revert

dreamify/classicdream.py

@@ -40,7 +40,7 @@ def classicdream(
 ):
     output_path = Path(output_path)
 
-    ft_ext = FeatureExtractor(model_name, dream_style="classic", layer_settings=None)
+    ft_ext = FeatureExtractor(model_name, "classic", layer_settings, channel_settings)
     get_tiled_gradients = TiledGradients(ft_ext.feature_extractor)
 
     img = get_image(image_path)

dreamify/deepdream.py

@@ -3,16 +3,12 @@
 import IPython.display as display
 import tensorflow as tf
 
-from dreamify.decorators import optional_global_determinism, validate_dream_params
+from dreamify.decorators import (optional_global_determinism,
+                                 validate_dream_params)
 from dreamify.lib import Config, FeatureExtractor
 from dreamify.lib.models import TiledGradients
-from dreamify.utils import (
-    deprocess_image,
-    get_image,
-    preprocess_image,
-    save_output,
-    show,
-)
+from dreamify.utils import (deprocess_image, get_image, preprocess_image,
+                            save_output, show)
 
 
 @optional_global_determinism
@@ -37,7 +33,7 @@ def deepdream(
 ):
     output_path = Path(output_path)
 
-    ft_ext = FeatureExtractor(model_name, dream_style="deep", layer_settings=None)
+    ft_ext = FeatureExtractor(model_name, "deep", layer_settings, channel_settings)
     get_tiled_gradients = TiledGradients(ft_ext.feature_extractor)
 
     img = get_image(image_path)

dreamify/dream.py

@@ -45,9 +45,7 @@ def dream(
 ):
     output_path = Path(output_path)
 
-    ft_ext = FeatureExtractor(
-        model_name=model_name, dream_style="shallow", layer_settings=None
-    )
+    ft_ext = FeatureExtractor(model_name, "shallow", layer_settings, channel_settings)
 
     original_img = get_image(image_path)
     original_img = preprocess_image(original_img, model_name)

dreamify/lib/feature_extractor.py

@@ -5,10 +5,11 @@
 
 
 class FeatureExtractor:
-    def __init__(self, model_name, dream_style, layer_settings):
+    def __init__(self, model_name, dream_style, layer_settings, channel_settings):
         self.model, self.layer_settings = choose_base_model(
             model_name, dream_style, layer_settings
         )
+        self.channel_settings = channel_settings
 
         # outputs is either:
         # - A list of layer outputs (for DeepDream)

dreamify/lib/models/base/base_models.py

@@ -1,10 +1,10 @@
 import random
 
 from dreamify.lib.models.base.constants import MODEL_MAP, ModelType
-from dreamify.lib.models.base.layer_settings import ModelLayerNames
+from dreamify.lib.models.base.layer_names import ModelLayerNames
 
 
-def get_layer_settings(model_name_enum: ModelType, dream_style="deep"):
+def get_layer_names(model_name_enum: ModelType, dream_style):
     model_settings = ModelLayerNames[model_name_enum.name]
     if dream_style == "deep":
         return model_settings.deep
@@ -27,7 +27,7 @@ def choose_base_model(model_name: str, dream_style="deep", layer_settings=None):
     model_fn = MODEL_MAP[model_name_enum]
     base_model = model_fn(weights="imagenet", include_top=False)
 
-    layer_names = get_layer_settings(model_name_enum, dream_style)
+    layer_names = get_layer_names(model_name_enum, dream_style)
 
     if layer_settings is not None:
         if layers_settings == "all":

