keras-hub-nightly 0.16.0.dev202409201942__py3-none-any.whl → 0.16.1.dev202409210335__py3-none-any.whl

keras_hub/src/api_export.py

@@ -52,8 +52,8 @@ else:
 
     class keras_hub_export:
         def __init__(self, path):
-            pass
+            self.path = path
 
         def __call__(self, symbol):
-            maybe_register_serializable(symbol)
+            maybe_register_serializable(self.path, symbol)
             return symbol

keras_hub/src/layers/preprocessing/resizing_image_converter.py

@@ -12,9 +12,11 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 import keras
+from keras import ops
 
 from keras_hub.src.api_export import keras_hub_export
 from keras_hub.src.layers.preprocessing.image_converter import ImageConverter
+from keras_hub.src.utils.keras_utils import standardize_data_format
 from keras_hub.src.utils.tensor_utils import preprocessing_function
 
 
@@ -23,13 +25,23 @@ class ResizingImageConverter(ImageConverter):
     """An `ImageConverter` that simply resizes the input image.
 
     The `ResizingImageConverter` is a subclass of `ImageConverter` for models
-    that simply need to resize image tensors before using them for modeling.
-    The layer will take as input a raw image tensor (batched or unbatched) in the
-    channels last or channels first format, and output a resize tensor.
+    that need to resize (and optionally rescale) image tensors before using them
+    for modeling. The layer will take as input a raw image tensor (batched or
+    unbatched) in the channels last or channels first format, and output a
+    resize tensor.
 
     Args:
-        height: Integer, the height of the output shape.
-        width: Integer, the width of the output shape.
+        height: int, the height of the output shape.
+        width: int, the width of the output shape.
+        scale: float or `None`. If set, the image we be rescaled with a
+            `keras.layers.Rescaling` layer, multiplying the image by this
+            scale.
+        mean: tuples of floats per channel or `None`. If set, the image will be
+            normalized per channel by subtracting mean.
+            If set, also set `variance`.
+        variance: tuples of floats per channel or `None`. If set, the image will
+            be normalized per channel by dividing by `sqrt(variance)`.
+            If set, also set `mean`.
         crop_to_aspect_ratio: If `True`, resize the images without aspect
             ratio distortion. When the original aspect ratio differs
             from the target aspect ratio, the output image will be
@@ -64,6 +76,9 @@ class ResizingImageConverter(ImageConverter):
         self,
         height,
         width,
+        scale=None,
+        mean=None,
+        variance=None,
         crop_to_aspect_ratio=True,
         interpolation="bilinear",
         data_format=None,
@@ -78,7 +93,26 @@ class ResizingImageConverter(ImageConverter):
             crop_to_aspect_ratio=crop_to_aspect_ratio,
             interpolation=interpolation,
             data_format=data_format,
+            dtype=self.dtype_policy,
+            name="resizing",
         )
+        if scale is not None:
+            self.rescaling = keras.layers.Rescaling(
+                scale=scale,
+                dtype=self.dtype_policy,
+                name="rescaling",
+            )
+        else:
+            self.rescaling = None
+        if (mean is not None) != (variance is not None):
+            raise ValueError(
+                "Both `mean` and `variance` should be set or `None`. Received "
+                f"`mean={mean}`, `variance={variance}`."
+            )
+        self.scale = scale
+        self.mean = mean
+        self.variance = variance
+        self.data_format = standardize_data_format(data_format)
 
     def image_size(self):
         """Returns the preprocessed size of a single image."""
@@ -86,7 +120,20 @@ class ResizingImageConverter(ImageConverter):
 
     @preprocessing_function
     def call(self, inputs):
-        return self.resizing(inputs)
+        x = self.resizing(inputs)
+        if self.rescaling:
+            x = self.rescaling(x)
+        if self.mean is not None:
+            # Avoid `layers.Normalization` so this works batched and unbatched.
+            channels_first = self.data_format == "channels_first"
+            if len(ops.shape(inputs)) == 3:
+                broadcast_dims = (1, 2) if channels_first else (0, 1)
+            else:
+                broadcast_dims = (0, 2, 3) if channels_first else (0, 1, 2)
+            mean = ops.expand_dims(ops.array(self.mean), broadcast_dims)
+            std = ops.expand_dims(ops.sqrt(self.variance), broadcast_dims)
+            x = (x - mean) / std
+        return x
 
     def get_config(self):
         config = super().get_config()
@@ -96,6 +143,9 @@ class ResizingImageConverter(ImageConverter):
                 "width": self.resizing.width,
                 "interpolation": self.resizing.interpolation,
                 "crop_to_aspect_ratio": self.resizing.crop_to_aspect_ratio,
+                "scale": self.scale,
+                "mean": self.mean,
+                "variance": self.variance,
             }
         )
         return config

keras_hub/src/models/csp_darknet/csp_darknet_backbone.py

@@ -31,9 +31,6 @@ class CSPDarkNetBackbone(FeaturePyramidBackbone):
             level in the model.
         stackwise_depth: A list of ints, the depth for each dark level in the
             model.
-        include_rescaling: boolean. If `True`, rescale the input using
-            `Rescaling(1 / 255.0)` layer. If `False`, do nothing. Defaults to
-            `True`.
         block_type: str. One of `"basic_block"` or `"depthwise_block"`.
             Use `"depthwise_block"` for depthwise conv block
             `"basic_block"` for basic conv block.
@@ -55,7 +52,6 @@ class CSPDarkNetBackbone(FeaturePyramidBackbone):
     model = keras_hub.models.CSPDarkNetBackbone(
         stackwise_num_filters=[128, 256, 512, 1024],
         stackwise_depth=[3, 9, 9, 3],
-        include_rescaling=False,
     )
     model(input_data)
     ```
@@ -65,7 +61,6 @@ class CSPDarkNetBackbone(FeaturePyramidBackbone):
         self,
         stackwise_num_filters,
         stackwise_depth,
-        include_rescaling=True,
         block_type="basic_block",
         image_shape=(None, None, 3),
         **kwargs,
@@ -82,10 +77,7 @@ class CSPDarkNetBackbone(FeaturePyramidBackbone):
         base_channels = stackwise_num_filters[0] // 2
 
         image_input = layers.Input(shape=image_shape)
-        x = image_input
-        if include_rescaling:
-            x = layers.Rescaling(scale=1 / 255.0)(x)
-
+        x = image_input  # Intermediate result.
         x = apply_focus(channel_axis, name="stem_focus")(x)
         x = apply_darknet_conv_block(
             base_channels,
@@ -130,7 +122,6 @@ class CSPDarkNetBackbone(FeaturePyramidBackbone):
         # === Config ===
         self.stackwise_num_filters = stackwise_num_filters
         self.stackwise_depth = stackwise_depth
-        self.include_rescaling = include_rescaling
         self.block_type = block_type
         self.image_shape = image_shape
         self.pyramid_outputs = pyramid_outputs
@@ -141,7 +132,6 @@ class CSPDarkNetBackbone(FeaturePyramidBackbone):
             {
                 "stackwise_num_filters": self.stackwise_num_filters,
                 "stackwise_depth": self.stackwise_depth,
-                "include_rescaling": self.include_rescaling,
                 "block_type": self.block_type,
                 "image_shape": self.image_shape,
             }

