keras-hub-nightly 0.16.1.dev202410200345__py3-none-any.whl → 0.19.0.dev202412070351__py3-none-any.whl

keras_hub/src/models/clip/clip_vision_encoder.py

@@ -0,0 +1,159 @@
+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.clip.clip_encoder_block import CLIPEncoderBlock
+from keras_hub.src.models.clip.clip_vision_embedding import CLIPVisionEmbedding
+from keras_hub.src.utils.keras_utils import standardize_data_format
+
+
+@keras_hub_export("keras_hub.models.CLIPVisionEncoder")
+class CLIPVisionEncoder(Backbone):
+    """CLIP vision core network with hyperparameters.
+
+    Args:
+        patch_size: int. The size of each square patch in the input image.
+        hidden_dim: int. The size of the transformer hidden state at the end
+            of each transformer layer.
+        num_layers: int. The number of transformer layers.
+        num_heads: int. The number of attention heads for each transformer.
+        intermediate_dim: int. The output dimension of the first Dense layer in
+            a two-layer feedforward network for each transformer.
+        intermediate_activation: activation function. The activation that
+            is used for the first Dense layer in a two-layer feedforward network
+            for each transformer.
+        intermediate_output_index: optional int. The index of the intermediate
+            output. If specified, the output will become a dictionary with two
+            keys `"sequence_output"` and `"intermediate_output"`.
+        image_shape: tuple. The input shape without the batch size. Defaults to
+            `(224, 224, 3)`.
+        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.
+    """
+
+    def __init__(
+        self,
+        patch_size,
+        hidden_dim,
+        num_layers,
+        num_heads,
+        intermediate_dim,
+        intermediate_activation="quick_gelu",
+        intermediate_output_index=None,
+        image_shape=(224, 224, 3),
+        data_format=None,
+        dtype=None,
+        name=None,
+        **kwargs,
+    ):
+        data_format = standardize_data_format(data_format)
+        if data_format == "channels_last":
+            height, width = image_shape[0], image_shape[1]
+        else:
+            height, width = image_shape[1], image_shape[2]
+        if height != width:
+            raise ValueError(
+                "`CLIPVisionEncoder` expects the height and width to be the "
+                f"same in `image_shape`. Received: image_shape={image_shape}"
+            )
+
+        if (
+            intermediate_output_index is not None
+            and intermediate_output_index < 0
+        ):
+            intermediate_output_index += num_layers
+
+        # `prefix` is used to prevent duplicate name when utilizing multiple
+        # CLIP models within a single model, such as in StableDiffusion3.
+        prefix = str(name) + "_" if name is not None else ""
+
+        # === Layers ===
+        self.embedding = CLIPVisionEmbedding(
+            hidden_dim=hidden_dim,
+            patch_size=patch_size,
+            image_size=height,
+            data_format=data_format,
+            dtype=dtype,
+            name=f"{prefix}embedding",
+        )
+        self.pre_layer_norm = layers.LayerNormalization(
+            epsilon=1e-5, dtype=dtype, name=f"{prefix}pre_layer_norm"
+        )
+        self.encoder_layers = [
+            CLIPEncoderBlock(
+                hidden_dim,
+                num_heads,
+                intermediate_dim,
+                intermediate_activation,
+                use_causal_mask=False,  # `False` in the vision encoder.
+                dtype=dtype,
+                name=f"{prefix}encoder_block_{i}",
+            )
+            for i in range(num_layers)
+        ]
+        self.layer_norm = layers.LayerNormalization(
+            epsilon=1e-5, dtype=dtype, name=f"{prefix}layer_norm"
+        )
+
+        # === Functional Model ===
+        image_input = layers.Input(shape=image_shape, name="images")
+        x = self.embedding(image_input)
+        x = self.pre_layer_norm(x)
+        intermediate_output = None
+        for i, block in enumerate(self.encoder_layers):
+            x = block(x)
+            if i == intermediate_output_index:
+                intermediate_output = x
+        sequence_output = self.layer_norm(x)
+
+        if intermediate_output_index is not None:
+            outputs = {
+                "sequence_output": sequence_output,
+                "intermediate_output": intermediate_output,
+            }
+        else:
+            outputs = sequence_output
+        super().__init__(
+            inputs={"images": image_input},
+            outputs=outputs,
+            dtype=dtype,
+            name=name,
+            **kwargs,
+        )
+
+        # === Config ===
+        self.patch_size = patch_size
+        self.hidden_dim = hidden_dim
+        self.num_layers = num_layers
+        self.num_heads = num_heads
+        self.intermediate_dim = intermediate_dim
+        self.intermediate_activation = intermediate_activation
+        self.intermediate_output_index = intermediate_output_index
+        self.image_shape = image_shape
+
+    def get_config(self):
+        config = super().get_config()
+        config.update(
+            {
+                "patch_size": self.patch_size,
+                "hidden_dim": self.hidden_dim,
+                "num_layers": self.num_layers,
+                "num_heads": self.num_heads,
+                "intermediate_dim": self.intermediate_dim,
+                "intermediate_activation": self.intermediate_activation,
+                "intermediate_output_index": self.intermediate_output_index,
+                "image_shape": self.image_shape,
+            }
+        )
+        return config

