add 3d axon seg

Author: Donglai Wei

CHECKERBOARD_FIX.md

@@ -147,6 +147,7 @@ class ModelConfig:
     kernel_size: int = 3  # Convolution kernel size
     strides: Optional[List[int]] = None  # Downsampling strides (e.g., [2, 2, 2, 2] for 4 levels)
     act: str = "relu"  # Activation function: 'relu', 'prelu', 'elu', etc.
+    upsample: str = "deconv"  # Upsampling mode for MONAI BasicUNet: 'deconv' (transposed conv), 'nontrainable' (interpolation + conv), or 'pixelshuffle'
 
     # Transformer-specific (UNETR, etc.)
     feature_size: int = 16
@@ -926,17 +927,15 @@ class BinaryPostprocessingConfig:
     """Binary segmentation postprocessing configuration.
 
     Applies morphological operations and connected components filtering to binary predictions.
+    Input should already be binary (from binary_thresholding decoding).
     Pipeline order:
-        1. Threshold predictions to binary mask (using threshold_range if provided)
+        1. Ensure input is binary (auto-threshold if needed: 0.5 for [0,1], 0 for >1)
         2. Apply morphological opening (erosion + dilation)
         3. Extract connected components
         4. Keep top-k largest components
     """
 
     enabled: bool = False  # Enable binary postprocessing pipeline
-    threshold_range: Optional[List[float]] = (
-        None  # Threshold range [min, max] for binarization. If None, uses 0.5 for [0,1] or 0 for >1
-    )
     median_filter_size: Optional[Tuple[int, ...]] = (
         None  # Median filter kernel size (e.g., (3, 3) for 2D)
     )
@@ -960,16 +959,16 @@ class PostprocessingConfig:
     """Postprocessing configuration for inference output.
 
     Controls how predictions are transformed before saving:
-    - Thresholding: Binarize predictions using threshold_range
+    - Binary refinement: Morphological operations and connected components filtering
     - Scaling: Multiply intensity values (e.g., 255 for uint8)
     - Dtype conversion: Convert to target data type with proper clamping
     - Transpose: Reorder axes (e.g., [2,1,0] for zyx->xyz)
     """
 
-    # Thresholding configuration
+    # Binary segmentation refinement (morphological ops, connected components)
     binary: Optional[Dict[str, Any]] = field(
         default_factory=dict
-    )  # Binary thresholding config (e.g., {'threshold_range': [0.5, 1.0]})
+    )  # Binary postprocessing config (e.g., {'opening_iterations': 2})
 
     # Intensity scaling
     intensity_scale: Optional[float] = (

connectomics/config/hydra_config.py

@@ -291,8 +291,43 @@ def _combine_path(base_path: str, file_path: Optional[Union[str, List[str]]]) ->
         if base_path and not os.path.isabs(file_path):
             file_path = os.path.join(base_path, file_path)
 
-        # Expand glob patterns
-        if "*" in file_path or "?" in file_path:
+        # Expand glob patterns with optional selector support
+        # Format: path/*.tiff[0] or path/*.tiff[filename]
+        import re
+        selector_match = re.match(r'^(.+)\[(.+)\]$', file_path)
+
+        if selector_match:
+            # Has selector - extract glob pattern and selector
+            glob_pattern = selector_match.group(1)
+            selector = selector_match.group(2)
+
+            expanded = sorted(glob(glob_pattern))
+            if not expanded:
+                return file_path  # No matches - return original
+
+            # Select file based on selector
+            try:
+                # Try numeric index
+                index = int(selector)
+                if index < -len(expanded) or index >= len(expanded):
+                    print(f"Warning: Index {index} out of range for {len(expanded)} files, using first")
+                    return expanded[0]
+                return expanded[index]
+            except ValueError:
+                # Not a number, try filename match
+                from pathlib import Path
+                matching = [f for f in expanded if Path(f).name == selector or Path(f).stem == selector]
+                if not matching:
+                    # Try partial match
+                    matching = [f for f in expanded if selector in Path(f).name]
+                if matching:
+                    return matching[0]
+                else:
+                    print(f"Warning: No file matches selector '{selector}', using first of {len(expanded)} files")
+                    return expanded[0]
+
+        elif "*" in file_path or "?" in file_path:
+            # Standard glob without selector
             expanded = sorted(glob(file_path))
             if expanded:
                 return expanded
@@ -302,32 +337,34 @@ def _combine_path(base_path: str, file_path: Optional[Union[str, List[str]]]) ->
 
