bioimageio.core.model_adapters

 36class ModelAdapter(ABC):
 37    """
 38    Represents model *without* any preprocessing or postprocessing.
 39
 40    ```
 41    from bioimageio.core import load_description
 42
 43    model = load_description(...)
 44
 45    # option 1:
 46    adapter = ModelAdapter.create(model)
 47    adapter.forward(...)
 48    adapter.unload()
 49
 50    # option 2:
 51    with ModelAdapter.create(model) as adapter:
 52        adapter.forward(...)
 53    ```
 54    """
 55
 56    def __init__(self, model_description: AnyModelDescr):
 57        super().__init__()
 58        self._model_descr = model_description
 59        self._input_ids = get_member_ids(model_description.inputs)
 60        self._output_ids = get_member_ids(model_description.outputs)
 61        self._input_axes = [
 62            tuple(a.id for a in get_axes_infos(t)) for t in model_description.inputs
 63        ]
 64        self._output_axes = [
 65            tuple(a.id for a in get_axes_infos(t)) for t in model_description.outputs
 66        ]
 67        if isinstance(model_description, v0_4.ModelDescr):
 68            self._input_is_optional = [False] * len(model_description.inputs)
 69        else:
 70            self._input_is_optional = [ipt.optional for ipt in model_description.inputs]
 71
 72    @final
 73    @classmethod
 74    def create(
 75        cls,
 76        model_description: Union[v0_4.ModelDescr, v0_5.ModelDescr],
 77        *,
 78        devices: Optional[Sequence[str]] = None,
 79        weight_format_priority_order: Optional[Sequence[SupportedWeightsFormat]] = None,
 80    ):
 81        """
 82        Creates model adapter based on the passed spec
 83        Note: All specific adapters should happen inside this function to prevent different framework
 84        initializations interfering with each other
 85        """
 86        if not isinstance(model_description, (v0_4.ModelDescr, v0_5.ModelDescr)):
 87            raise TypeError(
 88                f"expected v0_4.ModelDescr or v0_5.ModelDescr, but got {type(model_description)}"
 89            )
 90
 91        weights = model_description.weights
 92        errors: List[Exception] = []
 93        weight_format_priority_order = (
 94            DEFAULT_WEIGHT_FORMAT_PRIORITY_ORDER
 95            if weight_format_priority_order is None
 96            else weight_format_priority_order
 97        )
 98        # limit weight formats to the ones present
 99        weight_format_priority_order_present: Sequence[SupportedWeightsFormat] = [
100            w for w in weight_format_priority_order if getattr(weights, w) is not None
101        ]
102        if not weight_format_priority_order_present:
103            raise ValueError(
104                f"None of the specified weight formats ({weight_format_priority_order}) is present ({weight_format_priority_order_present})"
105            )
106
107        for wf in weight_format_priority_order_present:
108            if wf == "pytorch_state_dict":
109                assert weights.pytorch_state_dict is not None
110                try:
111                    from .pytorch_backend import PytorchModelAdapter
112
113                    return PytorchModelAdapter(
114                        model_description=model_description, devices=devices
115                    )
116                except Exception as e:
117                    errors.append(e)
118            elif wf == "tensorflow_saved_model_bundle":
119                assert weights.tensorflow_saved_model_bundle is not None
120                try:
121                    from .tensorflow_backend import create_tf_model_adapter
122
123                    return create_tf_model_adapter(
124                        model_description=model_description, devices=devices
125                    )
126                except Exception as e:
127                    errors.append(e)
128            elif wf == "onnx":
129                assert weights.onnx is not None
130                try:
131                    from .onnx_backend import ONNXModelAdapter
132
133                    return ONNXModelAdapter(
134                        model_description=model_description, devices=devices
135                    )
136                except Exception as e:
137                    errors.append(e)
138            elif wf == "torchscript":
139                assert weights.torchscript is not None
140                try:
141                    from .torchscript_backend import TorchscriptModelAdapter
142
143                    return TorchscriptModelAdapter(
144                        model_description=model_description, devices=devices
145                    )
146                except Exception as e:
147                    errors.append(e)
148            elif wf == "keras_hdf5":
149                assert weights.keras_hdf5 is not None
150                # keras can either be installed as a separate package or used as part of tensorflow
151                # we try to first import the keras model adapter using the separate package and,
152                # if it is not available, try to load the one using tf
153                try:
154                    try:
155                        from .keras_backend import KerasModelAdapter
156                    except Exception:
157                        from .tensorflow_backend import KerasModelAdapter
158
159                    return KerasModelAdapter(
160                        model_description=model_description, devices=devices
161                    )
162                except Exception as e:
163                    errors.append(e)
164            else:
165                assert_never(wf)
166
167        assert errors
168        if len(weight_format_priority_order) == 1:
169            assert len(errors) == 1
170            raise errors[0]
171
172        else:
173            msg = (
174                "None of the weight format specific model adapters could be created"
175                + " in this environment."
176            )
177            if sys.version_info[:2] >= (3, 11):
178                raise ExceptionGroup(msg, errors)
179            else:
180                raise ValueError(msg) from Exception(errors)
181
182    @final
183    def load(self, *, devices: Optional[Sequence[str]] = None) -> None:
184        warnings.warn("Deprecated. ModelAdapter is loaded on initialization")
185
186    def forward(
187        self, input_sample: Union[Sample, SampleBlock, SampleBlockWithOrigin]
188    ) -> Sample:
189        """
190        Run forward pass of model to get model predictions
191
192        Note: sample id and stample stat attributes are passed through
193        """
194        unexpected = [mid for mid in input_sample.members if mid not in self._input_ids]
195        if unexpected:
196            warnings.warn(f"Got unexpected input tensor IDs: {unexpected}")
197
198        input_arrays = [
199            (
200                None
201                if (a := input_sample.members.get(in_id)) is None
202                else a.transpose(in_order).data.data
203            )
204            for in_id, in_order in zip(self._input_ids, self._input_axes)
205        ]
206        output_arrays = self._forward_impl(input_arrays)
207        assert len(output_arrays) <= len(self._output_ids)
208        output_tensors = [
209            None if a is None else Tensor(a, dims=d)
210            for a, d in zip(output_arrays, self._output_axes)
211        ]
212        return Sample(
213            members={
214                tid: out
215                for tid, out in zip(
216                    self._output_ids,
217                    output_tensors,
218                )
219                if out is not None
220            },
221            stat=input_sample.stat,
222            id=(
223                input_sample.id
224                if isinstance(input_sample, Sample)
225                else input_sample.sample_id
226            ),
227        )
228
229    @abstractmethod
230    def _forward_impl(
231        self, input_arrays: Sequence[Optional[NDArray[Any]]]
232    ) -> Union[List[Optional[NDArray[Any]]], Tuple[Optional[NDArray[Any]]]]:
233        """framework specific forward implementation"""
234
235    @abstractmethod
236    def unload(self):
237        """
238        Unload model from any devices, freeing their memory.
239        The moder adapter should be considered unusable afterwards.
240        """
241
242    def _get_input_args_numpy(self, input_sample: Sample):
243        """helper to extract tensor args as transposed numpy arrays"""