keras_hub/src/models/deberta_v3/deberta_v3_presets.py

@@ -8,9 +8,7 @@ backbone_presets = {
                 "Trained on English Wikipedia, BookCorpus and OpenWebText."
             ),
             "params": 70682112,
-            "official_name": "DeBERTaV3",
             "path": "deberta_v3",
-            "model_card": "https://huggingface.co/microsoft/deberta-v3-xsmall",
         },
         "kaggle_handle": "kaggle://keras/deberta_v3/keras/deberta_v3_extra_small_en/2",
     },
@@ -21,9 +19,7 @@ backbone_presets = {
                 "Trained on English Wikipedia, BookCorpus and OpenWebText."
             ),
             "params": 141304320,
-            "official_name": "DeBERTaV3",
             "path": "deberta_v3",
-            "model_card": "https://huggingface.co/microsoft/deberta-v3-small",
         },
         "kaggle_handle": "kaggle://keras/deberta_v3/keras/deberta_v3_small_en/2",
     },
@@ -34,9 +30,7 @@ backbone_presets = {
                 "Trained on English Wikipedia, BookCorpus and OpenWebText."
             ),
             "params": 183831552,
-            "official_name": "DeBERTaV3",
             "path": "deberta_v3",
-            "model_card": "https://huggingface.co/microsoft/deberta-v3-base",
         },
         "kaggle_handle": "kaggle://keras/deberta_v3/keras/deberta_v3_base_en/2",
     },
@@ -47,9 +41,7 @@ backbone_presets = {
                 "Trained on English Wikipedia, BookCorpus and OpenWebText."
             ),
             "params": 434012160,
-            "official_name": "DeBERTaV3",
             "path": "deberta_v3",
-            "model_card": "https://huggingface.co/microsoft/deberta-v3-large",
         },
         "kaggle_handle": "kaggle://keras/deberta_v3/keras/deberta_v3_large_en/2",
     },
@@ -60,9 +52,7 @@ backbone_presets = {
                 "Trained on the 2.5TB multilingual CC100 dataset."
             ),
             "params": 278218752,
-            "official_name": "DeBERTaV3",
             "path": "deberta_v3",
-            "model_card": "https://huggingface.co/microsoft/mdeberta-v3-base",
         },
         "kaggle_handle": "kaggle://keras/deberta_v3/keras/deberta_v3_base_multi/2",
     },

keras_hub/src/models/deeplab_v3/deeplab_v3_presets.py

@@ -9,9 +9,7 @@ backbone_presets = {
                 "which is having categorical accuracy of 90.01 and 0.63 Mean IoU."
             ),
             "params": 39190656,
-            "official_name": "DeepLabV3",
             "path": "deeplab_v3",
-            "model_card": "https://arxiv.org/abs/1802.02611",
         },
         "kaggle_handle": "kaggle://keras/deeplabv3plus/keras/deeplab_v3_plus_resnet50_pascalvoc/3",
     },

