Add FeaturePyramidBackbone and port weights from timm for ResNetBackbone

keras_nlp/api/models/__init__.py

@@ -106,6 +106,7 @@
 )
 from keras_nlp.src.models.falcon.falcon_preprocessor import FalconPreprocessor
 from keras_nlp.src.models.falcon.falcon_tokenizer import FalconTokenizer
+from keras_nlp.src.models.feature_pyramid_backbone import FeaturePyramidBackbone
 from keras_nlp.src.models.gemma.gemma_backbone import GemmaBackbone
 from keras_nlp.src.models.gemma.gemma_causal_lm import GemmaCausalLM
 from keras_nlp.src.models.gemma.gemma_causal_lm_preprocessor import (

keras_nlp/src/models/backbone.py

@@ -30,6 +30,7 @@
 from keras_nlp.src.utils.preset_utils import save_metadata
 from keras_nlp.src.utils.preset_utils import save_serialized_object
 from keras_nlp.src.utils.python_utils import classproperty
+from keras_nlp.src.utils.timm.convert import load_timm_backbone
 from keras_nlp.src.utils.transformers.convert import load_transformers_backbone
 
 
@@ -204,6 +205,8 @@ class like `keras_nlp.models.Backbone.from_preset()`, or from
 
         if format == "transformers":
             return load_transformers_backbone(cls, preset, load_weights)
+        elif format == "timm":
+            return load_timm_backbone(cls, preset, load_weights, **kwargs)
 
         preset_cls = check_config_class(preset)
         if not issubclass(preset_cls, cls):

keras_nlp/src/models/feature_pyramid_backbone.py

@@ -0,0 +1,73 @@
+# Copyright 2024 The KerasNLP Authors
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     https://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import keras
+
+from keras_nlp.src.api_export import keras_nlp_export
+from keras_nlp.src.models.backbone import Backbone
+
+
+@keras_nlp_export("keras_nlp.models.FeaturePyramidBackbone")
+class FeaturePyramidBackbone(Backbone):
+    """A backbone with feature pyramid outputs.
+
+    `FeaturePyramidBackbone` extends `Backbone` with a single `pyramid_outputs`
+    property for accessing the feature pyramid outputs of the model. Subclassers
+    should set the `pyramid_outputs` property during the model constructor.
+
+    Example:
+
+    ```python
+    input_data = np.random.uniform(0, 255, size=(2, 224, 224, 3))
+
+    # Convert to feature pyramid output format using ResNet.
+    backbone = ResNetBackbone.from_preset("resnet50")
+    model = keras.Model(
+        inputs=backbone.inputs, outputs=backbone.pyramid_outputs
+    )
+    model(input_data)  # A dict containing the keys ["P2", "P3", "P4", "P5"]
+    ```
+    """
+
+    @property
+    def pyramid_outputs(self):
+        """A dict for feature pyramid outputs.
+
+        The key is a string represents the name of the feature output and the
+        value is a `keras.KerasTensor`. A typical feature pyramid has multiple
+        levels corresponding to scales such as `["P2", "P3", "P4", "P5"]`. Scale
+        `Pn` represents a feature map `2^n` times smaller in width and height
+        than the inputs.
+        """
+        return getattr(self, "_pyramid_outputs", {})
+
+    @pyramid_outputs.setter
+    def pyramid_outputs(self, value):
+        if not isinstance(value, dict):
+            raise TypeError(
+                "`pyramid_outputs` must be a dictionary. "
+                f"Received: value={value} of type {type(value)}"
+            )
+        for k, v in value.items():
+            if not isinstance(k, str):
+                raise TypeError(
+                    "The key of `pyramid_outputs` must be a string. "
+                    f"Received: key={k} of type {type(k)}"
+                )
+            if not isinstance(v, keras.KerasTensor):
+                raise TypeError(
+                    "The value of `pyramid_outputs` must be a "
+                    "`keras.KerasTensor`. "
+                    f"Received: value={v} of type {type(v)}"
+                )
+        self._pyramid_outputs = value