keras-hub-nightly 0.19.0.dev202412180349__py3-none-any.whl → 0.19.0.dev202412200346__py3-none-any.whl

keras_hub/api/layers/__init__.py

@@ -35,6 +35,9 @@ from keras_hub.src.layers.preprocessing.multi_segment_packer import (
 from keras_hub.src.layers.preprocessing.random_deletion import RandomDeletion
 from keras_hub.src.layers.preprocessing.random_swap import RandomSwap
 from keras_hub.src.layers.preprocessing.start_end_packer import StartEndPacker
+from keras_hub.src.models.basnet.basnet_image_converter import (
+    BASNetImageConverter,
+)
 from keras_hub.src.models.clip.clip_image_converter import CLIPImageConverter
 from keras_hub.src.models.deeplab_v3.deeplab_v3_image_converter import (
     DeepLabV3ImageConverter,

keras_hub/api/models/__init__.py

@@ -29,6 +29,9 @@ from keras_hub.src.models.bart.bart_seq_2_seq_lm_preprocessor import (
     BartSeq2SeqLMPreprocessor,
 )
 from keras_hub.src.models.bart.bart_tokenizer import BartTokenizer
+from keras_hub.src.models.basnet.basnet import BASNetImageSegmenter
+from keras_hub.src.models.basnet.basnet_backbone import BASNetBackbone
+from keras_hub.src.models.basnet.basnet_preprocessor import BASNetPreprocessor
 from keras_hub.src.models.bert.bert_backbone import BertBackbone
 from keras_hub.src.models.bert.bert_masked_lm import BertMaskedLM
 from keras_hub.src.models.bert.bert_masked_lm_preprocessor import (

keras_hub/src/models/basnet/__init__.py

@@ -0,0 +1,5 @@
+from keras_hub.src.models.basnet.basnet_backbone import BASNetBackbone
+from keras_hub.src.models.basnet.basnet_presets import basnet_presets
+from keras_hub.src.utils.preset_utils import register_presets
+
+register_presets(basnet_presets, BASNetBackbone)

keras_hub/src/models/basnet/basnet.py

@@ -0,0 +1,122 @@
+import keras
+
+from keras_hub.src.api_export import keras_hub_export
+from keras_hub.src.models.basnet.basnet_backbone import BASNetBackbone
+from keras_hub.src.models.basnet.basnet_preprocessor import BASNetPreprocessor
+from keras_hub.src.models.image_segmenter import ImageSegmenter
+
+
+@keras_hub_export("keras_hub.models.BASNetImageSegmenter")
+class BASNetImageSegmenter(ImageSegmenter):
+    """BASNet image segmentation task.
+
+    Args:
+        backbone: A `keras_hub.models.BASNetBackbone` instance.
+        preprocessor: `None`, a `keras_hub.models.Preprocessor` instance,
+            a `keras.Layer` instance, or a callable. If `None` no preprocessing
+            will be applied to the inputs.
+
+    Example:
+    ```python
+    import keras_hub
+
+    images = np.ones(shape=(1, 288, 288, 3))
+    labels = np.zeros(shape=(1, 288, 288, 1))
+
+    image_encoder = keras_hub.models.ResNetBackbone.from_preset(
+        "resnet_18_imagenet",
+        load_weights=False
+    )
+    backbone = keras_hub.models.BASNetBackbone(
+        image_encoder,
+        num_classes=1,
+        image_shape=[288, 288, 3]
+    )
+    model = keras_hub.models.BASNetImageSegmenter(backbone)
+
+    # Evaluate the model
+    pred_labels = model(images)
+
+    # Train the model
+    model.compile(
+        optimizer="adam",
+        loss=keras.losses.BinaryCrossentropy(from_logits=False),
+        metrics=["accuracy"],
+    )
+    model.fit(images, labels, epochs=3)
+    ```
+    """
+
+    backbone_cls = BASNetBackbone
+    preprocessor_cls = BASNetPreprocessor
+
+    def __init__(
+        self,
+        backbone,
+        preprocessor=None,
+        **kwargs,
+    ):
+        # === Functional Model ===
+        x = backbone.input
+        outputs = backbone(x)
+        # only return the refinement module's output as final prediction
+        outputs = outputs["refine_out"]
+        super().__init__(inputs=x, outputs=outputs, **kwargs)
+
+        # === Config ===
+        self.backbone = backbone
+        self.preprocessor = preprocessor
+
+    def compute_loss(self, x, y, y_pred, *args, **kwargs):
+        # train BASNet's prediction and refinement module outputs against the
+        # same ground truth data
+        outputs = self.backbone(x)
+        losses = []
+        for output in outputs.values():
+            losses.append(super().compute_loss(x, y, output, *args, **kwargs))
+        return keras.ops.sum(losses, axis=0)
+
+    def compile(
+        self,
+        optimizer="auto",
+        loss="auto",
+        metrics="auto",
+        **kwargs,
+    ):
+        """Configures the `BASNet` task for training.
+
+        `BASNet` extends the default compilation signature
+        of `keras.Model.compile` with defaults for `optimizer` and `loss`. To
+        override these defaults, pass any value to these arguments during
+        compilation.
+
+        Args:
+            optimizer: `"auto"`, an optimizer name, or a `keras.Optimizer`
+                instance. Defaults to `"auto"`, which uses the default
+                optimizer for `BASNet`. See `keras.Model.compile` and
+                `keras.optimizers` for more info on possible `optimizer`
+                values.
+            loss: `"auto"`, a loss name, or a `keras.losses.Loss` instance.
+                Defaults to `"auto"`, in which case the default loss
+                computation of `BASNet` will be applied.
+                See `keras.Model.compile` and `keras.losses` for more info on
+                possible `loss` values.
+            metrics: `"auto"`, or a list of metrics to be evaluated by
+                the model during training and testing. Defaults to `"auto"`,
+                where a `keras.metrics.Accuracy` will be applied to track the
+                accuracy of the model during training.
+                See `keras.Model.compile` and `keras.metrics` for
+                more info on possible `metrics` values.
+            **kwargs: See `keras.Model.compile` for a full list of arguments
+                supported by the compile method.
+        """
+        if loss == "auto":
+            loss = keras.losses.BinaryCrossentropy()
+        if metrics == "auto":
+            metrics = [keras.metrics.Accuracy()]
+        super().compile(
+            optimizer=optimizer,
+            loss=loss,
+            metrics=metrics,
+            **kwargs,
+        )

