Merge branch 'master' of https://github.com/zudi-lin/pytorch_connectomics

Author: Donglai Wei

connectomics/config/hydra_config.py

@@ -329,10 +329,14 @@ class DataConfig:
     - Multi-channel label transformations
     - Train/validation splitting options
     - Caching and performance optimization
+    - 2D data support with do_2d parameter
     """
 
     # Dataset type
     dataset_type: Optional[str] = None  # Type of dataset: None (volume), 'filename', 'tile', etc.
+    
+    # 2D data support
+    do_2d: bool = False  # Enable 2D data processing (extract 2D slices from 3D volumes)
 
     # Base path (prepended to train_image, train_label, etc. if set)
     train_path: str = ""  # Base path for training data (e.g., "/path/to/dataset/")
@@ -765,6 +769,9 @@ class InferenceDataConfig:
         default_factory=list
     )  # Axis permutation for test data (e.g., [2,1,0] for xyz->zyx)
     output_path: str = "results/"
+    
+    # 2D data support
+    do_2d: bool = False  # Enable 2D data processing for inference
 
 
 @dataclass

connectomics/data/dataset/dataset_base.py

@@ -70,7 +70,11 @@ def __init__(
         super().__init__(data=data_dicts, transform=transforms)
 
         # Store connectomics-specific parameters
-        self.sample_size = ensure_tuple_rep(sample_size, 3)
+        # For 2D data, use 2D dimensions; otherwise use 3D
+        if do_2d:
+            self.sample_size = ensure_tuple_rep(sample_size, 2)
+        else:
+            self.sample_size = ensure_tuple_rep(sample_size, 3)
         self.mode = mode
         self.iter_num = iter_num
         self.valid_ratio = valid_ratio

connectomics/lightning/lit_model.py

@@ -156,6 +156,15 @@ def _setup_sliding_window_inferer(self):
         if not hasattr(self.cfg, 'inference'):
             return
 
+        # For 2D models with do_2d=True, disable sliding window inference
+        if getattr(self.cfg.data, 'do_2d', False):
+            warnings.warn(
+                "Sliding-window inference disabled for 2D models with do_2d=True. "
+                "Using direct inference instead.",
+                UserWarning,
+            )
+            return
+
         roi_size = self._resolve_inferer_roi_size()
         if roi_size is None:
             warnings.warn(
@@ -200,12 +209,20 @@ def _resolve_inferer_roi_size(self) -> Optional[Tuple[int, ...]]:
         if hasattr(self.cfg, 'model') and hasattr(self.cfg.model, 'output_size'):
             output_size = getattr(self.cfg.model, 'output_size', None)
             if output_size:
-                return tuple(int(v) for v in output_size)
+                roi_size = tuple(int(v) for v in output_size)
+                # For 2D models with do_2d=True, convert to 3D ROI size
+                if getattr(self.cfg.data, 'do_2d', False) and len(roi_size) == 2:
+                    roi_size = (1,) + roi_size  # Add depth dimension
+                return roi_size
 
         if hasattr(self.cfg, 'data') and hasattr(self.cfg.data, 'patch_size'):
             patch_size = getattr(self.cfg.data, 'patch_size', None)
             if patch_size:
-                return tuple(int(v) for v in patch_size)
+                roi_size = tuple(int(v) for v in patch_size)
+                # For 2D models with do_2d=True, convert to 3D ROI size
+                if getattr(self.cfg.data, 'do_2d', False) and len(roi_size) == 2:
+                    roi_size = (1,) + roi_size  # Add depth dimension
+                return roi_size
 
         return None
 
@@ -367,6 +384,10 @@ def _predict_with_tta(self, images: torch.Tensor, mask: Optional[torch.Tensor] =
                 f"Expected shapes: (D, H, W), (B, D, H, W), or (B, C, D, H, W)"
             )
 