 72    @final
 73    @classmethod
 74    def create(
 75        cls,
 76        model_description: Union[v0_4.ModelDescr, v0_5.ModelDescr],
 77        *,
 78        devices: Optional[Sequence[str]] = None,
 79        weight_format_priority_order: Optional[Sequence[SupportedWeightsFormat]] = None,
 80    ):
 81        """
 82        Creates model adapter based on the passed spec
 83        Note: All specific adapters should happen inside this function to prevent different framework
 84        initializations interfering with each other
 85        """
 86        if not isinstance(model_description, (v0_4.ModelDescr, v0_5.ModelDescr)):
 87            raise TypeError(
 88                f"expected v0_4.ModelDescr or v0_5.ModelDescr, but got {type(model_description)}"
 89            )
 90
 91        weights = model_description.weights
 92        errors: List[Exception] = []
 93        weight_format_priority_order = (
 94            DEFAULT_WEIGHT_FORMAT_PRIORITY_ORDER
 95            if weight_format_priority_order is None
 96            else weight_format_priority_order
 97        )
 98        # limit weight formats to the ones present
 99        weight_format_priority_order_present: Sequence[SupportedWeightsFormat] = [
100            w for w in weight_format_priority_order if getattr(weights, w) is not None
101        ]
102        if not weight_format_priority_order_present:
103            raise ValueError(
104                f"None of the specified weight formats ({weight_format_priority_order}) is present ({weight_format_priority_order_present})"
105            )
106
107        for wf in weight_format_priority_order_present:
108            if wf == "pytorch_state_dict":
109                assert weights.pytorch_state_dict is not None
110                try:
111                    from .pytorch_backend import PytorchModelAdapter
112
113                    return PytorchModelAdapter(
114                        model_description=model_description, devices=devices
115                    )
116                except Exception as e:
117                    errors.append(e)
118            elif wf == "tensorflow_saved_model_bundle":
119                assert weights.tensorflow_saved_model_bundle is not None
120                try:
121                    from .tensorflow_backend import create_tf_model_adapter
122
123                    return create_tf_model_adapter(
124                        model_description=model_description, devices=devices
125                    )
126                except Exception as e:
127                    errors.append(e)
128            elif wf == "onnx":
129                assert weights.onnx is not None
130                try:
131                    from .onnx_backend import ONNXModelAdapter
132
133                    return ONNXModelAdapter(
134                        model_description=model_description, devices=devices
135                    )
136                except Exception as e:
137                    errors.append(e)
138            elif wf == "torchscript":
139                assert weights.torchscript is not None
140                try:
141                    from .torchscript_backend import TorchscriptModelAdapter
142
143                    return TorchscriptModelAdapter(
144                        model_description=model_description, devices=devices
145                    )
146                except Exception as e:
147                    errors.append(e)
148            elif wf == "keras_hdf5":
149                assert weights.keras_hdf5 is not None
150                # keras can either be installed as a separate package or used as part of tensorflow
151                # we try to first import the keras model adapter using the separate package and,
152                # if it is not available, try to load the one using tf
153                try:
154                    try:
155                        from .keras_backend import KerasModelAdapter
156                    except Exception:
157                        from .tensorflow_backend import KerasModelAdapter
158
159                    return KerasModelAdapter(
160                        model_description=model_description, devices=devices
161                    )
162                except Exception as e:
163                    errors.append(e)
164            else:
165                assert_never(wf)
166
167        assert errors
168        if len(weight_format_priority_order) == 1:
169            assert len(errors) == 1
170            raise errors[0]
171
172        else:
173            msg = (
174                "None of the weight format specific model adapters could be created"
175                + " in this environment."
176            )
177            if sys.version_info[:2] >= (3, 11):
178                raise ExceptionGroup(msg, errors)
179            else:
180                raise ValueError(msg) from Exception(errors)

19def get_weight_formats() -> List[str]:
20    """
21    Return list of supported weight types
22    """
23    return list(DEFAULT_WEIGHT_FORMAT_PRIORITY_ORDER)