keras-hub-nightly 0.23.0.dev202509190415__tar.gz → 0.23.0.dev202509280419__tar.gz

{keras_hub_nightly-0.23.0.dev202509190415 → keras_hub_nightly-0.23.0.dev202509280419}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: keras-hub-nightly
-Version: 0.23.0.dev202509190415
+Version: 0.23.0.dev202509280419
 Summary: Pretrained models for Keras.
 Author-email: Keras team <keras-users@googlegroups.com>
 License-Expression: Apache-2.0

{keras_hub_nightly-0.23.0.dev202509190415 → keras_hub_nightly-0.23.0.dev202509280419}/keras_hub/api/layers/__init__.py

@@ -87,6 +87,9 @@ from keras_hub.src.models.deit.deit_image_converter import (
 from keras_hub.src.models.densenet.densenet_image_converter import (
     DenseNetImageConverter as DenseNetImageConverter,
 )
+from keras_hub.src.models.depth_anything.depth_anything_image_converter import (
+    DepthAnythingImageConverter as DepthAnythingImageConverter,
+)
 from keras_hub.src.models.dinov2.dinov2_image_converter import (
     DINOV2ImageConverter as DINOV2ImageConverter,
 )

{keras_hub_nightly-0.23.0.dev202509190415 → keras_hub_nightly-0.23.0.dev202509280419}/keras_hub/api/models/__init__.py

@@ -166,6 +166,21 @@ from keras_hub.src.models.densenet.densenet_image_classifier import (
 from keras_hub.src.models.densenet.densenet_image_classifier_preprocessor import (
     DenseNetImageClassifierPreprocessor as DenseNetImageClassifierPreprocessor,
 )
+from keras_hub.src.models.depth_anything.depth_anything_backbone import (
+    DepthAnythingBackbone as DepthAnythingBackbone,
+)
+from keras_hub.src.models.depth_anything.depth_anything_depth_estimator import (
+    DepthAnythingDepthEstimator as DepthAnythingDepthEstimator,
+)
+from keras_hub.src.models.depth_anything.depth_anything_depth_estimator_preprocessor import (
+    DepthAnythingDepthEstimatorPreprocessor as DepthAnythingDepthEstimatorPreprocessor,
+)
+from keras_hub.src.models.depth_estimator import (
+    DepthEstimator as DepthEstimator,
+)
+from keras_hub.src.models.depth_estimator_preprocessor import (
+    DepthEstimatorPreprocessor as DepthEstimatorPreprocessor,
+)
 from keras_hub.src.models.dinov2.dinov2_backbone import (
     DINOV2Backbone as DINOV2Backbone,
 )
@@ -512,6 +527,15 @@ from keras_hub.src.models.qwen3.qwen3_causal_lm_preprocessor import (
 from keras_hub.src.models.qwen3.qwen3_tokenizer import (
     Qwen3Tokenizer as Qwen3Tokenizer,
 )
+from keras_hub.src.models.qwen3_moe.qwen3_moe_backbone import (
+    Qwen3MoeBackbone as Qwen3MoeBackbone,
+)
+from keras_hub.src.models.qwen3_moe.qwen3_moe_causal_lm import (
+    Qwen3MoeCausalLM as Qwen3MoeCausalLM,
+)
+from keras_hub.src.models.qwen3_moe.qwen3_moe_causal_lm_preprocessor import (
+    Qwen3MoeCausalLMPreprocessor as Qwen3MoeCausalLMPreprocessor,
+)
 from keras_hub.src.models.qwen_moe.qwen_moe_backbone import (
     QwenMoeBackbone as QwenMoeBackbone,
 )

{keras_hub_nightly-0.23.0.dev202509190415 → keras_hub_nightly-0.23.0.dev202509280419}/keras_hub/api/tokenizers/__init__.py

@@ -78,6 +78,9 @@ from keras_hub.src.models.qwen.qwen_tokenizer import (
 from keras_hub.src.models.qwen.qwen_tokenizer import (
     QwenTokenizer as QwenTokenizer,
 )
+from keras_hub.src.models.qwen3_moe.qwen3_moe_tokenizer import (
+    Qwen3MoeTokenizer as Qwen3MoeTokenizer,
+)
 from keras_hub.src.models.qwen_moe.qwen_moe_tokenizer import (
     QwenMoeTokenizer as QwenMoeTokenizer,
 )

keras_hub_nightly-0.23.0.dev202509280419/keras_hub/src/models/depth_anything/__init__.py

