Docs + change norm in inference

Normalize each window instead of whole image for memory issues

Author: C-Achard

docs/res/guides/training_module_guide.rst

@@ -95,22 +95,16 @@ Function                  Reference
 ========================  ================================================================================================
 Dice loss                 `Dice Loss from MONAI`_ with ``sigmoid=true``
 Generalized Dice loss     `Generalized dice Loss from MONAI`_ with ``sigmoid=true``
-Dice-CE loss              `Dice-CE Loss from MONAI`_ with ``sigmoid=true``
+Dice-CE loss              `Dice-CrossEntropy Loss from MONAI`_ with ``sigmoid=true``
 Tversky loss              `Tversky Loss from MONAI`_ with ``sigmoid=true``
 ========================  ================================================================================================
 
-
-.. Binary cross-entropy      `Binary cross entropy (BCE) loss from PyTorch`_
-   BCE with logits           `BCE loss with logits from PyTorch`_
-
 .. _Dice Loss from MONAI: https://docs.monai.io/en/stable/losses.html#diceloss
 .. _Focal Loss from MONAI: https://docs.monai.io/en/stable/losses.html#focalloss
 .. _Dice-focal Loss from MONAI: https://docs.monai.io/en/stable/losses.html#dicefocalloss
 .. _Generalized dice Loss from MONAI: https://docs.monai.io/en/stable/losses.html#generalizeddiceloss
-.. _Dice-CE Loss from MONAI: https://docs.monai.io/en/stable/losses.html#diceceloss
+.. _Dice-CrossEntropy Loss from MONAI: https://docs.monai.io/en/stable/losses.html#diceceloss
 .. _Tversky Loss from MONAI: https://docs.monai.io/en/stable/losses.html#tverskyloss
-.. _Binary cross entropy (BCE) loss from PyTorch:
-.. _BCE loss with logits from PyTorch:
 
 Once you are ready, press the Start button to begin training. The module will automatically load your dataset,
 perform data augmentation if you chose to, select a CUDA device if one is present, and train the model.

napari_cellseg3d/code_models/worker_inference.py

@@ -33,7 +33,6 @@
     LogSignal,
     ONNXModelWrapper,
     QuantileNormalization,
-    QuantileNormalizationd,
     RemapTensor,
     Threshold,
     WeightsDownloader,
@@ -194,7 +193,7 @@ def load_folder(self):
                     EnsureChannelFirstd(keys=["image"]),
                     # Orientationd(keys=["image"], axcodes="PLI"),
                     # anisotropic_transform,
-                    QuantileNormalizationd(keys=["image"]),
+                    # QuantileNormalizationd(keys=["image"]),
                     EnsureTyped(keys=["image"]),
                 ]
             )
@@ -204,7 +203,7 @@ def load_folder(self):
                     LoadImaged(keys=["image"]),
                     # AddChanneld(keys=["image"]), #already done
                     EnsureChannelFirstd(keys=["image"]),
-                    QuantileNormalizationd(keys=["image"]),
+                    # QuantileNormalizationd(keys=["image"]),
                     # Orientationd(keys=["image"], axcodes="PLI"),
                     # anisotropic_transform,
                     SpatialPadd(keys=["image"], spatial_size=pad),
@@ -248,7 +247,7 @@ def load_layer(self):
                     # anisotropic_transform,
                     AddChannel(),
                     # SpatialPad(spatial_size=pad),
-                    QuantileNormalization(),
+                    # QuantileNormalization(),
                     AddChannel(),
                     EnsureType(),
                 ],
@@ -263,7 +262,7 @@ def load_layer(self):
                     ToTensor(),
                     # anisotropic_transform,
                     AddChannel(),
-                    QuantileNormalization(),
+                    # QuantileNormalization(),
                     SpatialPad(spatial_size=pad),
                     AddChannel(),
                     EnsureType(),
@@ -301,10 +300,12 @@ def model_output(
             # logger.debug(f"model : {model}")
             logger.debug(f"inputs shape : {inputs.shape}")
             logger.debug(f"inputs type : {inputs.dtype}")
+            normalizazion = QuantileNormalization()
             try:
                 # outputs = model(inputs)
 
                 def model_output_wrapper(inputs):
+                    inputs = normalizazion(inputs)
                     result = model(inputs)
                     return post_process_transforms(result)