keras_hub/src/models/basnet/basnet_backbone.py

@@ -0,0 +1,366 @@
+import keras
+
+from keras_hub.src.api_export import keras_hub_export
+from keras_hub.src.models.backbone import Backbone
+from keras_hub.src.models.resnet.resnet_backbone import (
+    apply_basic_block as resnet_basic_block,
+)
+
+
+@keras_hub_export("keras_hub.models.BASNetBackbone")
+class BASNetBackbone(Backbone):
+    """BASNet architecture for semantic segmentation.
+
+    A Keras model implementing the BASNet architecture described in [BASNet:
+    Boundary-Aware Segmentation Network for Mobile and Web Applications](
+    https://arxiv.org/abs/2101.04704). BASNet uses a predict-refine
+    architecture for highly accurate image segmentation.
+
+    Args:
+        image_encoder: A `keras_hub.models.ResNetBackbone` instance. The
+            backbone network for the model that is used as a feature extractor
+            for BASNet prediction encoder. Currently supported backbones are
+            ResNet18 and ResNet34.
+            (Note: Do not specify `image_shape` within the backbone.
+            Please provide these while initializing the 'BASNetBackbone' model)
+        num_classes: int, the number of classes for the segmentation model.
+        image_shape: optional shape tuple, defaults to (None, None, 3).
+        projection_filters: int, number of filters in the convolution layer
+            projecting low-level features from the `backbone`.
+        prediction_heads: (Optional) List of `keras.layers.Layer` defining
+            the prediction module head for the model. If not provided, a
+            default head is created with a Conv2D layer followed by resizing.
+        refinement_head: (Optional) a `keras.layers.Layer` defining the
+            refinement module head for the model. If not provided, a default
+            head is created with a Conv2D layer.
+        dtype: `None` or str or `keras.mixed_precision.DTypePolicy`. The dtype
+            to use for the model's computations and weights.
+    """
+
+    def __init__(
+        self,
+        image_encoder,
+        num_classes,
+        image_shape=(None, None, 3),
+        projection_filters=64,
+        prediction_heads=None,
+        refinement_head=None,
+        dtype=None,
+        **kwargs,
+    ):
+        if not isinstance(image_encoder, keras.layers.Layer) or not isinstance(
+            image_encoder, keras.Model
+        ):
+            raise ValueError(
+                "Argument `image_encoder` must be a `keras.layers.Layer`"
+                f" instance or `keras.Model`. Received instead"
+                f" image_encoder={image_encoder} (of type"
+                f" {type(image_encoder)})."
+            )
+
+        if tuple(image_encoder.image_shape) != (None, None, 3):
+            raise ValueError(
+                "Do not specify `image_shape` within the"
+                " `BASNetBackbone`'s image_encoder. \nPlease provide"
+                " `image_shape` while initializing the 'BASNetBackbone' model."
+            )
+
+        # === Functional Model ===
+        inputs = keras.layers.Input(shape=image_shape)
+        x = inputs
+
+        if prediction_heads is None:
+            prediction_heads = []
+            for size in (1, 2, 4, 8, 16, 32, 32):
+                head_layers = [
+                    keras.layers.Conv2D(
+                        num_classes,
+                        kernel_size=(3, 3),
+                        padding="same",
+                        dtype=dtype,
+                    )
+                ]
+                if size != 1:
+                    head_layers.append(
+                        keras.layers.UpSampling2D(
+                            size=size, interpolation="bilinear", dtype=dtype
+                        )
+                    )
+                prediction_heads.append(keras.Sequential(head_layers))
+
+        if refinement_head is None:
+            refinement_head = keras.Sequential(
+                [
+                    keras.layers.Conv2D(
+                        num_classes,
+                        kernel_size=(3, 3),
+                        padding="same",
+                        dtype=dtype,
+                    ),
+                ]
+            )
+
+        # Prediction model.
+        predict_model = basnet_predict(
+            x, image_encoder, projection_filters, prediction_heads, dtype=dtype
+        )
+
+        # Refinement model.
+        refine_model = basnet_rrm(
+            predict_model, projection_filters, refinement_head, dtype=dtype
+        )
+
+        outputs = refine_model.outputs  # Combine outputs.