keras_hub/src/models/csp_darknet/csp_darknet_image_classifier.py

@@ -76,7 +76,6 @@ class CSPDarkNetImageClassifier(ImageClassifier):
     backbone = keras_hub.models.CSPDarkNetBackbone(
         stackwise_num_filters=[128, 256, 512, 1024],
         stackwise_depth=[3, 9, 9, 3],
-        include_rescaling=False,
         block_type="basic_block",
         image_shape = (224, 224, 3),
     )

keras_hub/src/models/densenet/densenet_backbone.py

@@ -31,9 +31,6 @@ class DenseNetBackbone(FeaturePyramidBackbone):
     Args:
         stackwise_num_repeats: list of ints, number of repeated convolutional
             blocks per dense block.
-        include_rescaling: bool, whether to rescale the inputs. If set
-            to `True`, inputs will be passed through a `Rescaling(1/255.0)`
-            layer. Defaults to `True`.
         image_shape: optional shape tuple, defaults to (None, None, 3).
         compression_ratio: float, compression rate at transition layers,
             defaults to 0.5.
@@ -51,7 +48,6 @@ class DenseNetBackbone(FeaturePyramidBackbone):
     # Randomly initialized backbone with a custom config
     model = keras_hub.models.DenseNetBackbone(
         stackwise_num_repeats=[6, 12, 24, 16],
-        include_rescaling=False,
     )
     model(input_data)
     ```
@@ -60,7 +56,6 @@ class DenseNetBackbone(FeaturePyramidBackbone):
     def __init__(
         self,
         stackwise_num_repeats,
-        include_rescaling=True,
         image_shape=(None, None, 3),
         compression_ratio=0.5,
         growth_rate=32,
@@ -71,10 +66,7 @@ class DenseNetBackbone(FeaturePyramidBackbone):
         channel_axis = -1 if data_format == "channels_last" else 1
         image_input = keras.layers.Input(shape=image_shape)
 
-        x = image_input
-        if include_rescaling:
-            x = keras.layers.Rescaling(1 / 255.0)(x)
-
+        x = image_input  # Intermediate result.
         x = keras.layers.Conv2D(
             64,
             7,
@@ -124,7 +116,6 @@ class DenseNetBackbone(FeaturePyramidBackbone):
 
         # === Config ===
         self.stackwise_num_repeats = stackwise_num_repeats
-        self.include_rescaling = include_rescaling
         self.compression_ratio = compression_ratio
         self.growth_rate = growth_rate
         self.image_shape = image_shape
@@ -135,7 +126,6 @@ class DenseNetBackbone(FeaturePyramidBackbone):
         config.update(
             {
                 "stackwise_num_repeats": self.stackwise_num_repeats,
-                "include_rescaling": self.include_rescaling,
                 "compression_ratio": self.compression_ratio,
                 "growth_rate": self.growth_rate,
                 "image_shape": self.image_shape,

keras_hub/src/models/densenet/densenet_image_classifier.py

@@ -74,7 +74,6 @@ class DenseNetImageClassifier(ImageClassifier):
     backbone = keras_hub.models.DenseNetBackbone(
         stackwise_num_filters=[128, 256, 512, 1024],
         stackwise_depth=[3, 9, 9, 3],
-        include_rescaling=False,
         block_type="basic_block",
         image_shape = (224, 224, 3),
     )

keras_hub/src/models/efficientnet/efficientnet_backbone.py

@@ -67,8 +67,6 @@ class EfficientNetBackbone(FeaturePyramidBackbone):
             MBConvBlock, but instead of using a depthwise convolution and a 1x1
             output convolution blocks fused blocks use a single 3x3 convolution
             block.
-        include_rescaling: bool, whether to rescale the inputs. If set to
-            True, inputs will be passed through a `Rescaling(1/255.0)` layer.
         min_depth: integer, minimum number of filters. Can be None and ignored
             if use_depth_divisor_as_min_depth is set to True.
         include_initial_padding: bool, whether to include initial zero padding
@@ -96,7 +94,6 @@ class EfficientNetBackbone(FeaturePyramidBackbone):
         stackwise_block_types=[["fused"] * 3 + ["unfused"] * 3],
         width_coefficient=1.0,
         depth_coefficient=1.0,
-        include_rescaling=False,
     )
     images = np.ones((1, 256, 256, 3))
     outputs = efficientnet.predict(images)
@@ -116,7 +113,6 @@ class EfficientNetBackbone(FeaturePyramidBackbone):
         stackwise_squeeze_and_excite_ratios,
         stackwise_strides,
         stackwise_block_types,
-        include_rescaling=True,
         dropout=0.2,
         depth_divisor=8,
         min_depth=8,
@@ -129,14 +125,9 @@ class EfficientNetBackbone(FeaturePyramidBackbone):
         batch_norm_momentum=0.9,
         **kwargs,
     ):
-        img_input = keras.layers.Input(shape=input_shape)
-
-        x = img_input
-
-        if include_rescaling:
-            # Use common rescaling strategy across keras
-            x = keras.layers.Rescaling(scale=1.0 / 255.0)(x)
+        image_input = keras.layers.Input(shape=input_shape)
 
+        x = image_input  # Intermediate result.
         if include_initial_padding:
             x = keras.layers.ZeroPadding2D(
                 padding=self._correct_pad_downsample(x, 3),
@@ -282,10 +273,9 @@ class EfficientNetBackbone(FeaturePyramidBackbone):
         curr_pyramid_level += 1
 
         # Create model.
-        super().__init__(inputs=img_input, outputs=x, **kwargs)
+        super().__init__(inputs=image_input, outputs=x, **kwargs)
 
         # === Config ===
-        self.include_rescaling = include_rescaling
         self.width_coefficient = width_coefficient
         self.depth_coefficient = depth_coefficient
         self.dropout = dropout
@@ -313,7 +303,6 @@ class EfficientNetBackbone(FeaturePyramidBackbone):
         config = super().get_config()
         config.update(
             {
-                "include_rescaling": self.include_rescaling,
                 "width_coefficient": self.width_coefficient,
                 "depth_coefficient": self.depth_coefficient,
                 "dropout": self.dropout,

keras_hub/src/models/mix_transformer/mix_transformer_backbone.py

@@ -36,7 +36,6 @@ class MiTBackbone(FeaturePyramidBackbone):
         end_value,
         patch_sizes,
         strides,
-        include_rescaling=True,
         image_shape=(None, None, 3),
         hidden_dims=None,
         **kwargs,
@@ -60,9 +59,6 @@ class MiTBackbone(FeaturePyramidBackbone):
                 value projections. If set to > 1, a `Conv2D` layer is used to
                 reduce the length of the sequence.
             end_value: The end value of the sequence.
-            include_rescaling: bool, whether to rescale the inputs. If set
-                to `True`, inputs will be passed through a `Rescaling(1/255.0)`
-                layer. Defaults to `True`.
             image_shape: optional shape tuple, defaults to (None, None, 3).
             hidden_dims: the embedding dims per hierarchical layer, used as
                 the levels of the feature pyramid.
@@ -123,11 +119,7 @@ class MiTBackbone(FeaturePyramidBackbone):
 