keras_hub/src/models/deeplab_v3/deeplab_v3_segmenter.py

@@ -31,9 +31,9 @@ class DeepLabV3ImageSegmenter(ImageSegmenter):
     Load a DeepLabV3 preset with all the 21 class, pretrained segmentation head.
     ```python
     images = np.ones(shape=(1, 96, 96, 3))
-    labels = np.zeros(shape=(1, 96, 96, 1))
+    labels = np.zeros(shape=(1, 96, 96, 2))
     segmenter = keras_hub.models.DeepLabV3ImageSegmenter.from_preset(
-        "deeplabv3_resnet50_pascalvoc",
+        "deeplab_v3_plus_resnet50_pascalvoc",
     )
     segmenter.predict(images)
     ```
@@ -41,12 +41,14 @@ class DeepLabV3ImageSegmenter(ImageSegmenter):
     Specify `num_classes` to load randomly initialized segmentation head.
     ```python
     segmenter = keras_hub.models.DeepLabV3ImageSegmenter.from_preset(
-        "deeplabv3_resnet50_pascalvoc",
+        "deeplab_v3_plus_resnet50_pascalvoc",
         num_classes=2,
     )
+    segmenter.preprocessor.image_size = (96, 96)
     segmenter.fit(images, labels, epochs=3)
     segmenter.predict(images)  # Trained 2 class segmentation.
     ```
+
     Load DeepLabv3+ presets a extension of DeepLabv3 by adding a simple yet
     effective decoder module to refine the segmentation results especially
     along object boundaries.

keras_hub/src/models/densenet/densenet_backbone.py

@@ -29,7 +29,7 @@ class DenseNetBackbone(FeaturePyramidBackbone):
     input_data = np.ones(shape=(8, 224, 224, 3))
 
     # Pretrained backbone
-    model = keras_hub.models.DenseNetBackbone.from_preset("densenet121_imagenet")
+    model = keras_hub.models.DenseNetBackbone.from_preset("densenet_121_imagenet")
     model(input_data)
 
     # Randomly initialized backbone with a custom config

keras_hub/src/models/densenet/densenet_presets.py

@@ -8,9 +8,7 @@ backbone_presets = {
                 "at a 224x224 resolution."
             ),
             "params": 7037504,
-            "official_name": "DenseNet",
             "path": "densenet",
-            "model_card": "https://arxiv.org/abs/1608.06993",
         },
         "kaggle_handle": "kaggle://keras/densenet/keras/densenet_121_imagenet/2",
     },
@@ -21,9 +19,7 @@ backbone_presets = {
                 "at a 224x224 resolution."
             ),
             "params": 12642880,
-            "official_name": "DenseNet",
             "path": "densenet",
-            "model_card": "https://arxiv.org/abs/1608.06993",
         },
         "kaggle_handle": "kaggle://keras/densenet/keras/densenet_169_imagenet/2",
     },
@@ -34,9 +30,7 @@ backbone_presets = {
                 "at a 224x224 resolution."
             ),
             "params": 18321984,
-            "official_name": "DenseNet",
             "path": "densenet",
-            "model_card": "https://arxiv.org/abs/1608.06993",
         },
         "kaggle_handle": "kaggle://keras/densenet/keras/densenet_201_imagenet/2",
     },

keras_hub/src/models/distil_bert/distil_bert_presets.py

@@ -9,9 +9,7 @@ backbone_presets = {
                 "teacher model."
             ),
             "params": 66362880,
-            "official_name": "DistilBERT",
             "path": "distil_bert",
-            "model_card": "https://huggingface.co/distilbert-base-uncased",
         },
         "kaggle_handle": "kaggle://keras/distil_bert/keras/distil_bert_base_en_uncased/2",
     },
@@ -23,9 +21,7 @@ backbone_presets = {
                 "teacher model."
             ),
             "params": 65190912,
-            "official_name": "DistilBERT",
             "path": "distil_bert",
-            "model_card": "https://huggingface.co/distilbert-base-cased",
         },
         "kaggle_handle": "kaggle://keras/distil_bert/keras/distil_bert_base_en/2",
     },
@@ -35,9 +31,7 @@ backbone_presets = {
                 "6-layer DistilBERT model where case is maintained. Trained on Wikipedias of 104 languages"
             ),
             "params": 134734080,
-            "official_name": "DistilBERT",
             "path": "distil_bert",
-            "model_card": "https://huggingface.co/distilbert-base-multilingual-cased",
         },
         "kaggle_handle": "kaggle://keras/distil_bert/keras/distil_bert_base_multi/2",
     },

keras_hub/src/models/efficientnet/__init__.py

@@ -0,0 +1,9 @@
+from keras_hub.src.models.efficientnet.efficientnet_backbone import (
+    EfficientNetBackbone,
+)
+from keras_hub.src.models.efficientnet.efficientnet_presets import (
+    backbone_presets,
+)
+from keras_hub.src.utils.preset_utils import register_presets
+
+register_presets(backbone_presets, EfficientNetBackbone)