+        outputs.extend(predict_model.outputs)
+
+        output_names = ["refine_out"] + [
+            f"predict_out_{i}" for i in range(1, len(outputs))
+        ]
+
+        outputs = {
+            output_name: keras.layers.Activation(
+                "sigmoid", name=output_name, dtype=dtype
+            )(output)
+            for output, output_name in zip(outputs, output_names)
+        }
+
+        super().__init__(inputs=inputs, outputs=outputs, dtype=dtype, **kwargs)
+
+        # === Config ===
+        self.image_encoder = image_encoder
+        self.num_classes = num_classes
+        self.image_shape = image_shape
+        self.projection_filters = projection_filters
+        self.prediction_heads = prediction_heads
+        self.refinement_head = refinement_head
+
+    def get_config(self):
+        config = super().get_config()
+        config.update(
+            {
+                "image_encoder": keras.saving.serialize_keras_object(
+                    self.image_encoder
+                ),
+                "num_classes": self.num_classes,
+                "image_shape": self.image_shape,
+                "projection_filters": self.projection_filters,
+                "prediction_heads": [
+                    keras.saving.serialize_keras_object(prediction_head)
+                    for prediction_head in self.prediction_heads
+                ],
+                "refinement_head": keras.saving.serialize_keras_object(
+                    self.refinement_head
+                ),
+            }
+        )
+        return config
+
+    @classmethod
+    def from_config(cls, config):
+        if "image_encoder" in config:
+            config["image_encoder"] = keras.layers.deserialize(
+                config["image_encoder"]
+            )
+        if "prediction_heads" in config and isinstance(
+            config["prediction_heads"], list
+        ):
+            for i in range(len(config["prediction_heads"])):
+                if isinstance(config["prediction_heads"][i], dict):
+                    config["prediction_heads"][i] = keras.layers.deserialize(
+                        config["prediction_heads"][i]
+                    )
+
+        if "refinement_head" in config and isinstance(
+            config["refinement_head"], dict
+        ):
+            config["refinement_head"] = keras.layers.deserialize(
+                config["refinement_head"]
+            )
+        return super().from_config(config)
+
+
+def convolution_block(x_input, filters, dilation=1, dtype=None):
+    """Apply convolution + batch normalization + ReLU activation.
+
+    Args:
+        x_input: Input keras tensor.
+        filters: int, number of output filters in the convolution.
+        dilation: int, dilation rate for the convolution operation.
+            Defaults to 1.
+        dtype: `None` or str or `keras.mixed_precision.DTypePolicy`. The dtype
+            to use for the model's computations and weights.
+
+    Returns:
+        A tensor with convolution, batch normalization, and ReLU
+        activation applied.
+    """
+    x = keras.layers.Conv2D(
+        filters, (3, 3), padding="same", dilation_rate=dilation, dtype=dtype
+    )(x_input)
+    x = keras.layers.BatchNormalization(dtype=dtype)(x)
+    return keras.layers.Activation("relu", dtype=dtype)(x)
+
+
+def get_resnet_block(_resnet, block_num):
+    """Extract and return a specific ResNet block.
+
+    Args:
+        _resnet: `keras.Model`. ResNet model instance.
+        block_num: int, block number to extract.
+
+    Returns:
+        A Keras Model representing the specified ResNet block.
+    """
+
+    extractor_levels = ["P2", "P3", "P4", "P5"]
+    num_blocks = _resnet.stackwise_num_blocks
+    if block_num == 0:
+        x = _resnet.get_layer("pool1_pool").output
+    else:
+        x = _resnet.pyramid_outputs[extractor_levels[block_num - 1]]
+    y = _resnet.get_layer(
+        f"stack{block_num}_block{num_blocks[block_num]-1}_add"
+    ).output
+    return keras.models.Model(
+        inputs=x,
+        outputs=y,
+        name=f"resnet_block{block_num + 1}",
+    )
+
+
+def basnet_predict(x_input, backbone, filters, segmentation_heads, dtype=None):
+    """BASNet Prediction Module.
+
+    This module outputs a coarse label map by integrating heavy
+    encoder, bridge, and decoder blocks.
+
+    Args:
+        x_input: Input keras tensor.
+        backbone: `keras.Model`. The backbone network used as a feature
+            extractor for BASNet prediction encoder.
+        filters: int, the number of filters.
+        segmentation_heads: List of `keras.layers.Layer`, A list of Keras
