Merge PR #184 from hci-unihd/fixflowGUI | fix #173, fix #183

Fix Flow in GUI & Napari: key, channel, LMC in GUI; multi-channel UNet layers in Napari

environment-dev.yaml

@@ -0,0 +1,22 @@
+name: plant-seg-dev
+channels:
+  - pytorch
+  - nvidia
+  - conda-forge
+dependencies:
+  - python
+  - h5py
+  - zarr
+  - requests
+  - pyyaml
+  - scikit-image
+  - tifffile
+  - vigra
+  - cudnn
+  - pytorch
+  - pytorch-cuda=12.1
+  - python-elf
+  - pyqt
+  - napari
+  - python-graphviz
+# then `pip install -e .`

examples/config_advanced.yaml

@@ -36,7 +36,7 @@ cnn_prediction:
   # channel to use if input image has shape CZYX or CYX, otherwise set to null; null is recommended if the previous steps has state True
   channel: Null
   # Trained model name, more info on available models and custom models in the README
-  model_name: 'RIKEN20230824v160'
+  model_name: 'generic_plant_nuclei_3D'
   # If a CUDA capable gpu is available and corrected setup use "cuda", if not you can use "cpu" for cpu only inference (slower)
   device: 'cuda'
   # how many subprocesses to use for data loading

plantseg/dataprocessing/functional/dataprocessing.py

@@ -77,7 +77,15 @@ def image_crop(image: np.array, crop_str: str) -> np.array:
     return image[slices]
 
 
-def fix_input_shape(data: np.array) -> np.array:
+def fix_input_shape(data: np.array, ndim=3) -> np.array:
+    assert ndim in [3, 4]
+    if ndim == 3:
+        return fix_input_shape_to_3D(data)
+    else:
+        return fix_input_shape_to_4D(data)
+
+
+def fix_input_shape_to_3D(data: np.array) -> np.array:
     """
     fix array ndim to be always 3
     """
@@ -94,6 +102,25 @@ def fix_input_shape(data: np.array) -> np.array:
         raise RuntimeError(f"Expected input data to be 2d, 3d or 4d, but got {data.ndim}d input")
 
 
+def fix_input_shape_to_4D(data: np.array) -> np.array:
+    """
+    Fix array ndim to be 4 and return it in (C x Z x Y x X) e.g. 2 x 1 x 512 x 512
+
+    Only used for multi-channel network output, e.g. 2-channel 3D pmaps.
+    """
+    if data.ndim == 4:
+        return data
+
+    elif data.ndim == 3:
+        return data.reshape(data.shape[0], 1, data.shape[1], data.shape[2])
+
+    # elif data.ndim == 2:
+    #     return data.reshape(1, 1, data.shape[0], data.shape[1])
+
+    else:
+        raise RuntimeError(f"Expected input data to be 3d or 4d, but got {data.ndim}d input")
+
+
 def normalize_01(data: np.array) -> np.array:
     """
     normalize a numpy array between 0 and 1