File size: 16,160 Bytes
f2ae8ec
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
# Copyright (c) MONAI Consortium
# 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
#     http://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.

from __future__ import annotations

from typing import TYPE_CHECKING, Any, Callable, Iterable, Sequence

import numpy as np
import torch
from monai.engines.evaluator import SupervisedEvaluator
from monai.engines.utils import IterationEvents, default_metric_cmp_fn, default_prepare_batch
from monai.inferers import Inferer, SimpleInferer
from monai.transforms import Transform, reset_ops_id
from monai.utils import ForwardMode, IgniteInfo, RankFilter, min_version, optional_import
from monai.utils.enums import CommonKeys as Keys
from torch.utils.data import DataLoader

rearrange, _ = optional_import("einops", name="rearrange")

if TYPE_CHECKING:
    from ignite.engine import Engine, EventEnum
    from ignite.metrics import Metric
else:
    Engine, _ = optional_import("ignite.engine", IgniteInfo.OPT_IMPORT_VERSION, min_version, "Engine")
    Metric, _ = optional_import("ignite.metrics", IgniteInfo.OPT_IMPORT_VERSION, min_version, "Metric")
    EventEnum, _ = optional_import("ignite.engine", IgniteInfo.OPT_IMPORT_VERSION, min_version, "EventEnum")

__all__ = ["Vista3dEvaluator"]


class Vista3dEvaluator(SupervisedEvaluator):
    """
    Supervised detection evaluation method with image and label, inherits from ``SupervisedEvaluator`` and ``Workflow``.
    Args:
        device: an object representing the device on which to run.
        val_data_loader: Ignite engine use data_loader to run, must be Iterable, typically be torch.DataLoader.
        network: detector to evaluate in the evaluator, should be regular PyTorch `torch.nn.Module`.
        epoch_length: number of iterations for one epoch, default to `len(val_data_loader)`.
        non_blocking: if True and this copy is between CPU and GPU, the copy may occur asynchronously
            with respect to the host. For other cases, this argument has no effect.
        prepare_batch: function to parse expected data (usually `image`, `label` and other network args)
            from `engine.state.batch` for every iteration, for more details please refer to:
            https://pytorch.org/ignite/generated/ignite.engine.create_supervised_trainer.html.
        iteration_update: the callable function for every iteration, expect to accept `engine`
            and `engine.state.batch` as inputs, return data will be stored in `engine.state.output`.
            if not provided, use `self._iteration()` instead. for more details please refer to:
            https://pytorch.org/ignite/generated/ignite.engine.engine.Engine.html.
        inferer: inference method that execute model forward on input data, like: SlidingWindow, etc.
        postprocessing: execute additional transformation for the model output data.
            Typically, several Tensor based transforms composed by `Compose`.
        key_val_metric: compute metric when every iteration completed, and save average value to
            engine.state.metrics when epoch completed. key_val_metric is the main metric to compare and save the
            checkpoint into files.
        additional_metrics: more Ignite metrics that also attach to Ignite Engine.
        metric_cmp_fn: function to compare current key metric with previous best key metric value,
            it must accept 2 args (current_metric, previous_best) and return a bool result: if `True`, will update
            `best_metric` and `best_metric_epoch` with current metric and epoch, default to `greater than`.
        val_handlers: every handler is a set of Ignite Event-Handlers, must have `attach` function, like:
            CheckpointHandler, StatsHandler, etc.
        amp: whether to enable auto-mixed-precision evaluation, default is False.
        mode: model forward mode during evaluation, should be 'eval' or 'train',
            which maps to `model.eval()` or `model.train()`, default to 'eval'.
        event_names: additional custom ignite events that will register to the engine.
            new events can be a list of str or `ignite.engine.events.EventEnum`.
        event_to_attr: a dictionary to map an event to a state attribute, then add to `engine.state`.
            for more details, check: https://pytorch.org/ignite/generated/ignite.engine.engine.Engine.html
            #ignite.engine.engine.Engine.register_events.
        decollate: whether to decollate the batch-first data to a list of data after model computation,
            recommend `decollate=True` when `postprocessing` uses components from `monai.transforms`.
            default to `True`.
        to_kwargs: dict of other args for `prepare_batch` API when converting the input data, except for
            `device`, `non_blocking`.
        amp_kwargs: dict of the args for `torch.amp.autocast()` API, for more details:
            https://pytorch.org/docs/stable/amp.html#torch.amp.autocast.
    """