         # === Functional Model ===
         image_input = keras.layers.Input(shape=image_shape)
-        x = image_input
-
-        if include_rescaling:
-            x = keras.layers.Rescaling(scale=1 / 255)(x)
-
+        x = image_input  # Intermediate result.
         pyramid_outputs = {}
         for i in range(num_layers):
             # Compute new height/width after the `proj`
@@ -151,7 +143,6 @@ class MiTBackbone(FeaturePyramidBackbone):
 
         # === Config ===
         self.depths = depths
-        self.include_rescaling = include_rescaling
         self.image_shape = image_shape
         self.hidden_dims = hidden_dims
         self.pyramid_outputs = pyramid_outputs
@@ -167,7 +158,6 @@ class MiTBackbone(FeaturePyramidBackbone):
         config.update(
             {
                 "depths": self.depths,
-                "include_rescaling": self.include_rescaling,
                 "hidden_dims": self.hidden_dims,
                 "image_shape": self.image_shape,
                 "num_layers": self.num_layers,

keras_hub/src/models/mix_transformer/mix_transformer_classifier.py

@@ -76,7 +76,6 @@ class MiTImageClassifier(ImageClassifier):
     backbone = keras_hub.models.MiTBackbone(
         stackwise_num_filters=[128, 256, 512, 1024],
         stackwise_depth=[3, 9, 9, 3],
-        include_rescaling=False,
         block_type="basic_block",
         image_shape = (224, 224, 3),
     )

keras_hub/src/models/mobilenet/mobilenet_backbone.py

@@ -54,9 +54,6 @@ class MobileNetBackbone(Backbone):
             model. 0 if dont want to add Squeeze and Excite layer.
         stackwise_activation: list of activation functions, for each inverted
              residual block in the model.
-        include_rescaling: bool, whether to rescale the inputs. If set to True,
-            inputs will be passed through a `Rescaling(scale=1 / 255)`
-            layer.
         image_shape: optional shape tuple, defaults to (224, 224, 3).
         depth_multiplier: float, controls the width of the network.
             - If `depth_multiplier` < 1.0, proportionally decreases the number
@@ -92,7 +89,6 @@ class MobileNetBackbone(Backbone):
         stackwise_num_strides=[2, 2, 1],
         stackwise_se_ratio=[0.25, None, 0.25],
         stackwise_activation=["relu", "relu6", "hard_swish"],
-        include_rescaling=False,
         output_num_filters=1280,
         input_activation='hard_swish',
         output_activation='hard_swish',
@@ -111,7 +107,6 @@ class MobileNetBackbone(Backbone):
         stackwise_num_strides,
         stackwise_se_ratio,
         stackwise_activation,
-        include_rescaling,
         output_num_filters,
         inverted_res_block,
         image_shape=(224, 224, 3),
@@ -126,12 +121,8 @@ class MobileNetBackbone(Backbone):
             -1 if keras.config.image_data_format() == "channels_last" else 1
         )
 
-        inputs = keras.layers.Input(shape=image_shape)
-        x = inputs
-
-        if include_rescaling:
-            x = keras.layers.Rescaling(scale=1 / 255)(x)
-
+        image_input = keras.layers.Input(shape=image_shape)
+        x = image_input  # Intermediate result.
         input_num_filters = adjust_channels(input_num_filters)
         x = keras.layers.Conv2D(
             input_num_filters,
@@ -195,7 +186,7 @@ class MobileNetBackbone(Backbone):
             )(x)
             x = keras.layers.Activation(output_activation)(x)
 
-        super().__init__(inputs=inputs, outputs=x, **kwargs)
+        super().__init__(inputs=image_input, outputs=x, **kwargs)
 
         # === Config ===
         self.stackwise_expansion = stackwise_expansion
@@ -204,7 +195,6 @@ class MobileNetBackbone(Backbone):
         self.stackwise_num_strides = stackwise_num_strides
         self.stackwise_se_ratio = stackwise_se_ratio
         self.stackwise_activation = stackwise_activation
-        self.include_rescaling = include_rescaling
         self.depth_multiplier = depth_multiplier
         self.input_num_filters = input_num_filters
         self.output_num_filters = output_num_filters
@@ -223,7 +213,6 @@ class MobileNetBackbone(Backbone):
                 "stackwise_num_strides": self.stackwise_num_strides,
                 "stackwise_se_ratio": self.stackwise_se_ratio,
                 "stackwise_activation": self.stackwise_activation,
-                "include_rescaling": self.include_rescaling,
                 "image_shape": self.image_shape,
                 "depth_multiplier": self.depth_multiplier,
                 "input_num_filters": self.input_num_filters,

keras_hub/src/models/mobilenet/mobilenet_image_classifier.py

@@ -56,7 +56,6 @@ class MobileNetImageClassifier(ImageClassifier):
         stackwise_stride = [2, 2, 1],
         stackwise_se_ratio = [ 0.25, None, 0.25],
         stackwise_activation = ["relu", "relu", "hard_swish"],
-        include_rescaling = False,
         output_filter=1280,
         activation="hard_swish",
         inverted_res_block=True,

keras_hub/src/models/pali_gemma/pali_gemma_vit.py

@@ -476,6 +476,9 @@ class PaliGemmaVit(keras.Model):
             shape=(image_size, image_size, 3), name="images"
         )
         x = image_input  # Intermediate result.
+        # TODO we have moved this rescaling to preprocessing layers for most
+        # models. We should consider removing it here, though it would break
+        # compatibility.
         if include_rescaling:
             rescaling = keras.layers.Rescaling(
                 scale=1.0 / 127.5, offset=-1.0, name="rescaling"

keras_hub/src/models/resnet/resnet_backbone.py

@@ -44,9 +44,6 @@ class ResNetBackbone(FeaturePyramidBackbone):
     additional pooling operation rather than performing downsampling within
     the convolutional layers themselves.
 
-    Note that `ResNetBackbone` expects the inputs to be images with a value
-    range of `[0, 255]` when `include_rescaling=True`.
-
     Args:
         input_conv_filters: list of ints. The number of filters of the initial
             convolution(s).
@@ -65,9 +62,6 @@ class ResNetBackbone(FeaturePyramidBackbone):
             variants.
         use_pre_activation: boolean. Whether to use pre-activation or not.
             `True` for ResNetV2, `False` for ResNet.
-        include_rescaling: boolean. If `True`, rescale the input using
-            `Rescaling` and `Normalization` layers. If `False`, do nothing.
-            Defaults to `True`.
         image_shape: tuple. The input shape without the batch size.
             Defaults to `(None, None, 3)`.
         pooling: `None` or str. Pooling mode for feature extraction. Defaults
@@ -124,7 +118,6 @@ class ResNetBackbone(FeaturePyramidBackbone):
         stackwise_num_strides,
         block_type,
         use_pre_activation=False,
-        include_rescaling=True,
         image_shape=(None, None, 3),
         data_format=None,
         dtype=None,
@@ -170,18 +163,7 @@ class ResNetBackbone(FeaturePyramidBackbone):
 