+            layers serving as the segmentation head for prediction module.
+        dtype: `None` or str or `keras.mixed_precision.DTypePolicy`. The dtype
+            to use for the model's computations and weights.
+
+
+    Returns:
+        A Keras Model that integrates the encoder, bridge, and decoder
+        blocks for coarse label map prediction.
+    """
+    num_stages = 6
+
+    x = x_input
+
+    # -------------Encoder--------------
+    x = keras.layers.Conv2D(
+        filters, kernel_size=(3, 3), padding="same", dtype=dtype
+    )(x)
+
+    encoder_blocks = []
+    for i in range(num_stages):
+        if i < 4:  # First four stages are adopted from ResNet backbone.
+            x = get_resnet_block(backbone, i)(x)
+            encoder_blocks.append(x)
+        else:  # Last 2 stages consist of three basic resnet blocks.
+            x = keras.layers.MaxPool2D(
+                pool_size=(2, 2), strides=(2, 2), dtype=dtype
+            )(x)
+            for j in range(3):
+                x = resnet_basic_block(
+                    x,
+                    filters=x.shape[3],
+                    conv_shortcut=False,
+                    name=f"v1_basic_block_{i + 1}_{j + 1}",
+                    dtype=dtype,
+                )
+            encoder_blocks.append(x)
+
+    # -------------Bridge-------------
+    x = convolution_block(x, filters=filters * 8, dilation=2, dtype=dtype)
+    x = convolution_block(x, filters=filters * 8, dilation=2, dtype=dtype)
+    x = convolution_block(x, filters=filters * 8, dilation=2, dtype=dtype)
+    encoder_blocks.append(x)
+
+    # -------------Decoder-------------
+    decoder_blocks = []
+    for i in reversed(range(num_stages)):
+        if i != (num_stages - 1):  # Except first, scale other decoder stages.
+            x = keras.layers.UpSampling2D(
+                size=2, interpolation="bilinear", dtype=dtype
+            )(x)
+
+        x = keras.layers.concatenate([encoder_blocks[i], x], axis=-1)
+        x = convolution_block(x, filters=filters * 8, dtype=dtype)
+        x = convolution_block(x, filters=filters * 8, dtype=dtype)
+        x = convolution_block(x, filters=filters * 8, dtype=dtype)
+        decoder_blocks.append(x)
+
+    decoder_blocks.reverse()  # Change order from last to first decoder stage.
+    decoder_blocks.append(encoder_blocks[-1])  # Copy bridge to decoder.
+
+    # -------------Side Outputs--------------
+    decoder_blocks = [
+        segmentation_head(decoder_block)  # Prediction segmentation head.
+        for segmentation_head, decoder_block in zip(
+            segmentation_heads, decoder_blocks
+        )
+    ]
+
+    return keras.models.Model(inputs=[x_input], outputs=decoder_blocks)
+
+
+def basnet_rrm(base_model, filters, segmentation_head, dtype=None):
+    """BASNet Residual Refinement Module (RRM).
+
+    This module outputs a fine label map by integrating light encoder,
+    bridge, and decoder blocks.
+
+    Args:
+        base_model: Keras model used as the base or coarse label map.
+        filters: int, the number of filters.
+        segmentation_head: a `keras.layers.Layer`, A Keras layer serving
+            as the segmentation head for refinement module.
+        dtype: `None` or str or `keras.mixed_precision.DTypePolicy`. The dtype
+            to use for the model's computations and weights.
+
+    Returns:
+        A Keras Model that constructs the Residual Refinement Module (RRM).
+    """
+    num_stages = 4
+
+    x_input = base_model.output[0]
+
+    # -------------Encoder--------------
+    x = keras.layers.Conv2D(
+        filters, kernel_size=(3, 3), padding="same", dtype=dtype
+    )(x_input)
+
+    encoder_blocks = []
+    for _ in range(num_stages):
+        x = convolution_block(x, filters=filters)
+        encoder_blocks.append(x)
+        x = keras.layers.MaxPool2D(
+            pool_size=(2, 2), strides=(2, 2), dtype=dtype
+        )(x)
+
+    # -------------Bridge--------------
+    x = convolution_block(x, filters=filters, dtype=dtype)
+
+    # -------------Decoder--------------
+    for i in reversed(range(num_stages)):
+        x = keras.layers.UpSampling2D(
+            size=2, interpolation="bilinear", dtype=dtype
+        )(x)
+        x = keras.layers.concatenate([encoder_blocks[i], x], axis=-1)
+        x = convolution_block(x, filters=filters)
+
+    x = segmentation_head(x)  # Refinement segmentation head.
+
+    # ------------- refined = coarse + residual
+    x = keras.layers.Add(dtype=dtype)(
+        [x_input, x]
+    )  # Add prediction + refinement output
+
+    return keras.models.Model(inputs=base_model.input, outputs=[x])