@@ -0,0 +1,9 @@
+from keras_hub.src.models.depth_anything.depth_anything_backbone import (
+    DepthAnythingBackbone,
+)
+from keras_hub.src.models.depth_anything.depth_anything_presets import (
+    backbone_presets,
+)
+from keras_hub.src.utils.preset_utils import register_presets
+
+register_presets(backbone_presets, DepthAnythingBackbone)

keras_hub_nightly-0.23.0.dev202509280419/keras_hub/src/models/depth_anything/depth_anything_backbone.py

@@ -0,0 +1,232 @@
+import keras
+from keras import layers
+
+from keras_hub.src.api_export import keras_hub_export
+from keras_hub.src.models.backbone import Backbone
+from keras_hub.src.models.depth_anything.depth_anything_layers import (
+    DepthAnythingDepthEstimationHead,
+)
+from keras_hub.src.models.depth_anything.depth_anything_layers import (
+    DepthAnythingNeck,
+)
+from keras_hub.src.models.dinov2 import DINOV2Backbone
+from keras_hub.src.utils.keras_utils import standardize_data_format
+
+
+@keras_hub_export("keras_hub.models.DepthAnythingBackbone")
+class DepthAnythingBackbone(Backbone):
+    """DepthAnything core network with hyperparameters.
+
+    DepthAnything offers a powerful monocular depth estimation as described in
+    [Depth Anything V2](https://arxiv.org/abs/2406.09414).
+
+    The default constructor gives a fully customizable, randomly initialized
+    DepthAnything model with any number of layers, heads, and embedding
+    dimensions by providing the DINOV2 as the `image_encoder`. To load preset
+    architectures and weights, use the `from_preset` constructor.
+
+    Args:
+        image_encoder: The DINOV2 image encoder for encoding the input images.
+        reassemble_factors: List of float. The reassemble factor for each
+            feature map from the image encoder. The length of the list must be
+            equal to the number of feature maps from the image encoder.
+        neck_hidden_dims: int. The size of the neck hidden state.
+        fusion_hidden_dim: int. The size of the fusion hidden state.
+        head_hidden_dim: int. The size of the neck hidden state.
+        head_in_index: int. The index to select the feature from the neck
+            features as the input to the head.
+        feature_keys: List of string. The keys to select the feature maps from
+            the image encoder. If `None`, all feature maps from the image
+            encoder will be used. Defaults to `None`.
+        data_format: `None` or str. If specified, either `"channels_last"` or
+            `"channels_first"`. The ordering of the dimensions in the
+            inputs. `"channels_last"` corresponds to inputs with shape
+            `(batch_size, height, width, channels)`
+            while `"channels_first"` corresponds to inputs with shape
+            `(batch_size, channels, height, width)`. It defaults to the
+            `image_data_format` value found in your Keras config file at
+            `~/.keras/keras.json`. If you never set it, then it will be
+            `"channels_last"`.
+        dtype: string or `keras.mixed_precision.DTypePolicy`. The dtype to use
+            for the models computations and weights. Note that some
+            computations, such as softmax and layer normalization will always
+            be done a float32 precision regardless of dtype.
+
+    Example:
+    ```python
+    # Pretrained DepthAnything model.
+    input_data = {
+        "images": np.ones(shape=(1, 518, 518, 3), dtype="float32"),
+    }
+    model = keras_hub.models.DepthAnythingBackbone.from_preset(
+        "depth_anything_v2_small"
+    )
+    model(input_data)
+
+    # Pretrained DepthAnything model with custom image shape.
+    input_data = {
+        "images": np.ones(shape=(1, 224, 224, 3), dtype="float32"),
+    }
+    model = keras_hub.models.DepthAnythingBackbone.from_preset(
+        "depth_anything_v2_small", image_shape=(224, 224, 3)
+    )
+    model(input_data)
+
+    # Randomly initialized DepthAnything model with custom config.
+    image_encoder = keras_hub.models.DINOV2Backbone(
+        patch_size=14,
+        num_layers=4,
+        hidden_dim=32,
+        num_heads=2,
+        intermediate_dim=128,
+        image_shape=(224, 224, 3),
+        position_embedding_shape=(518, 518),
+    )
+    model = keras_hub.models.DepthAnythingBackbone(
+        image_encoder=image_encoder,
+        reassemble_factors=[4, 2, 1, 0.5],
+        neck_hidden_dims=[16, 32, 64, 128],
+        fusion_hidden_dim=128,
+        head_hidden_dim=16,
+        head_in_index=-1,
+        feature_keys=["Stage1", "Stage2", "Stage3", "Stage4"],
+    )
+    model(input_data)
+    ```
+    """
+
+    def __init__(
+        self,
+        image_encoder,
+        reassemble_factors,
+        neck_hidden_dims,
+        fusion_hidden_dim,
+        head_hidden_dim,
+        head_in_index,
+        feature_keys=None,
+        data_format=None,
+        dtype=None,
+        **kwargs,
+    ):
+        if not isinstance(image_encoder, DINOV2Backbone):