         # === Functional Model ===
         image_input = layers.Input(shape=image_shape)
-        if include_rescaling:
-            x = layers.Rescaling(scale=1 / 255.0, dtype=dtype)(image_input)
-            x = layers.Normalization(
-                axis=bn_axis,
-                mean=(0.485, 0.456, 0.406),
-                variance=(0.229**2, 0.224**2, 0.225**2),
-                dtype=dtype,
-                name="normalization",
-            )(x)
-        else:
-            x = image_input
-
+        x = image_input  # Intermediate result.
         # The padding between torch and tensorflow/jax differs when `strides>1`.
         # Therefore, we need to manually pad the tensor.
         x = layers.ZeroPadding2D(
@@ -299,7 +281,6 @@ class ResNetBackbone(FeaturePyramidBackbone):
         self.stackwise_num_strides = stackwise_num_strides
         self.block_type = block_type
         self.use_pre_activation = use_pre_activation
-        self.include_rescaling = include_rescaling
         self.image_shape = image_shape
         self.pyramid_outputs = pyramid_outputs
         self.data_format = data_format
@@ -315,7 +296,6 @@ class ResNetBackbone(FeaturePyramidBackbone):
                 "stackwise_num_strides": self.stackwise_num_strides,
                 "block_type": self.block_type,
                 "use_pre_activation": self.use_pre_activation,
-                "include_rescaling": self.include_rescaling,
                 "image_shape": self.image_shape,
             }
         )

keras_hub/src/models/resnet/resnet_image_classifier.py

@@ -85,7 +85,6 @@ class ResNetImageClassifier(ImageClassifier):
         stackwise_num_strides=[1, 2, 2],
         block_type="basic_block",
         use_pre_activation=True,
-        include_rescaling=False,
         pooling="avg",
     )
     classifier = keras_hub.models.ResNetImageClassifier(

keras_hub/src/models/resnet/resnet_presets.py

@@ -25,7 +25,7 @@ backbone_presets = {
             "path": "resnet",
             "model_card": "https://arxiv.org/abs/2110.00476",
         },
-        "kaggle_handle": "kaggle://kerashub/resnetv1/keras/resnet_18_imagenet/2",
+        "kaggle_handle": "kaggle://kerashub/resnetv1/keras/resnet_18_imagenet/3",
     },
     "resnet_50_imagenet": {
         "metadata": {
@@ -38,7 +38,7 @@ backbone_presets = {
             "path": "resnet",
             "model_card": "https://arxiv.org/abs/2110.00476",
         },
-        "kaggle_handle": "kaggle://kerashub/resnetv1/keras/resnet_50_imagenet/2",
+        "kaggle_handle": "kaggle://kerashub/resnetv1/keras/resnet_50_imagenet/3",
     },
     "resnet_101_imagenet": {
         "metadata": {
@@ -51,7 +51,7 @@ backbone_presets = {
             "path": "resnet",
             "model_card": "https://arxiv.org/abs/2110.00476",
         },
-        "kaggle_handle": "kaggle://kerashub/resnetv1/keras/resnet_101_imagenet/2",
+        "kaggle_handle": "kaggle://kerashub/resnetv1/keras/resnet_101_imagenet/3",
     },
     "resnet_152_imagenet": {
         "metadata": {
@@ -64,7 +64,7 @@ backbone_presets = {
             "path": "resnet",
             "model_card": "https://arxiv.org/abs/2110.00476",
         },
-        "kaggle_handle": "kaggle://kerashub/resnetv1/keras/resnet_152_imagenet/2",
+        "kaggle_handle": "kaggle://kerashub/resnetv1/keras/resnet_152_imagenet/3",
     },
     "resnet_v2_50_imagenet": {
         "metadata": {
@@ -77,7 +77,7 @@ backbone_presets = {
             "path": "resnet",
             "model_card": "https://arxiv.org/abs/2110.00476",
         },
-        "kaggle_handle": "kaggle://kerashub/resnetv2/keras/resnet_v2_50_imagenet/2",
+        "kaggle_handle": "kaggle://kerashub/resnetv2/keras/resnet_v2_50_imagenet/3",
     },
     "resnet_v2_101_imagenet": {
         "metadata": {
@@ -90,6 +90,6 @@ backbone_presets = {
             "path": "resnet",
             "model_card": "https://arxiv.org/abs/2110.00476",
         },
-        "kaggle_handle": "kaggle://kerashub/resnetv2/keras/resnet_v2_101_imagenet/2",
+        "kaggle_handle": "kaggle://kerashub/resnetv2/keras/resnet_v2_101_imagenet/3",
     },
 }

keras_hub/src/models/vgg/vgg_backbone.py

@@ -33,8 +33,6 @@ class VGGBackbone(Backbone):
       stackwise_num_filters: list of ints, filter size for convolutional
             blocks per VGG block. For both VGG16 and VGG19 this is [
             64, 128, 256, 512, 512].
-      include_rescaling: bool, whether to rescale the inputs. If set to
-        True, inputs will be passed through a `Rescaling(1/255.0)` layer.
       image_shape: tuple, optional shape tuple, defaults to (224, 224, 3).
       pooling: bool, Optional pooling mode for feature extraction
         when `include_top` is `False`.
@@ -61,7 +59,6 @@ class VGGBackbone(Backbone):
         stackwise_num_repeats = [2, 2, 3, 3, 3],
         stackwise_num_filters = [64, 128, 256, 512, 512],
         image_shape = (224, 224, 3),
-        include_rescaling = False,
         pooling = "avg",
     )
     model(input_data)
@@ -72,7 +69,6 @@ class VGGBackbone(Backbone):
         self,
         stackwise_num_repeats,
         stackwise_num_filters,
-        include_rescaling,
         image_shape=(224, 224, 3),
         pooling="avg",
         **kwargs,
@@ -82,8 +78,6 @@ class VGGBackbone(Backbone):
         img_input = keras.layers.Input(shape=image_shape)
         x = img_input
 
-        if include_rescaling:
-            x = layers.Rescaling(scale=1 / 255.0)(x)
         for stack_index in range(len(stackwise_num_repeats) - 1):
             x = apply_vgg_block(
                 x=x,
@@ -105,7 +99,6 @@ class VGGBackbone(Backbone):
         # === Config ===
         self.stackwise_num_repeats = stackwise_num_repeats
         self.stackwise_num_filters = stackwise_num_filters
-        self.include_rescaling = include_rescaling
         self.image_shape = image_shape
         self.pooling = pooling
 