    def __init__(
        self,
        device: torch.device,
        val_data_loader: Iterable | DataLoader,
        network: torch.nn.Module,
        epoch_length: int | None = None,
        non_blocking: bool = False,
        prepare_batch: Callable = default_prepare_batch,
        iteration_update: Callable[[Engine, Any], Any] | None = None,
        inferer: Inferer | None = None,
        postprocessing: Transform | None = None,
        key_val_metric: dict[str, Metric] | None = None,
        additional_metrics: dict[str, Metric] | None = None,
        metric_cmp_fn: Callable = default_metric_cmp_fn,
        val_handlers: Sequence | None = None,
        amp: bool = False,
        mode: ForwardMode | str = ForwardMode.EVAL,
        event_names: list[str | EventEnum | type[EventEnum]] | None = None,
        event_to_attr: dict | None = None,
        decollate: bool = True,
        to_kwargs: dict | None = None,
        amp_kwargs: dict | None = None,
        hyper_kwargs: dict | None = None,
    ) -> None:
        super().__init__(
            device=device,
            val_data_loader=val_data_loader,
            network=network,
            epoch_length=epoch_length,
            non_blocking=non_blocking,
            prepare_batch=prepare_batch,
            iteration_update=iteration_update,
            postprocessing=postprocessing,
            key_val_metric=key_val_metric,
            additional_metrics=additional_metrics,
            metric_cmp_fn=metric_cmp_fn,
            val_handlers=val_handlers,
            amp=amp,
            mode=mode,
            event_names=event_names,
            event_to_attr=event_to_attr,
            decollate=decollate,
            to_kwargs=to_kwargs,
            amp_kwargs=amp_kwargs,
        )

        self.network = network
        self.device = device
        self.inferer = SimpleInferer() if inferer is None else inferer
        self.hyper_kwargs = hyper_kwargs
        self.logger.addFilter(RankFilter())

    def transform_points(self, point, affine):
        """transform point to the coordinates of the transformed image
        point: numpy array [bs, N, 3]
        """
        bs, n = point.shape[:2]
        point = np.concatenate((point, np.ones((bs, n, 1))), axis=-1)
        point = rearrange(point, "b n d -> d (b n)")
        point = affine @ point
        point = rearrange(point, "d (b n)-> b n d", b=bs)[:, :, :3]
        return point

    def check_prompts_format(self, label_prompt, points, point_labels):
        """check the format of user prompts
        label_prompt: [1,2,3,4,...,B] List of tensors
        points: [[[x,y,z], [x,y,z], ...]] List of coordinates of a single object
        point_labels: [[1,1,0,...]] List of scalar that matches number of points
        """
        # check prompt is given
        if label_prompt is None and points is None:
            everything_labels = self.hyper_kwargs.get("everything_labels", None)
            if everything_labels is not None:
                label_prompt = [torch.tensor(_) for _ in everything_labels]
                return label_prompt, points, point_labels
            else:
                raise ValueError("Prompt must be given for inference.")
        # check label_prompt
        if label_prompt is not None:
            if isinstance(label_prompt, list):
                if not np.all([len(_) == 1 for _ in label_prompt]):
                    raise ValueError("Label prompt must be a list of single scalar, [1,2,3,4,...,].")
                if not np.all([(x < 255).item() for x in label_prompt]):
                    raise ValueError("Current bundle only supports label prompt smaller than 255.")
                if points is None:
                    supported_list = list({i + 1 for i in range(132)} - {16, 18, 129, 130, 131})
                    if not np.all([x in supported_list for x in label_prompt]):
                        raise ValueError("Undefined label prompt detected. Provide point prompts for zero-shot.")
            else:
                raise ValueError("Label prompt must be a list, [1,2,3,4,...,].")
        # check points
        if points is not None:
            if point_labels is None:
                raise ValueError("Point labels must be given if points are given.")
            if not np.all([len(_) == 3 for _ in points]):
                raise ValueError("Points must be three dimensional (x,y,z) in the shape of [[x,y,z],...,[x,y,z]].")
            if len(points) != len(point_labels):
                raise ValueError("Points must match point labels.")
            if not np.all([_ in [-1, 0, 1, 2, 3] for _ in point_labels]):
                raise ValueError("Point labels can only be -1,0,1 and 2,3 for special flags.")
        if label_prompt is not None and points is not None:
            if len(label_prompt) != 1:
                raise ValueError("Label prompt can only be a single object if provided with point prompts.")
        # check point_labels
        if point_labels is not None:
            if points is None:
                raise ValueError("Points must be given if point labels are given.")
        return label_prompt, points, point_labels

    def _iteration(self, engine: SupervisedEvaluator, batchdata: dict[str, torch.Tensor]) -> dict:
        """
        callback function for the Supervised Evaluation processing logic of 1 iteration in Ignite Engine.
        Return below items in a dictionary:
            - IMAGE: image Tensor data for model input, already moved to device.
            - LABEL: label Tensor data corresponding to the image, already moved to device.
            - PRED: prediction result of model.

