keras-hub-nightly 0.16.1.dev202409250340__py3-none-any.whl → 0.16.1.dev202409260340__py3-none-any.whl

keras_hub/src/models/densenet/densenet_image_classifier.py

@@ -15,6 +15,9 @@ import keras
 
 from keras_hub.src.api_export import keras_hub_export
 from keras_hub.src.models.densenet.densenet_backbone import DenseNetBackbone
+from keras_hub.src.models.densenet.densenet_image_classifier_preprocessor import (
+    DenseNetImageClassifierPreprocessor,
+)
 from keras_hub.src.models.image_classifier import ImageClassifier
 
 
@@ -32,7 +35,13 @@ class DenseNetImageClassifier(ImageClassifier):
         num_classes: int. The number of classes to predict.
         activation: `None`, str or callable. The activation function to use on
             the `Dense` layer. Set `activation=None` to return the output
-            logits. Defaults to `"softmax"`.
+            logits. Defaults to `None`.
+        pooling: A pooling layer to use before the final classification layer,
+            must be one of "avg" or "max". Use "avg" for
+            `GlobalAveragePooling2D` and "max" for "GlobalMaxPooling2D.
+        preprocessor: A `keras_hub.models.DenseNetImageClassifierPreprocessor`
+            or `None`. If `None`, this model will not apply preprocessing, and
+            inputs should be preprocessed before calling the model.
 
     Examples:
 
@@ -86,18 +95,29 @@ class DenseNetImageClassifier(ImageClassifier):
     """
 
     backbone_cls = DenseNetBackbone
+    preprocessor_cls = DenseNetImageClassifierPreprocessor
 
     def __init__(
         self,
         backbone,
         num_classes,
-        activation="softmax",
-        preprocessor=None,  # adding this dummy arg for saved model test
-        # TODO: once preprocessor flow is figured out, this needs to be updated
+        activation=None,
+        pooling="avg",
+        preprocessor=None,
         **kwargs,
     ):
         # === Layers ===
         self.backbone = backbone
+        self.preprocessor = preprocessor
+        if pooling == "avg":
+            self.pooler = keras.layers.GlobalAveragePooling2D()
+        elif pooling == "max":
+            self.pooler = keras.layers.GlobalMaxPooling2D()
+        else:
+            raise ValueError(
+                "Unknown `pooling` type. Polling should be either `'avg'` or "
+                f"`'max'`. Received: pooling={pooling}."
+            )
         self.output_dense = keras.layers.Dense(
             num_classes,
             activation=activation,
@@ -107,6 +127,7 @@ class DenseNetImageClassifier(ImageClassifier):
         # === Functional Model ===
         inputs = self.backbone.input
         x = self.backbone(inputs)
+        x = self.pooler(x)
         outputs = self.output_dense(x)
         super().__init__(
             inputs=inputs,
@@ -117,6 +138,7 @@ class DenseNetImageClassifier(ImageClassifier):
         # === Config ===
         self.num_classes = num_classes
         self.activation = activation
+        self.pooling = pooling
 
     def get_config(self):
         # Backbone serialized in `super`
@@ -125,6 +147,7 @@ class DenseNetImageClassifier(ImageClassifier):
             {
                 "num_classes": self.num_classes,
                 "activation": self.activation,
+                "pooling": self.pooling,
             }
         )
         return config

keras_hub/src/models/densenet/densenet_image_classifier_preprocessor.py

@@ -0,0 +1,27 @@
+# Copyright 2024 The KerasHub Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from keras_hub.src.api_export import keras_hub_export
+from keras_hub.src.models.densenet.densenet_backbone import DenseNetBackbone
+from keras_hub.src.models.densenet.densenet_image_converter import (
+    DenseNetImageConverter,
+)
+from keras_hub.src.models.image_classifier_preprocessor import (
+    ImageClassifierPreprocessor,
+)
+
+
+@keras_hub_export("keras_hub.models.DenseNetImageClassifierPreprocessor")
+class DenseNetImageClassifierPreprocessor(ImageClassifierPreprocessor):
+    backbone_cls = DenseNetBackbone
+    image_converter_cls = DenseNetImageConverter

keras_hub/src/models/densenet/densenet_image_converter.py

@@ -0,0 +1,23 @@
+# Copyright 2024 The KerasHub Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from keras_hub.src.api_export import keras_hub_export
+from keras_hub.src.layers.preprocessing.resizing_image_converter import (
+    ResizingImageConverter,
+)
+from keras_hub.src.models.densenet.densenet_backbone import DenseNetBackbone
+
+
+@keras_hub_export("keras_hub.layers.DenseNetImageConverter")
+class DenseNetImageConverter(ResizingImageConverter):
+    backbone_cls = DenseNetBackbone