@@ -113,7 +106,6 @@ class VGGBackbone(Backbone):
         return {
             "stackwise_num_repeats": self.stackwise_num_repeats,
             "stackwise_num_filters": self.stackwise_num_filters,
-            "include_rescaling": self.include_rescaling,
             "image_shape": self.image_shape,
             "pooling": self.pooling,
         }

keras_hub/src/models/vgg/vgg_image_classifier.py

@@ -66,7 +66,6 @@ class VGGImageClassifier(ImageClassifier):
         stackwise_num_repeats = [2, 2, 3, 3, 3],
         stackwise_num_filters = [64, 128, 256, 512, 512],
         image_shape = (224, 224, 3),
-        include_rescaling = False,
         pooling = "avg",
     )
     classifier = keras_hub.models.VGGImageClassifier(

keras_hub/src/models/vit_det/vit_det_backbone.py

@@ -46,9 +46,6 @@ class ViTDetBackbone(Backbone):
             global attention.
         image_shape (tuple[int], optional): The size of the input image in
             `(H, W, C)` format. Defaults to `(1024, 1024, 3)`.
-        include_rescaling (bool, optional): Whether to rescale the inputs. If
-            set to `True`, inputs will be passed through a
-            `Rescaling(1/255.0)` layer. Defaults to `False`.
         patch_size (int, optional): the patch size to be supplied to the
             Patching layer to turn input images into a flattened sequence of
             patches. Defaults to `16`.
@@ -96,7 +93,6 @@ class ViTDetBackbone(Backbone):
         intermediate_dim,
         num_heads,
         global_attention_layer_indices,
-        include_rescaling=True,
         image_shape=(1024, 1024, 3),
         patch_size=16,
         num_output_channels=256,
@@ -123,9 +119,6 @@ class ViTDetBackbone(Backbone):
             )
         img_size = img_input.shape[-3]
         x = img_input
-        if include_rescaling:
-            # Use common rescaling strategy across keras_cv
-            x = keras.layers.Rescaling(1.0 / 255.0)(x)
         # VITDet scales inputs based on the standard ImageNet mean/stddev.
         x = (x - ops.array([0.485, 0.456, 0.406], dtype=x.dtype)) / (
             ops.array([0.229, 0.224, 0.225], dtype=x.dtype)
@@ -179,14 +172,12 @@ class ViTDetBackbone(Backbone):
         self.window_size = window_size
         self.global_attention_layer_indices = global_attention_layer_indices
         self.layer_norm_epsilon = layer_norm_epsilon
-        self.include_rescaling = include_rescaling
 
     def get_config(self):
         config = super().get_config()
         config.update(
             {
                 "image_shape": self.image_shape,
-                "include_rescaling": self.include_rescaling,
                 "patch_size": self.patch_size,
                 "hidden_size": self.hidden_size,
                 "num_layers": self.num_layers,

keras_hub/src/utils/timm/convert_resnet.py

@@ -151,14 +151,6 @@ def convert_weights(backbone, loader, timm_config):
     if version == "v2":
         port_batch_normalization("post_bn", "norm")
 
-    # Rebuild normalization layer with pretrained mean & std
-    mean = timm_config["pretrained_cfg"]["mean"]
-    std = timm_config["pretrained_cfg"]["std"]
-    normalization_layer = backbone.get_layer("normalization")
-    normalization_layer.input_mean = mean
-    normalization_layer.input_variance = [s**2 for s in std]
-    normalization_layer.build(normalization_layer._build_input_shape)
-
 
 def convert_head(task, loader, timm_config):
     v2 = "resnetv2_" in timm_config["architecture"]

keras_hub/src/utils/timm/preset_loader.py

@@ -62,5 +62,20 @@ class TimmPresetLoader(PresetLoader):
         pretrained_cfg = self.config.get("pretrained_cfg", None)
         if not pretrained_cfg or "input_size" not in pretrained_cfg:
             return None
+        # This assumes the same basic setup for all timm preprocessing, and that
+        # all our image conversion will be via a `ResizingImageConverter. We may
+        # need to extend this as we cover more model types.
         input_size = pretrained_cfg["input_size"]
-        return cls(width=input_size[1], height=input_size[2])
+        mean = pretrained_cfg["mean"]
+        variance = [s**2 for s in pretrained_cfg["std"]]
+        interpolation = pretrained_cfg["interpolation"]
+        if interpolation not in ("bilinear", "nearest", "bicubic"):
+            interpolation = "bilinear"  # Unsupported interpolation type.
+        return cls(
+            width=input_size[1],
+            height=input_size[2],
+            scale=1 / 255.0,
+            mean=mean,
+            variance=variance,
+            interpolation=interpolation,
+        )

keras_hub/src/version_utils.py

@@ -15,7 +15,7 @@
 from keras_hub.src.api_export import keras_hub_export
 
 # Unique source of truth for the version number.
-__version__ = "0.16.0.dev202409201942"
+__version__ = "0.16.1.dev202409210335"
 
 
 @keras_hub_export("keras_hub.version")

{keras_hub_nightly-0.16.0.dev202409201942.dist-info → keras_hub_nightly-0.16.1.dev202409210335.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: keras-hub-nightly
-Version: 0.16.0.dev202409201942
+Version: 0.16.1.dev202409210335
 Summary: Industry-strength Natural Language Processing extensions for Keras.
 Home-page: https://github.com/keras-team/keras-hub
 Author: Keras team

{keras_hub_nightly-0.16.0.dev202409201942.dist-info → keras_hub_nightly-0.16.1.dev202409210335.dist-info}/RECORD