        Args:
            engine: `SupervisedEvaluator` to execute operation for an iteration.
            batchdata: input data for this iteration, usually can be dictionary or tuple of Tensor data.

        Raises:
            ValueError: When ``batchdata`` is None.

        """
        if batchdata is None:
            raise ValueError("Must provide batch data for current iteration.")
        label_set = engine.hyper_kwargs.get("label_set", None)
        # this validation label set should be consistent with 'labels.unique()', used to generate fg/bg points
        val_label_set = engine.hyper_kwargs.get("val_label_set", label_set)
        # If user provide prompts in the inference, input image must contain original affine.
        # the point coordinates are from the original_affine space, while image here is after preprocess transforms.
        if engine.hyper_kwargs["user_prompt"]:
            inputs, label_prompt, points, point_labels = (
                batchdata["image"],
                batchdata.get("label_prompt", None),
                batchdata.get("points", None),
                batchdata.get("point_labels", None),
            )
            labels = None
            label_prompt, points, point_labels = self.check_prompts_format(label_prompt, points, point_labels)
            inputs = inputs.to(engine.device)
            # For N foreground object, label_prompt is [1, N], but the batch number 1 needs to be removed. Convert to [N, 1]
            label_prompt = (
                torch.as_tensor([label_prompt]).to(inputs.device)[0].unsqueeze(-1) if label_prompt is not None else None
            )
            # For points, the size can only be [1, K, 3], where K is the number of points for this single foreground object.
            if points is not None:
                points = torch.as_tensor([points])
                points = self.transform_points(
                    points, np.linalg.inv(inputs.affine[0]) @ inputs.meta["original_affine"][0].numpy()
                )
                points = torch.from_numpy(points).to(inputs.device)
            point_labels = torch.as_tensor([point_labels]).to(inputs.device) if point_labels is not None else None

        # If validation with ground truth label available.
        else:
            inputs, labels = engine.prepare_batch(
                batchdata, engine.state.device, engine.non_blocking, **engine.to_kwargs
            )
            # create label prompt, this should be consistent with the label prompt used for training.
            if label_set is None:
                output_classes = engine.hyper_kwargs["output_classes"]
                label_set = np.arange(output_classes).tolist()
            label_prompt = torch.tensor(label_set).to(engine.state.device).unsqueeze(-1)
            # point prompt is generated withing vista3d, provide empty points
            points = torch.zeros(label_prompt.shape[0], 1, 3).to(inputs.device)
            point_labels = -1 + torch.zeros(label_prompt.shape[0], 1).to(inputs.device)
            # validation for either auto or point.
            if engine.hyper_kwargs.get("val_head", "auto") == "auto":
                # automatic only validation
                # remove val_label_set, vista3d will not sample points from gt labels.
                val_label_set = None
            else:
                # point only validation
                label_prompt = None

        # put iteration outputs into engine.state
        engine.state.output = {Keys.IMAGE: inputs, Keys.LABEL: labels}
        # execute forward computation
        with engine.mode(engine.network):
            if engine.amp:
                with torch.amp.autocast("cuda", **engine.amp_kwargs):
                    engine.state.output[Keys.PRED] = engine.inferer(
                        inputs=inputs,
                        network=engine.network,
                        point_coords=points,
                        point_labels=point_labels,
                        class_vector=label_prompt,
                        labels=labels,
                        label_set=val_label_set,
                    )
            else:
                engine.state.output[Keys.PRED] = engine.inferer(
                    inputs=inputs,
                    network=engine.network,
                    point_coords=points,
                    point_labels=point_labels,
                    class_vector=label_prompt,
                    labels=labels,
                    label_set=val_label_set,
                )
        inputs = reset_ops_id(inputs)
        # Add dim 0 for decollate batch
        engine.state.output["label_prompt"] = label_prompt.unsqueeze(0) if label_prompt is not None else None
        engine.state.output["points"] = points.unsqueeze(0) if points is not None else None
        engine.state.output["point_labels"] = point_labels.unsqueeze(0) if point_labels is not None else None
        engine.fire_event(IterationEvents.FORWARD_COMPLETED)
        engine.fire_event(IterationEvents.MODEL_COMPLETED)
        if torch.cuda.is_available():
            torch.cuda.empty_cache()

        return engine.state.output