+        # For 2D models with do_2d=True, squeeze the depth dimension if present
+        if getattr(self.cfg.data, 'do_2d', False) and images.size(2) == 1:  # [B, C, 1, H, W] -> [B, C, H, W]
+            images = images.squeeze(2)
+
         # Get TTA configuration (default to no augmentation if not configured)
         if hasattr(self.cfg, 'inference') and hasattr(self.cfg.inference, 'test_time_augmentation'):
             tta_flip_axes_config = getattr(self.cfg.inference.test_time_augmentation, 'flip_axes', None)
@@ -385,23 +406,21 @@ def _predict_with_tta(self, images: torch.Tensor, mask: Optional[torch.Tensor] =
             ensemble_result = self._apply_tta_preprocessing(pred)
         else:
             if tta_flip_axes_config == 'all' or tta_flip_axes_config == []:
-                # "all" or []: All 8 flips (all combinations of Z, Y, X)
-                # IMPORTANT: MONAI Flip spatial_axis behavior for (B, C, D, H, W) tensors:
-                #   spatial_axis=[0] flips C (channel) - WRONG for TTA!
-                #   spatial_axis=[1] flips D (depth/Z) - CORRECT
-                #   spatial_axis=[2] flips H (height/Y) - CORRECT
-                #   spatial_axis=[3] flips W (width/X) - CORRECT
-                # Must use [1, 2, 3] for [D, H, W] flips, NOT [0, 1, 2]!
-                tta_flip_axes = [
-                    [],           # No flip
-                    [1],          # Flip Z (depth)
-                    [2],          # Flip Y (height)
-                    [3],          # Flip X (width)
-                    [1, 2],       # Flip Z+Y
-                    [1, 3],       # Flip Z+X
-                    [2, 3],       # Flip Y+X
-                    [1, 2, 3],    # Flip Z+Y+X
-                ]
+                # "all" or []: All flips (all combinations of spatial axes)
+                # Determine spatial axes based on data dimensions
+                if images.dim() == 5:  # 3D data: [B, C, D, H, W]
+                    spatial_axes = [1, 2, 3]  # [D, H, W]
+                elif images.dim() == 4:  # 2D data: [B, C, H, W]
+                    spatial_axes = [1, 2]  # [H, W]
+                else:
+                    raise ValueError(f"Unsupported data dimensions: {images.dim()}")
+                
+                # Generate all combinations of spatial axes
+                tta_flip_axes = [[]]  # No flip baseline
+                for r in range(1, len(spatial_axes) + 1):
+                    from itertools import combinations
+                    for combo in combinations(spatial_axes, r):
+                        tta_flip_axes.append(list(combo))
             elif isinstance(tta_flip_axes_config, (list, tuple)):
                 # Custom list: Add no-flip baseline + user-specified flips
                 tta_flip_axes = [[]] + list(tta_flip_axes_config)

connectomics/models/arch/monai_models.py

@@ -36,7 +36,18 @@ def __init__(self, model: nn.Module):
 
     def forward(self, x: torch.Tensor) -> torch.Tensor:
         """Forward pass through MONAI model."""
-        return self.model(x)
+        # For 2D models, squeeze the depth dimension if present
+        if x.dim() == 5 and x.size(2) == 1:  # [B, C, 1, H, W] -> [B, C, H, W]
+            x = x.squeeze(2)
+        
+        # Forward through model
+        output = self.model(x)
+        
+        # For 2D models, add back the depth dimension if needed for sliding window inference
+        if output.dim() == 4 and x.dim() == 5:  # [B, C, H, W] -> [B, C, 1, H, W]
+            output = output.unsqueeze(2)
+        
+        return output
 
 
 def _check_monai_available():

scripts/main.py

@@ -822,6 +822,7 @@ def setup(self, stage=None):
             cache_rate=cfg.data.cache_rate if use_cache else 0.0,
             iter_num=iter_num_for_dataset,
             sample_size=tuple(cfg.data.patch_size),
+            do_2d=cfg.data.do_2d,
         )
         # Setup datasets based on mode
         if mode == "train":

tutorials/monai2d_worm.yaml

@@ -58,6 +58,9 @@ model:
 
 # Data - Using automatic 80/20 train/val split (DeepEM-style)
 data:
+  # 2D data support
+  do_2d: true                     # Enable 2D data processing (extract 2D slices from 3D volumes)
+  
   # Volume configuration
   train_image: /projects/weilab/shenb/PyTC/datasets/Dataset001_worm_image96/imagesTr/*.tif
   train_label: /projects/weilab/shenb/PyTC/datasets/Dataset001_worm_image96/labelsTr/*.tif
@@ -152,8 +155,9 @@ monitor:
 # Inference - MONAI SlidingWindowInferer
 inference:
   data:
-    test_image: /projects/weilab/shenb/PyTC/datasets/Dataset001_worm_image96/imagesTr/Image96_00002_0000.tif
-    test_label: /projects/weilab/shenb/PyTC/datasets/Dataset001_worm_image96/labelsTr/Image96_00002.tif
+    do_2d: true                     # Enable 2D data processing for inference
+    test_image: /projects/weilab/shenb/PyTC/datasets/Dataset001_worm_image96/imagesTs/*.tif
+    test_label: /projects/weilab/shenb/PyTC/datasets/Dataset001_worm_image96/imagesTs/*.tif
     test_resolution: [5, 5]
     output_path: outputs/monai2d_worm/results/
 
@@ -188,7 +192,7 @@ inference:
   # Evaluation
   evaluation:
     enabled: true                        # Use eval mode for BatchNorm
-    metrics: [jaccard]             # Metrics to compute
+    metrics: [adapted_rand]        # Metrics to compute (adapted_rand for instance segmentation)
 
   # NOTE: batch_size=1 for inference
   #   During training: batch_size controls how many random patches to load