|
| 1 | +# Copyright 2024 The KerasNLP Authors |
| 2 | +# |
| 3 | +# Licensed under the Apache License, Version 2.0 (the "License"); |
| 4 | +# you may not use this file except in compliance with the License. |
| 5 | +# You may obtain a copy of the License at |
| 6 | +# |
| 7 | +# https://www.apache.org/licenses/LICENSE-2.0 |
| 8 | +# |
| 9 | +# Unless required by applicable law or agreed to in writing, software |
| 10 | +# distributed under the License is distributed on an "AS IS" BASIS, |
| 11 | +# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 12 | +# See the License for the specific language governing permissions and |
| 13 | +# limitations under the License. |
| 14 | +import keras |
| 15 | + |
| 16 | +from keras_nlp.src.api_export import keras_nlp_export |
| 17 | +from keras_nlp.src.models.backbone import Backbone |
| 18 | + |
| 19 | + |
| 20 | +@keras_nlp_export("keras_nlp.models.FeaturePyramidBackbone") |
| 21 | +class FeaturePyramidBackbone(Backbone): |
| 22 | + """A backbone with feature pyramid outputs. |
| 23 | +
|
| 24 | + `FeaturePyramidBackbone` extends `Backbone` with a single `pyramid_outputs` |
| 25 | + property for accessing the feature pyramid outputs of the model. Subclassers |
| 26 | + should set the `pyramid_outputs` property during the model constructor. |
| 27 | +
|
| 28 | + Example: |
| 29 | +
|
| 30 | + ```python |
| 31 | + input_data = np.random.uniform(0, 255, size=(2, 224, 224, 3)) |
| 32 | +
|
| 33 | + # Convert to feature pyramid output format using ResNet. |
| 34 | + backbone = ResNetBackbone.from_preset("resnet50") |
| 35 | + model = keras.Model( |
| 36 | + inputs=backbone.inputs, outputs=backbone.pyramid_outputs |
| 37 | + ) |
| 38 | + model(input_data) # A dict containing the keys ["P2", "P3", "P4", "P5"] |
| 39 | + ``` |
| 40 | + """ |
| 41 | + |
| 42 | + @property |
| 43 | + def pyramid_outputs(self): |
| 44 | + """A dict for feature pyramid outputs. |
| 45 | +
|
| 46 | + The key is a string represents the name of the feature output and the |
| 47 | + value is a `keras.KerasTensor`. A typical feature pyramid has multiple |
| 48 | + levels corresponding to scales such as `["P2", "P3", "P4", "P5"]`. Scale |
| 49 | + `Pn` represents a feature map `2^n` times smaller in width and height |
| 50 | + than the inputs. |
| 51 | + """ |
| 52 | + return getattr(self, "_pyramid_outputs", {}) |
| 53 | + |
| 54 | + @pyramid_outputs.setter |
| 55 | + def pyramid_outputs(self, value): |
| 56 | + if not isinstance(value, dict): |
| 57 | + raise TypeError( |
| 58 | + "`pyramid_outputs` must be a dictionary. " |
| 59 | + f"Received: value={value} of type {type(value)}" |
| 60 | + ) |
| 61 | + for k, v in value.items(): |
| 62 | + if not isinstance(k, str): |
| 63 | + raise TypeError( |
| 64 | + "The key of `pyramid_outputs` must be a string. " |
| 65 | + f"Received: key={k} of type {type(k)}" |
| 66 | + ) |
| 67 | + if not isinstance(v, keras.KerasTensor): |
| 68 | + raise TypeError( |
| 69 | + "The value of `pyramid_outputs` must be a " |
| 70 | + "`keras.KerasTensor`. " |
| 71 | + f"Received: value={v} of type {type(v)}" |
| 72 | + ) |
| 73 | + self._pyramid_outputs = value |
0 commit comments