keras_hub/src/models/basnet/basnet_image_converter.py

@@ -0,0 +1,8 @@
+from keras_hub.src.api_export import keras_hub_export
+from keras_hub.src.layers.preprocessing.image_converter import ImageConverter
+from keras_hub.src.models.basnet.basnet_backbone import BASNetBackbone
+
+
+@keras_hub_export("keras_hub.layers.BASNetImageConverter")
+class BASNetImageConverter(ImageConverter):
+    backbone_cls = BASNetBackbone

keras_hub/src/models/basnet/basnet_preprocessor.py

@@ -0,0 +1,14 @@
+from keras_hub.src.api_export import keras_hub_export
+from keras_hub.src.models.basnet.basnet_backbone import BASNetBackbone
+from keras_hub.src.models.basnet.basnet_image_converter import (
+    BASNetImageConverter,
+)
+from keras_hub.src.models.image_segmenter_preprocessor import (
+    ImageSegmenterPreprocessor,
+)
+
+
+@keras_hub_export("keras_hub.models.BASNetPreprocessor")
+class BASNetPreprocessor(ImageSegmenterPreprocessor):
+    backbone_cls = BASNetBackbone
+    image_converter_cls = BASNetImageConverter

keras_hub/src/models/basnet/basnet_presets.py

@@ -0,0 +1,3 @@
+"""BASNet model preset configurations."""
+
+basnet_presets = {}

keras_hub/src/models/vit/vit_presets.py

@@ -46,4 +46,81 @@ backbone_presets = {
         },
         "kaggle_handle": "kaggle://keras/vit/keras/vit_large_patch16_384_imagenet/1",
     },
+    "vit_base_patch32_384_imagenet": {
+        "metadata": {
+            "description": (
+                "ViT-B32 model pre-trained on the ImageNet 1k dataset with "
+                "image resolution of 384x384 "
+            ),
+            "params": 87528192,
+            "path": "vit",
+        },
+        "kaggle_handle": "kaggle://keras/vit/keras/vit_base_patch32_384_imagenet/1",
+    },
+    "vit_large_patch32_384_imagenet": {
+        "metadata": {
+            "description": (
+                "ViT-L32 model pre-trained on the ImageNet 1k dataset with "
+                "image resolution of 384x384 "
+            ),
+            "params": 305607680,
+            "path": "vit",
+        },
+        "kaggle_handle": "kaggle://keras/vit/keras/vit_large_patch32_384_imagenet/1",
+    },
+    "vit_base_patch16_224_imagenet21k": {
+        "metadata": {
+            "description": (
+                "ViT-B16 backbone pre-trained on the ImageNet 21k dataset with "
+                "image resolution of 224x224 "
+            ),
+            "params": 85798656,
+            "path": "vit",
+        },
+        "kaggle_handle": "kaggle://keras/vit/keras/vit_base_patch16_224_imagenet21k/1",
+    },
+    "vit_base_patch32_224_imagenet21k": {
+        "metadata": {
+            "description": (
+                "ViT-B32 backbone pre-trained on the ImageNet 21k dataset with "
+                "image resolution of 224x224 "
+            ),
+            "params": 87455232,
+            "path": "vit",
+        },
+        "kaggle_handle": "kaggle://keras/vit/keras/vit_base_patch32_224_imagenet21k/1",
+    },
+    "vit_huge_patch14_224_imagenet21k": {
+        "metadata": {
+            "description": (
+                "ViT-H14 backbone pre-trained on the ImageNet 21k dataset with "
+                "image resolution of 224x224 "
+            ),
+            "params": 630764800,
+            "path": "vit",
+        },
+        "kaggle_handle": "kaggle://keras/vit/keras/vit_huge_patch14_224_imagenet21k/1",
+    },
+    "vit_large_patch16_224_imagenet21k": {
+        "metadata": {
+            "description": (
+                "ViT-L16 backbone pre-trained on the ImageNet 21k dataset with "
+                "image resolution of 224x224 "
+            ),
+            "params": 303301632,
+            "path": "vit",
+        },
+        "kaggle_handle": "kaggle://keras/vit/keras/vit_large_patch16_224_imagenet21k/1",
+    },
+    "vit_large_patch32_224_imagenet21k": {
+        "metadata": {
+            "description": (
+                "ViT-L32 backbone pre-trained on the ImageNet 21k dataset with "
+                "image resolution of 224x224 "
+            ),
+            "params": 305510400,
+            "path": "vit",
+        },
+        "kaggle_handle": "kaggle://keras/vit/keras/vit_large_patch32_224_imagenet21k/1",
+    },
 }

keras_hub/src/utils/preset_utils.py

@@ -454,16 +454,6 @@ def load_json(preset, config_file=CONFIG_FILE):
     return config
 