@@ -8,8 +8,8 @@ keras_hub/api/samplers/__init__.py,sha256=l56H4y3h_HlRn_PpeMyZ6vC7228EH_BVFo4Caa
 keras_hub/api/tokenizers/__init__.py,sha256=nzMwKmxkMCOiYB35BIgxHNveCM9WoYRp7ChhmVK8MIM,3042
 keras_hub/api/utils/__init__.py,sha256=4IXDgmXqFzqrCK2MPgkih0Ye1s-8hrlBaUk-n5Kqwl4,800
 keras_hub/src/__init__.py,sha256=lY7spwqXeGX_75qOHiSCff7FPvFCvRamJMF5ua9OWCg,585
-keras_hub/src/api_export.py,sha256=agkICNX5rGcJy_Bj29vaNmhH3no9KqJBO-V3MaqR6HQ,2062
-keras_hub/src/version_utils.py,sha256=4-xR-e8seKXQoWJuaZyu1IxKkBNDAI-WbkioZF8KDPI,808
+keras_hub/src/api_export.py,sha256=82JzmDgnWTJR-PRJI9L_vjhW2Svz8gilbE1NMGZ2JgA,2085
+keras_hub/src/version_utils.py,sha256=n0vYCPXxWIvl26pl1OjjwU68CgOOokRrcKD05BhvVNY,808
 keras_hub/src/bounding_box/__init__.py,sha256=lY7spwqXeGX_75qOHiSCff7FPvFCvRamJMF5ua9OWCg,585
 keras_hub/src/bounding_box/converters.py,sha256=V2ti6xPpaBgeLKbTpCsHsABdYOYASerIKX9oWqeOjHo,18450
 keras_hub/src/bounding_box/formats.py,sha256=5bbHO-n2ADsKIOBJDHMvIPCeNBaV1_mj-NVCgBKNiu8,4453
@@ -40,7 +40,7 @@ keras_hub/src/layers/preprocessing/multi_segment_packer.py,sha256=0se5fOIz-2fMt4
 keras_hub/src/layers/preprocessing/preprocessing_layer.py,sha256=5jFBScsNWuYyokPt8mUoyYeOkKH9ZS7MkeC3j-nxYHU,1273
 keras_hub/src/layers/preprocessing/random_deletion.py,sha256=P4YkpDXgQnlXEgukk6V_iuIrRIQOOC9i8KMkpd7UDic,10349
 keras_hub/src/layers/preprocessing/random_swap.py,sha256=Wu6pNuQ1l_5VRGlRxcomrWyEnqYfA4PcK-mHNuvSjr0,10090
-keras_hub/src/layers/preprocessing/resizing_image_converter.py,sha256=xbDDbJUL2IJ7Zv-CWFH8qtNjvGDrsj4Kf2L3usohIC0,4282
+keras_hub/src/layers/preprocessing/resizing_image_converter.py,sha256=P7KDWTGSnf40iUGUXhCkxx7A5kQMsTF1s3PxYkYxa6U,6440
 keras_hub/src/layers/preprocessing/start_end_packer.py,sha256=3IvVoOE-0kovt_8o2w-uVYEPFhGg-tmv3cwuJQu7VPc,8560
 keras_hub/src/metrics/__init__.py,sha256=lY7spwqXeGX_75qOHiSCff7FPvFCvRamJMF5ua9OWCg,585
 keras_hub/src/metrics/bleu.py,sha256=r0vROmLVVNjc1d9fwJgc64lwmhEXHNaNT1ed1h7Y0E0,14259
@@ -95,8 +95,8 @@ keras_hub/src/models/bloom/bloom_decoder.py,sha256=hSoeVnwRQvGbpVhYmf7-k8FB3Wg4a
 keras_hub/src/models/bloom/bloom_presets.py,sha256=7GiGFPmcXd_UraNsWGQffpzjKDRF-7nqIoUsic78xf0,4696
 keras_hub/src/models/bloom/bloom_tokenizer.py,sha256=ZMx8mHhw0D50zmmvYdmpg-Lk2GcvHz7pPlRpPlhS_2s,3161
 keras_hub/src/models/csp_darknet/__init__.py,sha256=lY7spwqXeGX_75qOHiSCff7FPvFCvRamJMF5ua9OWCg,585
-keras_hub/src/models/csp_darknet/csp_darknet_backbone.py,sha256=Zc3liZuKV-lgAKSAGGKZzsYyFRQwMFMI1qIkUGVUMBM,14718
-keras_hub/src/models/csp_darknet/csp_darknet_image_classifier.py,sha256=h74Q_VHaoSAkwBsDV-ZufN6fb9NFX2gDVk7AOvX-HUk,4388
+keras_hub/src/models/csp_darknet/csp_darknet_backbone.py,sha256=h0eua1EZP0vBV416uOVMmMP1JXy7cVoEj0JEO0OO_lc,14312
+keras_hub/src/models/csp_darknet/csp_darknet_image_classifier.py,sha256=qLav7bxuzB0oaNJLs8gIiQbQVFjAlteDT7WKRfKoSmk,4355
 keras_hub/src/models/deberta_v3/__init__.py,sha256=NCuHFWsgQl-Wer7w3xETvqFtF75AyKabjAYdOlyN34w,874
 keras_hub/src/models/deberta_v3/deberta_v3_backbone.py,sha256=_J-PpSLubay58YO51BicDK0bF97aUeoC21ZQOt1O9r0,7831
 keras_hub/src/models/deberta_v3/deberta_v3_masked_lm.py,sha256=urcktTsXN3kDWnppplnC8yISGx37qGW5HdwHSC7VDLE,4773
@@ -109,8 +109,8 @@ keras_hub/src/models/deberta_v3/disentangled_attention_encoder.py,sha256=Zt10UPx
 keras_hub/src/models/deberta_v3/disentangled_self_attention.py,sha256=MxpWy30h9JB8nlEk7V9_wETzP-tpv1Sd1Wiz_pHGpkI,13708
 keras_hub/src/models/deberta_v3/relative_embedding.py,sha256=QT5MAnheJ1wSKFeN49pdnZzWkztz5K2oYYuNEtB_5xM,3472
 keras_hub/src/models/densenet/__init__.py,sha256=lY7spwqXeGX_75qOHiSCff7FPvFCvRamJMF5ua9OWCg,585
-keras_hub/src/models/densenet/densenet_backbone.py,sha256=cMTTaI1WogaSjt8x8bpPMvApYp5NVmeHTfupUmZZ774,7661
-keras_hub/src/models/densenet/densenet_image_classifier.py,sha256=bmmkNNpxwkwfqI_ZMmoEATClmgmmkW6NO5tDK8BCt2Y,4336
+keras_hub/src/models/densenet/densenet_backbone.py,sha256=BbTecC7gfigSC3t4L-kGsZHS7pjj8DtDIztyMxo_AoI,7238
+keras_hub/src/models/densenet/densenet_image_classifier.py,sha256=eECPZKHycVHNbgFuBHyiZGPWBn0M_pBdLasjmroc95g,4303
 keras_hub/src/models/distil_bert/__init__.py,sha256=EiJUA3y_b22rMacMbBD7jD0eBSzR-wbVtF73k2RsQow,889
 keras_hub/src/models/distil_bert/distil_bert_backbone.py,sha256=ZW2OgNlWXeRlfI5BrcJLYr4Oc2qNJZoDxjoL7-cGuIQ,7027
 keras_hub/src/models/distil_bert/distil_bert_masked_lm.py,sha256=1BFS1At_HYlLK21VWyhQPrPtActpmR52A8LJG2c6N8Y,4862
@@ -120,7 +120,7 @@ keras_hub/src/models/distil_bert/distil_bert_text_classifier.py,sha256=Q-qGmyl6i