keras_hub/src/models/efficientnet/cba.py

@@ -0,0 +1,141 @@
+import keras
+
+BN_AXIS = 3
+
+
+class CBABlock(keras.layers.Layer):
+    """
+    Args:
+        input_filters: int, the number of input filters
+        output_filters: int, the number of output filters
+        kernel_size: default 3, the kernel_size to apply to the expansion phase
+            convolutions
+        strides: default 1, the strides to apply to the expansion phase
+            convolutions
+        data_format: str, channels_last (default) or channels_first, expects
+            tensors to be of shape (N, H, W, C) or (N, C, H, W) respectively
+        batch_norm_momentum: default 0.9, the BatchNormalization momentum
+        batch_norm_epsilon: default 1e-3, the BatchNormalization epsilon
+        activation: default "swish", the activation function used between
+            convolution operations
+        dropout: float, the optional dropout rate to apply before the output
+            convolution, defaults to 0.2
+        nores: bool, default False, forces no residual connection if True,
+            otherwise allows it if False.
+
+    Returns:
+        A tensor representing a feature map, passed through the ConvBNAct
+        block
+
+    Note:
+        Not intended to be used outside of the EfficientNet architecture.
+    """
+
+    def __init__(
+        self,
+        input_filters,
+        output_filters,
+        kernel_size=3,
+        strides=1,
+        data_format="channels_last",
+        batch_norm_momentum=0.9,
+        batch_norm_epsilon=1e-3,
+        activation="swish",
+        dropout=0.2,
+        nores=False,
+        **kwargs
+    ):
+        super().__init__(**kwargs)
+        self.input_filters = input_filters
+        self.output_filters = output_filters
+        self.kernel_size = kernel_size
+        self.strides = strides
+        self.data_format = data_format
+        self.batch_norm_momentum = batch_norm_momentum
+        self.batch_norm_epsilon = batch_norm_epsilon
+        self.activation = activation
+        self.dropout = dropout
+        self.nores = nores
+
+        padding_pixels = kernel_size // 2
+        self.conv1_pad = keras.layers.ZeroPadding2D(
+            padding=(padding_pixels, padding_pixels),
+            name=self.name + "conv_pad",
+        )
+        self.conv1 = keras.layers.Conv2D(
+            filters=self.output_filters,
+            kernel_size=kernel_size,
+            strides=strides,
+            kernel_initializer=self._conv_kernel_initializer(),
+            padding="valid",
+            data_format=data_format,
+            use_bias=False,
+            name=self.name + "conv",
+        )
+        self.bn1 = keras.layers.BatchNormalization(
+            axis=BN_AXIS,
+            momentum=self.batch_norm_momentum,
+            epsilon=self.batch_norm_epsilon,
+            name=self.name + "bn",
+        )
+        self.act = keras.layers.Activation(
+            self.activation, name=self.name + "activation"
+        )
+
+        if self.dropout:
+            self.dropout_layer = keras.layers.Dropout(
+                self.dropout,
+                noise_shape=(None, 1, 1, 1),
+                name=self.name + "drop",
+            )
+
+    def _conv_kernel_initializer(
+        self,
+        scale=2.0,
+        mode="fan_out",
+        distribution="truncated_normal",
+        seed=None,
+    ):
+        return keras.initializers.VarianceScaling(
+            scale=scale, mode=mode, distribution=distribution, seed=seed
+        )
+
+    def build(self, input_shape):
+        if self.name is None:
+            self.name = keras.backend.get_uid("block0")
+
+    def call(self, inputs):
+        x = self.conv1_pad(inputs)
+        x = self.conv1(x)
+        x = self.bn1(x)
+        x = self.act(x)
+
+        # Residual:
+        if (
+            self.strides == 1
+            and self.input_filters == self.output_filters
+            and not self.nores
+        ):
+            if self.dropout:
+                x = self.dropout_layer(x)
+            x = keras.layers.Add(name=self.name + "add")([x, inputs])
+        return x
+
+    def get_config(self):
+        config = super().get_config()
+        config.update(
+            {
+                "input_filters": self.input_filters,
+                "output_filters": self.output_filters,
+                "kernel_size": self.kernel_size,
+                "strides": self.strides,
+                "data_format": self.data_format,
+                "batch_norm_momentum": self.batch_norm_momentum,
+                "batch_norm_epsilon": self.batch_norm_epsilon,
+                "activation": self.activation,
+                "dropout": self.dropout,
+                "nores": self.nores,
+            }
+        )
+
+        return config