Implemented check for multiple connected objects in mask frame, and optional...

Implemented check for multiple connected objects in mask frame, and optional restriction to the largest-area one; inference on autogenerated model now works with 100% accuracy when passing on training data

extensions/chaeo/examples/transfer_labels_to_ilastik_object_classifier.py

@@ -58,7 +58,7 @@ def infer_and_compare(classifier: PatchStackObjectClassifier, prefix, raw, mask,
 
 
 if __name__ == '__main__':
-    root = Path('c:/Users/rhodes/projects/proj0011-plankton-seg/exp0009/output/labeled_patches-20231030-0002')
+    root = Path('c:/Users/rhodes/projects/proj0011-plankton-seg/exp0009/output/labeled_patches-20231030-0007')
     template_ilp = Path('c:/Users/rhodes/projects/proj0011-plankton-seg/exp0014/template_obj.ilp')
 
     df_labels = pd.read_csv(root / 'labels_key.csv')
@@ -78,21 +78,21 @@ if __name__ == '__main__':
     )
     classifier = PatchStackObjectClassifier({'project_file': classifier_file})
 
-    # # verify self-consistency of training set
-    # infer_and_compare(
-    #     classifier,
-    #     'train',
-    #     MonoPatchStackFromFile(root / 'zstack_train_raw.tif'),
-    #     MonoPatchStackFromFile(root / 'zstack_train_mask.tif'),
-    #     MonoPatchStackFromFile(root / 'zstack_train_label.tif')
-    # )
-    #
-    # # run test set
-    # infer_and_compare(
-    #     classifier,
-    #     'test',
-    #     MonoPatchStackFromFile(root / 'zstack_test_raw.tif'),
-    #     MonoPatchStackFromFile(root / 'zstack_test_mask.tif'),
-    #     MonoPatchStackFromFile(root / 'zstack_test_label.tif'),
-    # )
+    # verify self-consistency of training set
+    infer_and_compare(
+        classifier,
+        'train',
+        MonoPatchStackFromFile(root / 'zstack_train_raw.tif'),
+        MonoPatchStackFromFile(root / 'zstack_train_mask.tif'),
+        MonoPatchStackFromFile(root / 'zstack_train_label.tif')
+    )
+
+    # run test set
+    infer_and_compare(
+        classifier,
+        'test',
+        MonoPatchStackFromFile(root / 'zstack_test_raw.tif'),
+        MonoPatchStackFromFile(root / 'zstack_test_mask.tif'),
+        MonoPatchStackFromFile(root / 'zstack_test_label.tif'),
+    )
 

extensions/chaeo/workflows.py

@@ -4,6 +4,7 @@ from uuid import uuid4
 
 import numpy as np
 import pandas as pd
+from skimage.measure import label, regionprops_table
 from skimage.morphology import dilation
 from sklearn.model_selection import train_test_split
 
@@ -292,14 +293,15 @@ def infer_object_map_from_zstack(
 
 
 def transfer_ecotaxa_labels_to_patch_stacks(
-        where_masks: str,
-        where_patches: str,
-        object_csv: str,
-        ecotaxa_tsv: str,
-        where_output: str,
-        patch_size: tuple = (256, 256),
-        tr_split=0.6,
-        dilate_label_mask: bool = True, # to mitigate connected components error in ilastik
+    where_masks: str,
+    where_patches: str,
+    object_csv: str,
+    ecotaxa_tsv: str,
+    where_output: str,
+    patch_size: tuple = (256, 256),
+    tr_split=0.6,
+    dilate_label_mask: bool = True, # to mitigate connected components error in ilastik
+    allow_multiple_objects: bool = False,
 ) -> Dict:
     assert tr_split > 0.5 # reduce chance that low-probability objects are omitted from training
 
@@ -381,6 +383,14 @@ def transfer_ecotaxa_labels_to_patch_stacks(
             mask = acc_bm.data[:, :, 0, 0]
             if dilate_label_mask:
                 mask = dilation(mask)
+            if not allow_multiple_objects:
+                ob_id = label(acc_bm.data[:, :, 0, 0])
+                num_obj = len(np.unique(ob_id)) - 1
+                if num_obj > 1:
+                    print(f'Found multiple nonzero unique values in mask {fi}; keeping the one with largest area')
+                    pr = regionprops_table(ob_id, properties=['label', 'area'])
+                    idx_max_area = pr['area'].argmax()
+                    mask = 255 * (ob_id == pr['label'][idx_max_area])
             zstacks[dfk + '_mask'][:, :, 0, fi] = mask
             zstacks[dfk + '_label'][:, :, 0, fi] = (mask == 255) * aci