 keras_hub/src/models/distil_bert/distil_bert_text_classifier_preprocessor.py,sha256=sad3XpW2HfjG2iQ4JRm1tw2jp4pZCN4LYwF1mM4GUps,5480
 keras_hub/src/models/distil_bert/distil_bert_tokenizer.py,sha256=VK7kZJEbsClp20uWVb6pj-WSUU5IMdRBk0jyUIM_RIg,3698
 keras_hub/src/models/efficientnet/__init__.py,sha256=lY7spwqXeGX_75qOHiSCff7FPvFCvRamJMF5ua9OWCg,585
-keras_hub/src/models/efficientnet/efficientnet_backbone.py,sha256=krz5lgw5cPs2EyKArq99XnIfUeBVbkeq2PhPFADO04c,21841
+keras_hub/src/models/efficientnet/efficientnet_backbone.py,sha256=i-K9kYwnl2Ninuebw6nNJ6X7D_4dvjMrV1Y9XAdt6I4,21392
 keras_hub/src/models/efficientnet/fusedmbconv.py,sha256=_6aNQKL2XdVNgoAdKvvTh_NDkWeU66q98EFUOjEQ1UM,7933
 keras_hub/src/models/efficientnet/mbconv.py,sha256=LNbEj7RpEZ0SqzEu-7ZpH1BKm6Ne2sXPckc5c2DMqUk,8212
 keras_hub/src/models/electra/__init__.py,sha256=ixE5hAkfTFfErqbYVyIUKMT8MUz-u_175QXxEBIiGBU,849
@@ -191,12 +191,12 @@ keras_hub/src/models/mistral/mistral_presets.py,sha256=uF1Q4zllcV1upIlqmn3gxhVWz
 keras_hub/src/models/mistral/mistral_tokenizer.py,sha256=pO7mpzYgRDFpIrsmLBL3zxkadrOE0xfFj30c2nHN42c,2591
 keras_hub/src/models/mistral/mistral_transformer_decoder.py,sha256=6CdaZt1lQ9VcLz_OoYroqiqvsZfq9H5VGaWab25aCRI,10127
 keras_hub/src/models/mix_transformer/__init__.py,sha256=lY7spwqXeGX_75qOHiSCff7FPvFCvRamJMF5ua9OWCg,585
-keras_hub/src/models/mix_transformer/mix_transformer_backbone.py,sha256=TYcQCAMTZedirh2L4z8LrjfhmxR2CoImzIvVXFTiTMc,6833
-keras_hub/src/models/mix_transformer/mix_transformer_classifier.py,sha256=QUTeq4f07nrCE-hIKoam_M6jJ6aM9l6s_At5sRTo0JY,4310
+keras_hub/src/models/mix_transformer/mix_transformer_backbone.py,sha256=1OUWvrI4y5rzoOsQkB8ZqQqeg5DwFIWRY-IKgR5qDfA,6426
+keras_hub/src/models/mix_transformer/mix_transformer_classifier.py,sha256=Kq-FIayi0yiJ1P4_AhwdBAC-vFnfhEK3FYlmBjw4jUc,4277
 keras_hub/src/models/mix_transformer/mix_transformer_layers.py,sha256=Bi4lHMfiKgI-XOt21BBfKoK05uU3GcDJ3mQrGfCXb6Y,10123
 keras_hub/src/models/mobilenet/__init__.py,sha256=lY7spwqXeGX_75qOHiSCff7FPvFCvRamJMF5ua9OWCg,585
-keras_hub/src/models/mobilenet/mobilenet_backbone.py,sha256=G02NFvx2xy2mbEBX6mtJzhPwygZDAhJ2TMk2ejAuLg0,19168
-keras_hub/src/models/mobilenet/mobilenet_image_classifier.py,sha256=Oo3URtyqjfnmsyO9uncxOVHO9Giv607LBJ3UE8pWacU,3794
+keras_hub/src/models/mobilenet/mobilenet_backbone.py,sha256=Y950Yx4s5fTmVk7YTiMFiyqZLLuB75_iJaVbefznOwo,18776
+keras_hub/src/models/mobilenet/mobilenet_image_classifier.py,sha256=35Px2z1E_ATSZIYNb_bXjJ6Qimbd2rnPi04S99ycTNg,3759
 keras_hub/src/models/opt/__init__.py,sha256=DiiylcsbseSQ8te8KWZ6BTIaKYSzXHUPGBgFssFNGFY,825
 keras_hub/src/models/opt/opt_backbone.py,sha256=cbm9I7d3QlGD8l2W1eK8esqc5gm77tpwxg4t9nC-FtA,6460
 keras_hub/src/models/opt/opt_causal_lm.py,sha256=z6M8cQV-c8q7HmikNA9RuvsMMvQYF21-ZcC0nVGfnp8,11438
@@ -211,7 +211,7 @@ keras_hub/src/models/pali_gemma/pali_gemma_decoder_block.py,sha256=fXLO4uHtWYTuE
 keras_hub/src/models/pali_gemma/pali_gemma_image_converter.py,sha256=Wm1A-HuOMxesAHFbEpP5ZkPbdDaVW5CTTwkyFpI-WdI,990
 keras_hub/src/models/pali_gemma/pali_gemma_presets.py,sha256=cG5cV2bkiDJlKDiHX76BpnClsY5PcmLDezDg7emeiA4,2986
 keras_hub/src/models/pali_gemma/pali_gemma_tokenizer.py,sha256=7F1TQql3DEN517iVbNL60u6fQPimrGQvWBYh16ng8JU,3000
-keras_hub/src/models/pali_gemma/pali_gemma_vit.py,sha256=JUfJuyobcEb60jp3sIxlq12gIH_qsn97h4hsecimipQ,19092
+keras_hub/src/models/pali_gemma/pali_gemma_vit.py,sha256=GUMAuFcpoi0TxJk7LzsKp0Tt0c_83gx645cz26GqFzA,19271
 keras_hub/src/models/phi3/__init__.py,sha256=ENAOZhScWf9RbPmkiuICR5gr36ZMUn4AniLvJOrykj8,831
 keras_hub/src/models/phi3/phi3_attention.py,sha256=BcYApteLjbrCzube7jHVagc0mMpDCReRyvsQhQcJzY8,9828
 keras_hub/src/models/phi3/phi3_backbone.py,sha256=MvTE5bMmVpFHinZIEDBM1lfJFbgu4zg-0e-8_4hK-No,9470
@@ -223,11 +223,11 @@ keras_hub/src/models/phi3/phi3_presets.py,sha256=S7_gIqPxU5FQAEnAE_68UrfGGSLOMvo
 keras_hub/src/models/phi3/phi3_rotary_embedding.py,sha256=QVJIgpOw6iMicGrsPdW8eF84vV_stf0Tqm2qBJdsKH0,5597
 keras_hub/src/models/phi3/phi3_tokenizer.py,sha256=hlA-u2sTRYARDW3ABICPeiOYW1AJwr-5kvZk3EB5z7M,2577
 keras_hub/src/models/resnet/__init__.py,sha256=41gttaQ7gt_ZaqDa_GKuMPfIk5c88-GrdC1h9fBUTXc,843
-keras_hub/src/models/resnet/resnet_backbone.py,sha256=n9aKIpQcJCsAZrBiiN1vxUMHeQgYudRHdu_MsdRQZqw,33260
-keras_hub/src/models/resnet/resnet_image_classifier.py,sha256=I-dmx0O_ES3m3W5D4ICCux5zzDMZ2cM0vYGM9CDi5AE,5395
+keras_hub/src/models/resnet/resnet_backbone.py,sha256=Qu2MuPBNYasQDD4zeY2rnUUqiEYRXqjbeXilcUdimkA,32451
+keras_hub/src/models/resnet/resnet_image_classifier.py,sha256=4Ksxhp4kB93mbkjh7K-uKcCyEO4MtMazHN7VtUCL-wg,5362
