Merge pull request #36 from PolarBean/development

Development

Former-commit-id: 79a73a0

DeepSlice/coord_post_processing/angle_methods.py

@@ -85,6 +85,9 @@ def get_mean_angle(DV_list, ML_list, method, depths=None, species=None):
             weighted_accuracy = plane_alignment.make_gaussian_weights(max)
         else:
             weighted_accuracy = [1.0] * len(df_center)
+        df_center = np.array(df_center)
+        df_center[df_center < min] = min
+        df_center[df_center > max] = max-1
         weighted_accuracy = [weighted_accuracy[int(y)] for y in df_center]
         print(weighted_accuracy)
         DV_angle = np.average(DV_list, weights=weighted_accuracy)
@@ -111,6 +114,7 @@ def propagate_angles(df, method, species):
     DV_angle, ML_angle = get_mean_angle(
         DV_angle_list, ML_angle_list, method, depths, species
     )
+    print(f"DV angle: {DV_angle}\nML angle: {ML_angle}")
     # adjust the angles for each section in the dataset
     df = set_angles(df, DV_angle, ML_angle)
     return df

DeepSlice/coord_post_processing/plane_alignment_functions/plane_alignment.py

@@ -99,6 +99,8 @@ def rotation_around_axis(axis, angle):
 def make_gaussian_weights(size):
     x = np.linspace(-np.pi, np.pi, size)
     weights = np.exp(-(x ** 2) / 2) / np.sqrt(2 * np.pi)
+    weights[weights>size-1] = size-1
+    weights[weights<0] = 0
     return weights
 
 

DeepSlice/main.py

@@ -24,10 +24,12 @@ def __init__(self, species):
 
     def predict(
         self,
-        image_directory: str,
+        image_directory: str = None,
         ensemble: bool = None,
         section_numbers: bool = True,
         legacy_section_numbers=False,
+        image_list = None,
+        use_secondary_model = False,
     ):
         """predicts the atlas position for a folder full of histological brain sections
 
@@ -47,18 +49,40 @@ def predict(
         if ensemble == None:
             ensemble = self.config["ensemble_status"][self.species]
             ensemble = eval(ensemble)
-
-        image_generator, width, height = neural_network.load_images(image_directory)
+        if image_list:
+            image_generator, width, height = neural_network.load_images_from_list(
+                image_list
+            )
+            if image_directory:
+                print(
+                    "WARNING: image_directory is set but image_list is also set. image_directory will be ignored."
+                )
+        else:
+            image_generator, width, height = neural_network.load_images_from_path(
+                image_directory
+            )
         primary_weights = metadata_loader.get_data_path(self.config["weight_file_paths"][self.species]["primary"], self.metadata_path)
 
         secondary_weights = metadata_loader.get_data_path(self.config["weight_file_paths"][self.species]["secondary"], self.metadata_path)
 
         if secondary_weights == "None":
             print(f"ensemble is not available for {self.species}")
+            if use_secondary_model:
+                print("WARNING: use_secondary_model is set but no secondary model is available. use_secondary_model will be ignored.")
+                use_secondary_model = False
             ensemble = False
-        predictions = neural_network.predictions_util(
-            self.model, image_generator, primary_weights, secondary_weights, ensemble
-        )
+        if use_secondary_model and ensemble:
+            print("WARNING: use_secondary_model is set but ensemble is also set. use_secondary_model will be ignored.")
+            use_secondary_model = False
+        if use_secondary_model:
+            print("Using secondary model")
+            predictions = neural_network.predictions_util(
+                self.model, image_generator, secondary_weights,None, ensemble
+            )
+        else:
+            predictions = neural_network.predictions_util(
+                self.model, image_generator, primary_weights, secondary_weights, ensemble
+            )
         predictions["width"] = width
         predictions["height"] = height
         if section_numbers:

DeepSlice/neural_network/neural_network.py

@@ -60,10 +60,10 @@ def initialise_network(xception_weights: str, weights: str, species: str) -> Seq
 
 
 
-def load_images(image_path: str) -> np.ndarray:
+
+def load_images_from_path(image_path: str) -> np.ndarray:
     """
     Load the images from the given path
-
     :param image_path: The path to the images
     :type image_path: str
     :return: an Image generator for the found images
@@ -94,13 +94,46 @@ def load_images(image_path: str) -> np.ndarray:
             y_col=None,
             target_size=(299, 299),
             batch_size=1,
-
             colormode="rgb",
             shuffle=False,
             class_mode=None,
         )
     return image_generator, width, height
-
+
+def load_images_from_list(image_list: list) -> np.ndarray:
+    """
+    Load the images from the given list
+    :param image_list: The list of images
+    :type image_list: list
+    :return: an Image generator for the found images
+    :rtype: keras.preprocessing.image.ImageDataGenerator
+    """
+    valid_formats = [".jpg", ".jpeg", ".png"]
+    images = [i for i in image_list if os.path.splitext(i)[1].lower() in valid_formats]
+    sizes = [get_image_size(i) for i in images]
+    width = [i[0] for i in sizes]
+    height = [i[1] for i in sizes]
+    if len(images) == 0:
+        raise ValueError(
+            f"No images found in the directory, please ensure image files are one of the following formats: {', '.join(valid_formats)}"
+        )
+    image_df = pd.DataFrame({"Filenames": images})
+    with warnings.catch_warnings():
+        ##throws warning about samplewise_std_normalization conflicting with samplewise_center which we don't use.
+        warnings.simplefilter("ignore")
+        image_generator = ImageDataGenerator(
+            preprocessing_function=gray_scale, samplewise_std_normalization=True
+        ).flow_from_dataframe(
+            image_df,
+            x_col="Filenames",
+            y_col=None,
+            target_size=(299, 299),
+            batch_size=1,
+            colormode="rgb",
+            shuffle=False,
+            class_mode=None,
+        )
+    return image_generator, width, height
 
 def predictions_util(
     model: Sequential,