8 @@ datamint/client_cmd_tools/datamint_upload.py,sha256=jPzvlNeBZfOOxuG6ryswJ8OG4jXu
 datamint/configs.py,sha256=Bdp6NydYwyCJ2dk19_gf_o3M2ZyQOmMHpLi8wEWNHUk,1426
 datamint/dataset/__init__.py,sha256=4PlUKSvVhdfQvvuq8jQXrkdqnot-iTTizM3aM1vgSwg,47
 datamint/dataset/annotation.py,sha256=qN1IMjdfLD2ceQ6va3l76jOXA8Vb_c-eBk1oWQu6hW0,7994
-datamint/dataset/base_dataset.py,sha256=Ai_fEMW1lmA9ZxGc_7Zuz3zjVUsbzpbnWeDMy7eZLVQ,49583
-datamint/dataset/dataset.py,sha256=It-HOTi83ls4ww2qCAvFYU0_OLLrFclj0QQapMYgDAE,27333
+datamint/dataset/base_dataset.py,sha256=xdWnYNZ6yKi__YGkrJ3hssiljRV3InGlJn8zDcgJnp8,49547
+datamint/dataset/dataset.py,sha256=huUOyBRGVtcx0tcpX2FrsWn7Vsqy5i5e_J52gxR_29A,28637
 datamint/entities/__init__.py,sha256=tbHE7rZb0R9Hm-Dc8VWEq3PlRl7BYOzffumrV0ZdsMs,444
 datamint/entities/annotation.py,sha256=ochAEh_JqxAe_FyYTNUfPT47KiIAG7CkBTim52bu7M8,6636
 datamint/entities/base_entity.py,sha256=DniakCgJ-gV7Hz8VKQA_dRYTp4DU5rcjLBVOuD1aZuA,1902
@@ -44,7 +44,7 @@ datamint/logging.yaml,sha256=tOMxtc2UmwlIMTK6ljtnBwTco1PNrPeq3mx2iMuSbiw,482
 datamint/utils/logging_utils.py,sha256=9pRoaPrWu2jOdDCiAoUsjEdP5ZwaealWL3hjUqFvx9g,4022
 datamint/utils/torchmetrics.py,sha256=lwU0nOtsSWfebyp7dvjlAggaqXtj5ohSEUXOg3L0hJE,2837
 datamint/utils/visualization.py,sha256=yaUVAOHar59VrGUjpAWv5eVvQSfztFG0eP9p5Vt3l-M,4470
-datamint-2.1.0.dist-info/METADATA,sha256=JFlip7Zr_9y7MzxTz4NUw3vNbDQtAI8rOtiSRtqPbwM,4203
-datamint-2.1.0.dist-info/WHEEL,sha256=M5asmiAlL6HEcOq52Yi5mmk9KmTVjY2RDPtO4p9DMrc,88
-datamint-2.1.0.dist-info/entry_points.txt,sha256=mn5H6jPjO-rY0W0CAZ6Z_KKWhMLvyVaSpoqk77jlTI4,145
-datamint-2.1.0.dist-info/RECORD,,
+datamint-2.1.2.dist-info/METADATA,sha256=K-uz7hcmvpOky8LA-r1hkOBMsXyRW1us9qT8F63mn6s,4203
+datamint-2.1.2.dist-info/WHEEL,sha256=zp0Cn7JsFoX2ATtOhtaFYIiE2rmFAD4OcMhtUki8W3U,88
+datamint-2.1.2.dist-info/entry_points.txt,sha256=mn5H6jPjO-rY0W0CAZ6Z_KKWhMLvyVaSpoqk77jlTI4,145
+datamint-2.1.2.dist-info/RECORD,,

{datamint-2.1.0.dist-info → datamint-2.1.2.dist-info}/WHEEL

@@ -1,4 +1,4 @@
 Wheel-Version: 1.0
-Generator: poetry-core 2.2.0
+Generator: poetry-core 2.2.1
 Root-Is-Purelib: true
 Tag: py3-none-any