Finished UNet + launcher + started widget

docs/conf.py

@@ -21,12 +21,12 @@
 
 # -- Project information -----------------------------------------------------
 
-project = 'Protoplasts swelling analyzer'
+project   = 'Microglia Analyzer'
 copyright = '2024, Clément H. Benedetti'
-author = 'Clément H. BENEDETTI'
+author    = 'Clément H. BENEDETTI'
 
 # The full version, including alpha/beta/rc tags
-release = '1.0.0'
+release = '0.0.1'
 
 
 # -- General configuration ---------------------------------------------------
@@ -69,4 +69,4 @@
 
 html_favicon = 'icon.png'
 
-autosummary_generate = True
+autosummary_generate = True

src/microglia_analyzer/_tests/workflow.py

@@ -1,6 +1,6 @@
 import tifffile
 import os
-from microglia_analyzer.microglia_analyzer import MicrogliaAnalyzer
+from microglia_analyzer.ma_worker import MicrogliaAnalyzer
 from microglia_analyzer.tiles import tiler
 
 

src/microglia_analyzer/_widget.py

@@ -19,7 +19,7 @@
 import re
 
 from microglia_analyzer import TIFF_REGEX
-from microglia_analyzer.microglia_analyzer import MicrogliaAnalyzer
+from microglia_analyzer.ma_worker import MicrogliaAnalyzer
 
 
 _IMAGE_LAYER_NAME = "µ-Image"

src/microglia_analyzer/_widget_yolo_annotations.py

@@ -322,7 +322,7 @@ def clear_mask(self):
         Clears the mask layer.
         """
         if _MASKS_LAYER in self.viewer.layers:
-            self.viewer.layers[_MASKS_LAYER].data = np.zeros_like(self.viewer.layers[_IMAGE_LAYER].data)
+            self.viewer.layers[_MASKS_LAYER].data = np.zeros_like(self.viewer.layers[_MASKS_LAYER].data)
 
     def fill_current_label(self):
         if not _MASKS_LAYER in self.viewer.layers:

src/microglia_analyzer/dl/losses.py

@@ -0,0 +1,60 @@
+import tensorflow as tf
+
+def jaccard_loss(y_true, y_pred):
+    y_true = tf.cast(y_true, tf.float32)
+    y_pred = tf.cast(y_pred, tf.float32)
+    intersection = tf.reduce_sum(y_true * y_pred)
+    union = tf.reduce_sum(y_true) + tf.reduce_sum(y_pred) - intersection
+    return 1 - (intersection + 1) / (union + 1)
+
+def dice_loss(y_true, y_pred):
+    y_true = tf.cast(y_true, tf.float32)
+    y_pred = tf.cast(y_pred, tf.float32)
+    intersection = tf.reduce_sum(y_true * y_pred)
+    return 1 - (2. * intersection + 1) / (tf.reduce_sum(y_true) + tf.reduce_sum(y_pred) + 1)
+
+def bce_dice_loss(bce_coef=0.5):
+    def bcl(y_true, y_pred):
+        bce = tf.keras.losses.binary_crossentropy(y_true, y_pred)
+        dice = dice_loss(y_true, y_pred)
+        return bce_coef * bce + (1.0 - bce_coef) * dice
+    return bcl
+
+def focal_loss(gamma=2.0, alpha=5.75):
+    def focal_loss_fixed(y_true, y_pred):
+        y_true = tf.cast(y_true, tf.float32)
+        y_pred = tf.cast(y_pred, tf.float32)
+        alpha_t = y_true * alpha + (1 - y_true) * (1 - alpha)
+        p_t = y_true * y_pred + (1 - y_true) * (1 - y_pred)
+        fl = - alpha_t * (1 - p_t) ** gamma * tf.math.log(p_t + 1e-5)
+        return tf.reduce_mean(fl)
+    return focal_loss_fixed
+
+def tversky_loss(alpha=0.5):
+    beta = 1 - alpha
+    def loss(y_true, y_pred):
+        y_true = tf.cast(y_true, tf.float32)
+        y_pred = tf.cast(y_pred, tf.float32)
+        true_pos = tf.reduce_sum(y_true * y_pred)
+        false_neg = tf.reduce_sum(y_true * (1 - y_pred))
+        false_pos = tf.reduce_sum((1 - y_true) * y_pred)
+        return 1 - (true_pos + 1) / (true_pos + alpha * false_neg + beta * false_pos + 1)
+    return loss
+
+def skeleton_recall(y_true, y_pred):
+    intersection = tf.reduce_sum(y_true * y_pred)
+    recall = intersection / (tf.reduce_sum(y_true) + 1e-8)
+    return 1 - recall
+
+def dice_skeleton_loss(skeleton_coef=0.5, bce_coef=0.5):
+    bdl = bce_dice_loss(bce_coef)
+    def dsl(y_true, y_pred):
+        y_pred = tf.square(y_pred)
+        return (1.0 - skeleton_coef) * bdl(y_true, y_pred) + skeleton_coef * skeleton_recall(y_true, y_pred)
+    return dsl
+
+def skeleton_loss(y_true, y_pred):
+    inter = tf.reduce_sum(y_true * y_pred) / tf.reduce_sum(y_true)
+    mse_score = tf.reduce_mean(tf.square(y_true - y_pred))
+    mean_constraint = tf.abs(tf.reduce_mean(y_pred) - tf.reduce_mean(y_true))
+    return 1.0 - inter + mse_score + 0.1 * mean_constraint