keras_hub/src/models/densenet/densenet_presets.py

@@ -0,0 +1,56 @@
+# Copyright 2024 The KerasHub Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""DenseNet preset configurations."""
+
+backbone_presets = {
+    "densenet_121_imagenet": {
+        "metadata": {
+            "description": (
+                "121-layer DenseNet model pre-trained on the ImageNet 1k dataset "
+                "at a 224x224 resolution."
+            ),
+            "params": 7037504,
+            "official_name": "DenseNet",
+            "path": "densenet",
+            "model_card": "https://arxiv.org/abs/1608.06993",
+        },
+        "kaggle_handle": "kaggle://kerashub/densenet/keras/densenet_121_imagenet",
+    },
+    "densenet_169_imagenet": {
+        "metadata": {
+            "description": (
+                "169-layer DenseNet model pre-trained on the ImageNet 1k dataset "
+                "at a 224x224 resolution."
+            ),
+            "params": 12642880,
+            "official_name": "DenseNet",
+            "path": "densenet",
+            "model_card": "https://arxiv.org/abs/1608.06993",
+        },
+        "kaggle_handle": "kaggle://kerashub/densenet/keras/densenet_169_imagenet",
+    },
+    "densenet_201_imagenet": {
+        "metadata": {
+            "description": (
+                "201-layer DenseNet model pre-trained on the ImageNet 1k dataset "
+                "at a 224x224 resolution."
+            ),
+            "params": 18321984,
+            "official_name": "DenseNet",
+            "path": "densenet",
+            "model_card": "https://arxiv.org/abs/1608.06993",
+        },
+        "kaggle_handle": "kaggle://kerashub/densenet/keras/densenet_201_imagenet",
+    },
+}

keras_hub/src/models/stable_diffusion_3/__init__.py

@@ -0,0 +1,13 @@
+# Copyright 2024 The KerasHub Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.

keras_hub/src/models/stable_diffusion_3/flow_match_euler_discrete_scheduler.py

@@ -0,0 +1,93 @@
+# Copyright 2024 The KerasHub Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+from keras import layers
+from keras import ops
+
+
+class FlowMatchEulerDiscreteScheduler(layers.Layer):
+    """Flow-matching sampling euler scheduler.
+
+    This layer is used to compute the discrete sigmas for the diffusion chain.
+    Typically, the sigma refers to the amount of noise added during the
+    diffusion process.
+
+    Args:
+        num_train_timesteps: int. The number of diffusion steps to train the
+            model.
+        shift: float. The shift value for the timestep schedule.
+        **kwargs: other keyword arguments passed to `keras.layers.Layer`,
+            including `name`, `dtype` etc.
+
+    Call arguments:
+        inputs: The current step of the diffusion process.
+        num_steps: The total number of steps in the diffusion process.
+
+    References:
+    - [Common Diffusion Noise Schedules and Sample Steps are Flawed](
+    https://arxiv.org/abs/2305.08891).
+    - [Scaling Rectified Flow Transformers for High-Resolution Image Synthesis](
+    https://arxiv.org/abs/2403.03206).
+    """
+
+    def __init__(self, num_train_timesteps=1000, shift=1.0, **kwargs):
+        super().__init__(**kwargs)
+        self.num_train_timesteps = int(num_train_timesteps)
+        self.shift = float(shift)
+
+        timesteps = ops.linspace(
+            1, num_train_timesteps, num_train_timesteps, dtype="float32"
+        )
+        timesteps = ops.flip(timesteps, axis=0)
+        sigmas = self._timestep_to_sigma(timesteps)
+
+        self.timesteps = ops.multiply(sigmas, num_train_timesteps)
+        self.sigma_min = sigmas[-1]
+        self.sigma_max = sigmas[0]
+
+    def _sigma_to_timestep(self, sigma):
+        return sigma * self.num_train_timesteps
+
+    def _timestep_to_sigma(self, timestep):
+        sigma = ops.divide(timestep, self.num_train_timesteps)
+        if self.shift != 1.0:
+            sigma = ops.divide(
+                ops.multiply(self.shift, sigma),
+                ops.add(1, ops.multiply(self.shift - 1.0, sigma)),
+            )
+        return sigma
+
+    def call(self, inputs, num_steps):
+        start = self._sigma_to_timestep(self.sigma_max)
+        end = self._sigma_to_timestep(self.sigma_min)
+        step_size = ops.divide(
+            ops.subtract(end, start), ops.subtract(num_steps, 1)
+        )
+        timestep = ops.add(start, ops.multiply(inputs, step_size))
+        sigma = ops.maximum(self._timestep_to_sigma(timestep), 0.0)
+        timestep = self._sigma_to_timestep(sigma)
+        return sigma, timestep
+
+    def get_config(self):
+        config = super().get_config()
+        config.update(
+            {
+                "num_train_timesteps": self.num_train_timesteps,
+                "shift": self.shift,
+            }
+        )
+        return config
+
+    def compute_output_shape(self):
+        # Returns a tuple of (sigma, timestep).
+        return (None,), (None,)