dreamify/lib/models/base/layer_names.py

@@ -10,18 +10,18 @@ class ModelLayerNames(Enum):
     INCEPTION_V3 = (
         ["mixed3", "mixed5"],
         generate_shallow_settings(["mixed4", "mixed5", "mixed6", "mixed7"]),
-        [f"mixed_{i}" for i in range(1, 12)],
+        [f"mixed_{i}" for i in range(1, 12)]
     )
     VGG19 = (
         ["block5_conv3", "block5_conv2"],
         generate_shallow_settings(
             ["block5_conv3", "block5_conv2", "block4_conv3", "block3_conv3"]
         ),
-        [f"block1_conv{i}" for i in range(1, 3)]
-        + [f"block2_conv{i}" for i in range(1, 3)]
-        + [f"block3_conv{i}" for i in range(1, 6)]
-        + [f"block4_conv{i}" for i in range(1, 6)]
-        + [f"block5_conv{i}" for i in range(1, 6)],
+        [f"block1_conv{i}" for i in range(1, 3)] + 
+        [f"block2_conv{i}" for i in range(1, 3)] + 
+        [f"block3_conv{i}" for i in range(1, 6)] + 
+        [f"block4_conv{i}" for i in range(1, 6)] + 
+        [f"block5_conv{i}" for i in range(1, 6)]
     )
     DENSENET121 = (
         ["conv5_block16_1_conv", "conv4_block24_1_conv"],
@@ -33,7 +33,7 @@ class ModelLayerNames(Enum):
                 "conv2_block12_1_conv",
             ]
         ),
-        [],
+        []
     )
     EFFICIENTNET_V2L = (
         ["block7a_project_bn", "block6a_expand_activation"],
@@ -45,12 +45,12 @@ class ModelLayerNames(Enum):
                 "block4a_expand_activation",
             ]
         ),
-        [],
+        []
     )
     INCEPTION_RESNET_V2 = (
         ["mixed_7a", "mixed_6a"],
         generate_shallow_settings(["mixed_7a", "mixed_6a", "mixed_5a", "mixed_4a"]),
-        [],
+        []
     )
     RESNET152V2 = (
         ["conv5_block3_out", "conv4_block6_out"],
@@ -62,7 +62,7 @@ class ModelLayerNames(Enum):
                 "conv2_block3_out",
             ]
         ),
-        [],
+        []
     )
     XCEPTION = (
         ["block14_sepconv2_act", "block13_sepconv2_act"],
@@ -74,7 +74,7 @@ class ModelLayerNames(Enum):
                 "block11_sepconv2_act",
             ]
         ),
-        [],
+        []
     )
     CONVNEXT_XL = (
         ["stage4_block2_depthwise_conv", "stage3_block2_depthwise_conv"],
@@ -86,7 +86,7 @@ class ModelLayerNames(Enum):
                 "stage1_block2_depthwise_conv",
             ]
         ),
-        [],
+        []
     )
     MOBILENET_V2 = (
         ["block_16_depthwise", "block_13_depthwise"],
@@ -98,7 +98,7 @@ class ModelLayerNames(Enum):
                 "block_5_depthwise",
             ]
         ),
-        [],
+        []
     )
 
     @property
@@ -115,3 +115,4 @@ def shallow(self):
     def classic(self):
         """Returns the convolutional blocks of the model"""
         return self.value[2]
+

dreamify/lib/models/tiled_gradients.py

@@ -1,3 +1,5 @@
+from random import choice
+
 import tensorflow as tf
 
 
@@ -54,10 +56,26 @@ def calc_loss(img, feature_extractor):
         """Calculate the DeepDream loss by maximizing activations."""
         img_batch = tf.expand_dims(img, axis=0)
         layer_activations = feature_extractor(img_batch)
+
+        if feature_extractor.channel_settings == "random":
+            # Select a random channel for each layer
+            for layer_name in layer_activations:
+                channels = layer_activations[layer_name].shape[-1]
+                random_channel = choice(channels)
+                layer_activations[layer_name] = layer_activations[layer_name][
+                    ..., random_channel
+                ]
+
         if len(layer_activations) == 1:
             layer_activations = [layer_activations]
 
-        return tf.reduce_sum([tf.math.reduce_mean(act) for act in layer_activations])
+        # Ensure activations are iterable
+        if isinstance(layer_activations, dict):
+            activations_list = list(layer_activations.values())
+        else:
+            activations_list = [layer_activations]
+
+        return tf.reduce_sum([tf.reduce_mean(act) for act in activations_list])
 
     @staticmethod
     def random_roll(img, maxroll):

dreamify/utils/dream_utils/utils.py

@@ -1,14 +1,26 @@
+from random import choice
+
 import tensorflow as tf
 from tqdm import trange
 
 
 def compute_loss(input_image, config):
     features = config.feature_extractor(input_image)
     loss = tf.zeros(shape=())
+
     for name in features.keys():
         coeff = config.layer_settings[name]
         activation = features[name]
+
+        # Apply channel selection logic
+        if config.feature_extractor.channel_settings == "random":
+            channels = activation.shape[-1]
+            random_channel = choice(channels)
+            activation = activation[..., random_channel]
+
+        # Compute loss on selected activations
         loss += coeff * tf.reduce_mean(tf.square(activation[:, 2:-2, 2:-2, :]))
+
     return loss
 
 

pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "dreamify"
-version = "1.0.0"
+version = "1.1.3"
 description = ""
 authors = [
     {name = "kenny",email = "97374837+kankenny@users.noreply.github.com"}

test

@@ -0,0 +1 @@
+test