 keras_hub/src/models/resnet/resnet_image_classifier_preprocessor.py,sha256=Vrs9NBZRL5fgDXXY27GZJg5xMa5_wovi8A2z8kFl2nc,1129
 keras_hub/src/models/resnet/resnet_image_converter.py,sha256=820drIU5Kkib7gC7T418mmrhsBHSkenfEiZ6-fkChv0,961
-keras_hub/src/models/resnet/resnet_presets.py,sha256=DZoufeJyrVDL4aHSztQNzZj8Cb_OGX53Fn0Ze4RuZCI,3550
+keras_hub/src/models/resnet/resnet_presets.py,sha256=6y8R-PviAnEyh-LFli9uMUNku4cJC9V7YqOd9V5PlV0,3550
 keras_hub/src/models/retinanet/__init__.py,sha256=lY7spwqXeGX_75qOHiSCff7FPvFCvRamJMF5ua9OWCg,585
 keras_hub/src/models/retinanet/anchor_generator.py,sha256=VQwgIAWh-6s28TU8MHFdl556U6h7rfF9B9iVI_zwI7c,7027
 keras_hub/src/models/retinanet/box_matcher.py,sha256=SvGn_6d5sfjq522UaHpxVCE2S5Nwml_aj5yAKApTNE4,11420
@@ -259,10 +259,10 @@ keras_hub/src/models/t5/t5_presets.py,sha256=2RT_NuJcqDdSeAsoSJXh5O_ax2H-s4YKTAo
 keras_hub/src/models/t5/t5_tokenizer.py,sha256=UnmZjiKhyb4AU7zALW3YAM_6_OGzYOVEGStBiw4ICvg,3103
 keras_hub/src/models/t5/t5_transformer_layer.py,sha256=wnu108InkHH9YMmFNTbmgIqcrKQQUxeJ7S1dcjUfBSY,5933
 keras_hub/src/models/vgg/__init__.py,sha256=lY7spwqXeGX_75qOHiSCff7FPvFCvRamJMF5ua9OWCg,585
-keras_hub/src/models/vgg/vgg_backbone.py,sha256=dMXIGypDQdLztvbHz0JgSdTGXXAZj11vLxG5oHk4ZNw,5479
-keras_hub/src/models/vgg/vgg_image_classifier.py,sha256=1bH6E46yHxN5tey2Mc62U3l4_5mTZ40U00bws-c6wqE,4106
+keras_hub/src/models/vgg/vgg_backbone.py,sha256=O6onZEduEPt1J4v2HFgtHsxu-SheqpUwY2pYoeLa6uE,5080
+keras_hub/src/models/vgg/vgg_image_classifier.py,sha256=cDcmHoHU1BZ211JakGPw3Z9lV22oMmK8J4-Ng8S07G0,4071
 keras_hub/src/models/vit_det/__init__.py,sha256=lY7spwqXeGX_75qOHiSCff7FPvFCvRamJMF5ua9OWCg,585
-keras_hub/src/models/vit_det/vit_det_backbone.py,sha256=Tyw3xTOW1rlHV-copzotzpaoPLWU8nA-LtViUGGgSlw,8541
+keras_hub/src/models/vit_det/vit_det_backbone.py,sha256=4b3CUk4zg8gjFJvDU-QJZP72CV8jqw3TnaoCzUC-vyo,8054
 keras_hub/src/models/vit_det/vit_layers.py,sha256=JeUzOT2jmSOoJ_OiHOfLSkkCUZ5mlK5Mfd21DwudRCQ,20436
 keras_hub/src/models/whisper/__init__.py,sha256=FI-xj6FwZDAAdCfKhOrE1_roQ8cXhD1gK4G6CLTvPQo,849
 keras_hub/src/models/whisper/whisper_audio_converter.py,sha256=JqtA2kLUMFKZ4FrI8g2piEjahE-0-F3Yp4qQXS1cYf4,8973
@@ -314,8 +314,8 @@ keras_hub/src/utils/tensor_utils.py,sha256=XpWORE8iUzHXv1E1akiYDep07ndZJRKvjsKVl
 keras_hub/src/utils/imagenet/__init__.py,sha256=AK2s8L-VARI5OmlT6G3vtlKIVyjwLfgVwXfxzhhSCq4,585
 keras_hub/src/utils/imagenet/imagenet_utils.py,sha256=0iHrAQbh5DCa9Dh7tJiQeJc7AGzNO7j0cFEWS2Of16w,39889
 keras_hub/src/utils/timm/__init__.py,sha256=lY7spwqXeGX_75qOHiSCff7FPvFCvRamJMF5ua9OWCg,585
-keras_hub/src/utils/timm/convert_resnet.py,sha256=hZNj_kpwSA9Jp3NRDHtCPzHFzRKKPnidKQUAoqcdENk,6810
-keras_hub/src/utils/timm/preset_loader.py,sha256=EgS5xBP3sWYiTgKmOAMmj3b3kRWcPnsWLieReLHZ178,2928
+keras_hub/src/utils/timm/convert_resnet.py,sha256=X2N9lk8sqRMzOMXkcIThAu6ZEtw8u8_Y4Kol82iTuW4,6417
+keras_hub/src/utils/timm/preset_loader.py,sha256=ac2PwGkfe-bikhQEFeIM25gDs3xk0E9SS5A1YEzZYQU,3602
 keras_hub/src/utils/transformers/__init__.py,sha256=lY7spwqXeGX_75qOHiSCff7FPvFCvRamJMF5ua9OWCg,585
 keras_hub/src/utils/transformers/convert_albert.py,sha256=7b9X1TLrWfHieoeX_K-EXTagkl4Rp9AfPjsPrwArBGY,8280
 keras_hub/src/utils/transformers/convert_bart.py,sha256=RXmPf_XUZrUyqDaOV9T7qVNEP4rAVR44oK1aRZI0v78,14996
@@ -328,7 +328,7 @@ keras_hub/src/utils/transformers/convert_mistral.py,sha256=4QStizMS6ESEPjSI-ls6j
 keras_hub/src/utils/transformers/convert_pali_gemma.py,sha256=BT5eX1QzbjCQCopbMstiejQQWQiB_N77bpD5FMUygEo,11234
 keras_hub/src/utils/transformers/preset_loader.py,sha256=9x9hLhDh_6PAHG5gay5rVoEVyt-gXTQGrnprjMLKvCM,3294
 keras_hub/src/utils/transformers/safetensor_utils.py,sha256=2O8lcCf9yIFt5xiRVOtF1ZkPb5pfhOfDJotBaanD9Zo,3547
-keras_hub_nightly-0.16.0.dev202409201942.dist-info/METADATA,sha256=veZ6kme7NK3BC_Hg4JywZPIyP66gWgsgas4PsvfXbkU,7061
-keras_hub_nightly-0.16.0.dev202409201942.dist-info/WHEEL,sha256=GV9aMThwP_4oNCtvEC2ec3qUYutgWeAzklro_0m4WJQ,91
-keras_hub_nightly-0.16.0.dev202409201942.dist-info/top_level.txt,sha256=N4J6piIWBKa38A4uV-CnIopnOEf8mHAbkNXafXm_CuA,10
-keras_hub_nightly-0.16.0.dev202409201942.dist-info/RECORD,,
+keras_hub_nightly-0.16.1.dev202409210335.dist-info/METADATA,sha256=XDJP6zqkNwmPIjLkOEAHp9hh8z1JLqO09nMcGLlnmtU,7061
+keras_hub_nightly-0.16.1.dev202409210335.dist-info/WHEEL,sha256=GV9aMThwP_4oNCtvEC2ec3qUYutgWeAzklro_0m4WJQ,91
+keras_hub_nightly-0.16.1.dev202409210335.dist-info/top_level.txt,sha256=N4J6piIWBKa38A4uV-CnIopnOEf8mHAbkNXafXm_CuA,10
+keras_hub_nightly-0.16.1.dev202409210335.dist-info/RECORD,,