         return file_path
 
-    # Resolve training paths
-    if cfg.data.train_path:
-        cfg.data.train_image = _combine_path(cfg.data.train_path, cfg.data.train_image)
-        cfg.data.train_label = _combine_path(cfg.data.train_path, cfg.data.train_label)
-        cfg.data.train_mask = _combine_path(cfg.data.train_path, cfg.data.train_mask)
-        cfg.data.train_json = _combine_path(cfg.data.train_path, cfg.data.train_json)
-
-    # Resolve validation paths
-    if cfg.data.val_path:
-        cfg.data.val_image = _combine_path(cfg.data.val_path, cfg.data.val_image)
-        cfg.data.val_label = _combine_path(cfg.data.val_path, cfg.data.val_label)
-        cfg.data.val_mask = _combine_path(cfg.data.val_path, cfg.data.val_mask)
-        cfg.data.val_json = _combine_path(cfg.data.val_path, cfg.data.val_json)
-
-    # Resolve test paths
-    if cfg.data.test_path:
-        cfg.data.test_image = _combine_path(cfg.data.test_path, cfg.data.test_image)
-        cfg.data.test_label = _combine_path(cfg.data.test_path, cfg.data.test_label)
-        cfg.data.test_mask = _combine_path(cfg.data.test_path, cfg.data.test_mask)
-        cfg.data.test_json = _combine_path(cfg.data.test_path, cfg.data.test_json)
+    # Resolve training paths (always expand globs, use train_path as base if available)
+    train_base = cfg.data.train_path if cfg.data.train_path else ""
+    cfg.data.train_image = _combine_path(train_base, cfg.data.train_image)
+    cfg.data.train_label = _combine_path(train_base, cfg.data.train_label)
+    cfg.data.train_mask = _combine_path(train_base, cfg.data.train_mask)
+    cfg.data.train_json = _combine_path(train_base, cfg.data.train_json)
+
+    # Resolve validation paths (always expand globs, use val_path as base if available)
+    val_base = cfg.data.val_path if cfg.data.val_path else ""
+    cfg.data.val_image = _combine_path(val_base, cfg.data.val_image)
+    cfg.data.val_label = _combine_path(val_base, cfg.data.val_label)
+    cfg.data.val_mask = _combine_path(val_base, cfg.data.val_mask)
+    cfg.data.val_json = _combine_path(val_base, cfg.data.val_json)
+
+    # Resolve test paths (always expand globs, use test_path as base if available)
+    test_base = cfg.data.test_path if cfg.data.test_path else ""
+    cfg.data.test_image = _combine_path(test_base, cfg.data.test_image)
+    cfg.data.test_label = _combine_path(test_base, cfg.data.test_label)
+    cfg.data.test_mask = _combine_path(test_base, cfg.data.test_mask)
+    cfg.data.test_json = _combine_path(test_base, cfg.data.test_json)
 
     # Resolve inference data paths (primary location for test_path)
+    inference_test_base = ""
     if hasattr(cfg.inference.data, 'test_path') and cfg.inference.data.test_path:
-        cfg.inference.data.test_image = _combine_path(cfg.inference.data.test_path, cfg.inference.data.test_image)
-        cfg.inference.data.test_label = _combine_path(cfg.inference.data.test_path, cfg.inference.data.test_label)
-        cfg.inference.data.test_mask = _combine_path(cfg.inference.data.test_path, cfg.inference.data.test_mask)
+        inference_test_base = cfg.inference.data.test_path
+    cfg.inference.data.test_image = _combine_path(inference_test_base, cfg.inference.data.test_image)
+    cfg.inference.data.test_label = _combine_path(inference_test_base, cfg.inference.data.test_label)
+    cfg.inference.data.test_mask = _combine_path(inference_test_base, cfg.inference.data.test_mask)
 
     return cfg