 
-def load_serialized_object(config, **kwargs):
-    # `dtype` in config might be a serialized `DTypePolicy` or `DTypePolicyMap`.
-    # Ensure that `dtype` is properly configured.
-    dtype = kwargs.pop("dtype", None)
-    config = set_dtype_in_config(config, dtype)
-
-    config["config"] = {**config["config"], **kwargs}
-    return keras.saving.deserialize_keras_object(config)
-
-
 def check_config_class(config):
     """Validate a preset is being loaded on the correct class."""
     registered_name = config["registered_name"]
@@ -631,7 +621,7 @@ class KerasPresetLoader(PresetLoader):
         return check_config_class(self.config)
 
     def load_backbone(self, cls, load_weights, **kwargs):
-        backbone = load_serialized_object(self.config, **kwargs)
+        backbone = self._load_serialized_object(self.config, **kwargs)
         if load_weights:
             jax_memory_cleanup(backbone)
             backbone.load_weights(get_file(self.preset, MODEL_WEIGHTS_FILE))
@@ -639,18 +629,18 @@ class KerasPresetLoader(PresetLoader):
 
     def load_tokenizer(self, cls, config_file=TOKENIZER_CONFIG_FILE, **kwargs):
         tokenizer_config = load_json(self.preset, config_file)
-        tokenizer = load_serialized_object(tokenizer_config, **kwargs)
+        tokenizer = self._load_serialized_object(tokenizer_config, **kwargs)
         if hasattr(tokenizer, "load_preset_assets"):
             tokenizer.load_preset_assets(self.preset)
         return tokenizer
 
     def load_audio_converter(self, cls, **kwargs):
         converter_config = load_json(self.preset, AUDIO_CONVERTER_CONFIG_FILE)
-        return load_serialized_object(converter_config, **kwargs)
+        return self._load_serialized_object(converter_config, **kwargs)
 
     def load_image_converter(self, cls, **kwargs):
         converter_config = load_json(self.preset, IMAGE_CONVERTER_CONFIG_FILE)
-        return load_serialized_object(converter_config, **kwargs)
+        return self._load_serialized_object(converter_config, **kwargs)
 
     def load_task(self, cls, load_weights, load_task_weights, **kwargs):
         # If there is no `task.json` or it's for the wrong class delegate to the
@@ -671,7 +661,7 @@ class KerasPresetLoader(PresetLoader):
             backbone_config = task_config["config"]["backbone"]["config"]
             backbone_config = {**backbone_config, **backbone_kwargs}
             task_config["config"]["backbone"]["config"] = backbone_config
-        task = load_serialized_object(task_config, **kwargs)
+        task = self._load_serialized_object(task_config, **kwargs)
         if task.preprocessor and hasattr(
             task.preprocessor, "load_preset_assets"
         ):
@@ -699,11 +689,20 @@ class KerasPresetLoader(PresetLoader):
         if not issubclass(check_config_class(preprocessor_json), cls):
             return super().load_preprocessor(cls, **kwargs)
         # We found a `preprocessing.json` with a complete config for our class.
-        preprocessor = load_serialized_object(preprocessor_json, **kwargs)
+        preprocessor = self._load_serialized_object(preprocessor_json, **kwargs)
         if hasattr(preprocessor, "load_preset_assets"):
             preprocessor.load_preset_assets(self.preset)
         return preprocessor
 
+    def _load_serialized_object(self, config, **kwargs):
+        # `dtype` in config might be a serialized `DTypePolicy` or
+        # `DTypePolicyMap`. Ensure that `dtype` is properly configured.
+        dtype = kwargs.pop("dtype", None)
+        config = set_dtype_in_config(config, dtype)
+
+        config["config"] = {**config["config"], **kwargs}
+        return keras.saving.deserialize_keras_object(config)
+
 
 class KerasPresetSaver:
     def __init__(self, preset_dir):
@@ -787,6 +786,8 @@ class KerasPresetSaver:
         tasks = list_subclasses(Task)
         tasks = filter(lambda x: x.backbone_cls is type(layer), tasks)
         tasks = [task.__base__.__name__ for task in tasks]
+        # Keep task list alphabetical.
+        tasks = sorted(tasks)
 
         keras_version = keras.version() if hasattr(keras, "version") else None
         metadata = {

keras_hub/src/version_utils.py

@@ -1,7 +1,7 @@
 from keras_hub.src.api_export import keras_hub_export
 
 # Unique source of truth for the version number.
-__version__ = "0.19.0.dev202412180349"
+__version__ = "0.19.0.dev202412200346"
 
 
 @keras_hub_export("keras_hub.version")

{keras_hub_nightly-0.19.0.dev202412180349.dist-info → keras_hub_nightly-0.19.0.dev202412200346.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: keras-hub-nightly
-Version: 0.19.0.dev202412180349
+Version: 0.19.0.dev202412200346
 Summary: Industry-strength Natural Language Processing extensions for Keras.
 Home-page: https://github.com/keras-team/keras-hub
 Author: Keras team

{keras_hub_nightly-0.19.0.dev202412180349.dist-info → keras_hub_nightly-0.19.0.dev202412200346.dist-info}/RECORD