+            raise ValueError(
+                "`image_encoder` must be a `DINOV2Backbone`. "
+                f"Received image_encoder={image_encoder} "
+                f"(of type {type(image_encoder)})."
+            )
+        if feature_keys is not None:
+            feature_keys = [str(key) for key in feature_keys]
+            for key in feature_keys:
+                if key not in image_encoder.pyramid_outputs:
+                    raise ValueError(
+                        "All `feature_keys` must be in "
+                        "`image_encoder.pyramid_outputs`. "
+                        f"Received feature_keys={feature_keys}, but "
+                        "`image_encoder.pyramid_outputs` contains "
+                        f"{list(image_encoder.pyramid_outputs.keys())}."
+                    )
+        else:
+            feature_keys = list(image_encoder.pyramid_outputs.keys())
+        if len(reassemble_factors) != len(feature_keys):
+            raise ValueError(
+                "The length of `reassemble_factors` must be equal to the "
+                "length of `feature_keys`. "
+                f"Received len(reassemble_factors)={len(reassemble_factors)}, "
+                f"len(feature_keys)={len(feature_keys)}."
+            )
+        data_format = standardize_data_format(data_format)
+        patch_size = image_encoder.patch_size
+        backbone_hidden_dim = image_encoder.hidden_dim
+        image_shape = image_encoder.image_shape
+        if data_format == "channels_last":
+            image_size = (image_shape[0], image_shape[1])
+        else:
+            image_size = (image_shape[1], image_shape[2])
+
+        # === Layers ===
+        pyramid_outputs = {
+            key: value
+            for key, value in image_encoder.pyramid_outputs.items()
+            if key in feature_keys
+        }
+        self.feature_extractor = keras.Model(
+            inputs=image_encoder.inputs,
+            outputs=pyramid_outputs,
+        )
+        self.feature_extractor.dtype_policy = image_encoder.dtype_policy
+        self.neck = DepthAnythingNeck(
+            patch_size=patch_size,
+            image_size=image_size,
+            backbone_hidden_dim=backbone_hidden_dim,
+            neck_hidden_dims=neck_hidden_dims,
+            reassemble_factors=reassemble_factors,
+            fusion_hidden_dim=fusion_hidden_dim,
+            num_cls_tokens=1,
+            num_register_tokens=image_encoder.num_register_tokens,
+            data_format=data_format,
+            dtype=dtype,
+            name="neck",
+        )
+        self.head = DepthAnythingDepthEstimationHead(
+            patch_size=patch_size,
+            patch_height=image_size[0] // patch_size,
+            patch_width=image_size[1] // patch_size,
+            fusion_hidden_dim=fusion_hidden_dim,
+            head_hidden_dim=head_hidden_dim,
+            head_in_index=head_in_index,
+            data_format=data_format,
+            dtype=dtype,
+            name="head",
+        )
+
+        # === Functional Model ===
+        image_input = layers.Input(shape=image_shape, name="images")
+        features = self.feature_extractor(image_input)
+        features = self.neck(list(features.values()))
+        depth_output = self.head(features)
+        super().__init__(
+            inputs=image_input,
+            outputs=depth_output,
+            dtype=dtype,
+            **kwargs,
+        )
+
+        # === Config ===
+        self.image_encoder = image_encoder
+        self.reassemble_factors = reassemble_factors
+        self.neck_hidden_dims = neck_hidden_dims
+        self.fusion_hidden_dim = fusion_hidden_dim
+        self.head_hidden_dim = head_hidden_dim
+        self.head_in_index = head_in_index
+        self.feature_keys = feature_keys
+
+    def get_config(self):
+        config = super().get_config()
+        config.update(
+            {
+                "image_encoder": layers.serialize(self.image_encoder),
+                "reassemble_factors": self.reassemble_factors,
+                "neck_hidden_dims": self.neck_hidden_dims,
+                "fusion_hidden_dim": self.fusion_hidden_dim,
+                "head_hidden_dim": self.head_hidden_dim,
+                "head_in_index": self.head_in_index,
+                "feature_keys": self.feature_keys,
+            }
+        )
+        return config
+
+    @classmethod
+    def from_config(cls, config, custom_objects=None):
+        config = config.copy()
+
+        # Propagate `dtype` to `image_encoder` if needed.
+        if "dtype" in config and config["dtype"] is not None:
+            dtype_config = config["dtype"]
+            if "dtype" not in config["image_encoder"]["config"]:
+                config["image_encoder"]["config"]["dtype"] = dtype_config
+
+        # We expect submodels to be instantiated.
+        config["image_encoder"] = layers.deserialize(
+            config["image_encoder"], custom_objects=custom_objects
+        )
+        return cls(**config)