@@ -1,15 +1,15 @@
 keras_hub/__init__.py,sha256=QGdXyHgYt6cMUAP1ebxwc6oR86dE0dkMxNy2eOCQtFo,855
 keras_hub/api/__init__.py,sha256=spMxsgqzjpeuC8rY4WP-2kAZ2qwwKRSbFwddXgUjqQE,524
 keras_hub/api/bounding_box/__init__.py,sha256=T8R_X7BPm0et1xaZq8565uJmid7dylsSFSj4V-rGuFQ,1097
-keras_hub/api/layers/__init__.py,sha256=7-EfzihzyE-TinUW078OyLDQnjsADB2binfim8rj6TA,3102
+keras_hub/api/layers/__init__.py,sha256=YO_YLbcxMEboFEgmFkzRf_JfQciQukX2AseOGpWEbDo,3195
 keras_hub/api/metrics/__init__.py,sha256=So8Ec-lOcTzn_UUMmAdzDm8RKkPu2dbRUm2px8gpUEI,381
-keras_hub/api/models/__init__.py,sha256=0VcBlT-Ee_-7KnMW6YmjSYd0fjligqNmIXbL7Vl0GTo,16577
+keras_hub/api/models/__init__.py,sha256=suTcar7FqO5w9nNtalqmfYn7Fs6XmNEGpbojK-gaMEY,16795
 keras_hub/api/samplers/__init__.py,sha256=n-_SEXxr2LNUzK2FqVFN7alsrkx1P_HOVTeLZKeGCdE,730
 keras_hub/api/tokenizers/__init__.py,sha256=mtJgQy1spfQnPAkeLoeinsT_W9iCWHlJXwzcol5W1aU,2524
 keras_hub/api/utils/__init__.py,sha256=Gp1E6gG-RtKQS3PBEQEOz9PQvXkXaJ0ySGMqZ7myN7A,215
 keras_hub/src/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/api_export.py,sha256=9pQZK27JObxWZ96QPLBp1OBsjWigh1iuV6RglPGMRk0,1499
-keras_hub/src/version_utils.py,sha256=2C2DeD2GCwzG2BMJUN9YMKu_w131-hIZyhHDcb9XDBA,222
+keras_hub/src/version_utils.py,sha256=Tk7iPoppgJnEhbF7HzURDrFGXD3R74NzoSPr_MJo9yY,222
 keras_hub/src/bounding_box/__init__.py,sha256=7i6KnGupN4AVivR_dFjQyuuTbI0GkHy8d-aMXeqZdU8,95
 keras_hub/src/bounding_box/converters.py,sha256=UUp1hwegpDZyIo8sh9TLNy1v6JjwmvwzL6wmHFMAtbk,21916
 keras_hub/src/bounding_box/formats.py,sha256=YmskOz2BOSat7NaE__J9VfpSNGPJJR0znSzA4lp8MMI,3868
@@ -85,6 +85,12 @@ keras_hub/src/models/bart/bart_presets.py,sha256=ppk9r_4Sm21XO6F9k3L946rkJBwWSLN
 keras_hub/src/models/bart/bart_seq_2_seq_lm.py,sha256=0r9snJsqqmH8F1_CDQZyFgqLNMYJM8AYFkmqfxUNB1U,19262
 keras_hub/src/models/bart/bart_seq_2_seq_lm_preprocessor.py,sha256=3_e-ULIcm_3DKgt7X7cvyLZEDIEkpu9HdANgH6MjZgg,4373
 keras_hub/src/models/bart/bart_tokenizer.py,sha256=Q7IXmIwXzhPSN427oQRyF9ufoExQGS184Yo_4boaOZo,2811
+keras_hub/src/models/basnet/__init__.py,sha256=4N6XvIUYYJl5xtoaL3_9fawUX_qP3WmTYNEEU7tn8Gw,253
+keras_hub/src/models/basnet/basnet.py,sha256=JA58Q9lmygdSOm5MUaPAlaL6B8XnmqCcRaGrk9c8P3Q,4287
+keras_hub/src/models/basnet/basnet_backbone.py,sha256=t_52WW6jetONS7AnPf9YsiMLDqOjVwjNuayQEv6ZAk4,13503
+keras_hub/src/models/basnet/basnet_image_converter.py,sha256=DwzAwtZeggYw_qyRQ-Abnnm885Wobv3wClxRzOTscI0,342
+keras_hub/src/models/basnet/basnet_preprocessor.py,sha256=uM504utaXODSqR5zpKnopRuaV_l84zCg06RkNoNSKIs,510
+keras_hub/src/models/basnet/basnet_presets.py,sha256=z6tR2q_EvYnUmGfsWIWYfmR_8gvWYPH3QmtpAu_T8f8,63
 keras_hub/src/models/bert/__init__.py,sha256=K_UmCqDgOFFvXgzjXRn5oG0WWi53rAsQMOmUrsiBe1k,245
 keras_hub/src/models/bert/bert_backbone.py,sha256=o8GXUpoKPXLpfFzx5u9wI_3rZJeabPfYJEYSI09Clos,8069
 keras_hub/src/models/bert/bert_masked_lm.py,sha256=8gb1g8h5VFVLmKNEPfLe26z7SOlFnzf9R9okK3rp8AU,4045
@@ -339,7 +345,7 @@ keras_hub/src/models/vit/vit_image_classifier.py,sha256=lMVxiD1_6drx7XQ7P7YzlqnF
 keras_hub/src/models/vit/vit_image_classifier_preprocessor.py,sha256=wu6YcBlXMWB9sKCPvmNdGBZKTLQt_HyHWS6P9nyDwsk,504
 keras_hub/src/models/vit/vit_image_converter.py,sha256=5xVF04BzMcdTDc6aErAYj3_BuGmVd3zoJMcH1ho4T0g,2561
 keras_hub/src/models/vit/vit_layers.py,sha256=s4j3n3qnJnv6W9AdUkNsO3Vsi_BhxEGECYkaLVCU6XY,13238
-keras_hub/src/models/vit/vit_presets.py,sha256=TK7fBafZsM6S8_KCMSdayof8qrB40H8GLYMVYA83HTA,1682
+keras_hub/src/models/vit/vit_presets.py,sha256=1QSyagzonaK4zpJdnjW2UL70T85xGxktsmLdSxcZTjk,4479
 keras_hub/src/models/vit_det/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/models/vit_det/vit_det_backbone.py,sha256=DOZ5J7c1t5PAZ6y0pMmBoQTMOUup7UoUrYVfCs69ltY,7697
 keras_hub/src/models/vit_det/vit_layers.py,sha256=mnwu56chMc6zxmfp_hsLdR7TXYy1_YsWy1KwGX9M5Ic,19840
@@ -387,7 +393,7 @@ keras_hub/src/tokenizers/word_piece_tokenizer_trainer.py,sha256=Zz1SGgArykxBVWnS
 keras_hub/src/utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 keras_hub/src/utils/keras_utils.py,sha256=lrZuC8HL2lmQfbHaS_t1JUyJann_ji2iTYE0Fzos8PU,1969
 keras_hub/src/utils/pipeline_model.py,sha256=jgzB6NQPSl0KOu08N-TazfOnXnUJbZjH2EXXhx25Ftg,9084
-keras_hub/src/utils/preset_utils.py,sha256=P3vqzVb3M-gJmPJDGOe1k0KLmHrhgi4ULeg-L_n5jhM,30976
+keras_hub/src/utils/preset_utils.py,sha256=MFQqOIIWvfYToiUHfpPX0lERmgCkz09bM9L67E44H3s,31115
 keras_hub/src/utils/python_utils.py,sha256=N8nWeO3san4YnGkffRXG3Ix7VEIMTKSN21FX5TuL7G8,202
 keras_hub/src/utils/tensor_utils.py,sha256=YVJesN91bk-OzJXY1mOKBppuY8noBU7zhPQNXPxZVGc,14646
 keras_hub/src/utils/imagenet/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -411,7 +417,7 @@ keras_hub/src/utils/transformers/convert_pali_gemma.py,sha256=B1leeDw96Yvu81hYum
 keras_hub/src/utils/transformers/convert_vit.py,sha256=9SUZ9utNJhW_5cj3acMn9cRy47u2eIcDsrhmzj77o9k,5187
 keras_hub/src/utils/transformers/preset_loader.py,sha256=DgGJXbTSB9Na8FIR-YWWVqQPOFxHwWrGm41EwcS_EFs,3797
 keras_hub/src/utils/transformers/safetensor_utils.py,sha256=CYUHyA4y-B61r7NDnCsFb4t_UmSwZ1k9L-8gzEd6KRg,3339
-keras_hub_nightly-0.19.0.dev202412180349.dist-info/METADATA,sha256=MC4QMK_Qao8jT7qOGSWYpMfxofLZcUxdIn-c-d-czd0,7263
-keras_hub_nightly-0.19.0.dev202412180349.dist-info/WHEEL,sha256=PZUExdf71Ui_so67QXpySuHtCi3-J3wvF4ORK6k_S8U,91
-keras_hub_nightly-0.19.0.dev202412180349.dist-info/top_level.txt,sha256=N4J6piIWBKa38A4uV-CnIopnOEf8mHAbkNXafXm_CuA,10
-keras_hub_nightly-0.19.0.dev202412180349.dist-info/RECORD,,
+keras_hub_nightly-0.19.0.dev202412200346.dist-info/METADATA,sha256=gnzV7FHJ7Cx0ZFRZ-1WimsdR_QeM-z_8z8ycCRAC808,7263
+keras_hub_nightly-0.19.0.dev202412200346.dist-info/WHEEL,sha256=PZUExdf71Ui_so67QXpySuHtCi3-J3wvF4ORK6k_S8U,91
+keras_hub_nightly-0.19.0.dev202412200346.dist-info/top_level.txt,sha256=N4J6piIWBKa38A4uV-CnIopnOEf8mHAbkNXafXm_CuA,10
+keras_hub_nightly-0.19.0.dev202412200346.dist-info/RECORD,,