keras_hub_nightly-0.23.0.dev202509280419/keras_hub/src/models/depth_anything/depth_anything_depth_estimator.py

@@ -0,0 +1,70 @@
+import keras
+
+from keras_hub.src.api_export import keras_hub_export
+from keras_hub.src.models.depth_anything.depth_anything_backbone import (
+    DepthAnythingBackbone,
+)
+from keras_hub.src.models.depth_anything.depth_anything_depth_estimator_preprocessor import (  # noqa: E501
+    DepthAnythingDepthEstimatorPreprocessor,
+)
+from keras_hub.src.models.depth_anything.depth_anything_loss import (
+    DepthAnythingLoss,
+)
+from keras_hub.src.models.depth_estimator import DepthEstimator
+
+
+@keras_hub_export("keras_hub.models.DepthAnythingDepthEstimator")
+class DepthAnythingDepthEstimator(DepthEstimator):
+    backbone_cls = DepthAnythingBackbone
+    preprocessor_cls = DepthAnythingDepthEstimatorPreprocessor
+
+    def compile(
+        self,
+        optimizer="auto",
+        loss="auto",
+        *,
+        metrics="auto",
+        **kwargs,
+    ):
+        """Configures the `DepthEstimator` task for training.
+
+        The `DepthEstimator` task extends the default compilation signature of
+        `keras.Model.compile` with defaults for `optimizer`, `loss`, and
+        `metrics`. 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 the given model and task. 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"`, where a `DepthAnythingLoss` loss will be
+                applied for the depth estimation task. See
+                `keras.Model.compile` and `keras.losses` for more info on
+                possible `loss` values.
+            metrics: `"auto"`, or a dict of metrics to be evaluated by
+                the model during training and testing. Defaults to `"auto"`,
+                where a `keras.metrics.RootMeanSquaredError` 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 optimizer == "auto":
+            optimizer = keras.optimizers.AdamW(5e-5)
+        if loss == "auto":
+            loss = {
+                "depths": DepthAnythingLoss(
+                    min_depth=self.min_depth, max_depth=self.max_depth
+                )
+            }
+        if metrics == "auto":
+            metrics = {"depths": keras.metrics.RootMeanSquaredError()}
+        super().compile(
+            optimizer=optimizer,
+            loss=loss,
+            metrics=metrics,
+            **kwargs,
+        )

keras_hub_nightly-0.23.0.dev202509280419/keras_hub/src/models/depth_anything/depth_anything_depth_estimator_preprocessor.py

@@ -0,0 +1,16 @@
+from keras_hub.src.api_export import keras_hub_export
+from keras_hub.src.models.depth_anything.depth_anything_backbone import (
+    DepthAnythingBackbone,
+)
+from keras_hub.src.models.depth_anything.depth_anything_image_converter import (
+    DepthAnythingImageConverter,
+)
+from keras_hub.src.models.depth_estimator_preprocessor import (
+    DepthEstimatorPreprocessor,
+)
+
+
+@keras_hub_export("keras_hub.models.DepthAnythingDepthEstimatorPreprocessor")
+class DepthAnythingDepthEstimatorPreprocessor(DepthEstimatorPreprocessor):
+    backbone_cls = DepthAnythingBackbone
+    image_converter_cls = DepthAnythingImageConverter

keras_hub_nightly-0.23.0.dev202509280419/keras_hub/src/models/depth_anything/depth_anything_image_converter.py

@@ -0,0 +1,10 @@
+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.depth_anything.depth_anything_backbone import (
+    DepthAnythingBackbone,
+)
+
+
+@keras_hub_export("keras_hub.layers.DepthAnythingImageConverter")
+class DepthAnythingImageConverter(ImageConverter):
+    backbone_cls = DepthAnythingBackbone