File size: 23,513 Bytes
61c2d32
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
# Copyright (c) Facebook, Inc. and its affiliates.
import numpy as np
from typing import Callable, Dict, List, Union
import fvcore.nn.weight_init as weight_init
import torch
from torch import nn
from torch.nn import functional as F

from detectron2.config import configurable
from detectron2.data import MetadataCatalog
from detectron2.layers import Conv2d, DepthwiseSeparableConv2d, ShapeSpec, get_norm
from detectron2.modeling import (
    META_ARCH_REGISTRY,
    SEM_SEG_HEADS_REGISTRY,
    build_backbone,
    build_sem_seg_head,
)
from detectron2.modeling.postprocessing import sem_seg_postprocess
from detectron2.projects.deeplab import DeepLabV3PlusHead
from detectron2.projects.deeplab.loss import DeepLabCE
from detectron2.structures import BitMasks, ImageList, Instances
from detectron2.utils.registry import Registry

from .post_processing import get_panoptic_segmentation

__all__ = ["PanopticDeepLab", "INS_EMBED_BRANCHES_REGISTRY", "build_ins_embed_branch"]


INS_EMBED_BRANCHES_REGISTRY = Registry("INS_EMBED_BRANCHES")
INS_EMBED_BRANCHES_REGISTRY.__doc__ = """
Registry for instance embedding branches, which make instance embedding
predictions from feature maps.
"""


@META_ARCH_REGISTRY.register()
class PanopticDeepLab(nn.Module):
    """
    Main class for panoptic segmentation architectures.
    """

    def __init__(self, cfg):
        super().__init__()
        self.backbone = build_backbone(cfg)
        self.sem_seg_head = build_sem_seg_head(cfg, self.backbone.output_shape())
        self.ins_embed_head = build_ins_embed_branch(cfg, self.backbone.output_shape())
        self.register_buffer("pixel_mean", torch.tensor(cfg.MODEL.PIXEL_MEAN).view(-1, 1, 1), False)
        self.register_buffer("pixel_std", torch.tensor(cfg.MODEL.PIXEL_STD).view(-1, 1, 1), False)
        self.meta = MetadataCatalog.get(cfg.DATASETS.TRAIN[0])
        self.stuff_area = cfg.MODEL.PANOPTIC_DEEPLAB.STUFF_AREA
        self.threshold = cfg.MODEL.PANOPTIC_DEEPLAB.CENTER_THRESHOLD
        self.nms_kernel = cfg.MODEL.PANOPTIC_DEEPLAB.NMS_KERNEL
        self.top_k = cfg.MODEL.PANOPTIC_DEEPLAB.TOP_K_INSTANCE
        self.predict_instances = cfg.MODEL.PANOPTIC_DEEPLAB.PREDICT_INSTANCES
        self.use_depthwise_separable_conv = cfg.MODEL.PANOPTIC_DEEPLAB.USE_DEPTHWISE_SEPARABLE_CONV
        assert (
            cfg.MODEL.SEM_SEG_HEAD.USE_DEPTHWISE_SEPARABLE_CONV
            == cfg.MODEL.PANOPTIC_DEEPLAB.USE_DEPTHWISE_SEPARABLE_CONV
        )
        self.size_divisibility = cfg.MODEL.PANOPTIC_DEEPLAB.SIZE_DIVISIBILITY
        self.benchmark_network_speed = cfg.MODEL.PANOPTIC_DEEPLAB.BENCHMARK_NETWORK_SPEED

    @property
    def device(self):
        return self.pixel_mean.device

    def forward(self, batched_inputs):
        """
        Args:
            batched_inputs: a list, batched outputs of :class:`DatasetMapper`.
                Each item in the list contains the inputs for one image.
                For now, each item in the list is a dict that contains:
                   * "image": Tensor, image in (C, H, W) format.
                   * "sem_seg": semantic segmentation ground truth
                   * "center": center points heatmap ground truth
                   * "offset": pixel offsets to center points ground truth
                   * Other information that's included in the original dicts, such as:
                     "height", "width" (int): the output resolution of the model (may be different
                     from input resolution), used in inference.
        Returns:
            list[dict]:
                each dict is the results for one image. The dict contains the following keys:

                * "panoptic_seg", "sem_seg": see documentation
                    :doc:`/tutorials/models` for the standard output format
                * "instances": available if ``predict_instances is True``. see documentation
                    :doc:`/tutorials/models` for the standard output format
        """
        images = [x["image"].to(self.device) for x in batched_inputs]
        images = [(x - self.pixel_mean) / self.pixel_std for x in images]
        # To avoid error in ASPP layer when input has different size.
        size_divisibility = (
            self.size_divisibility
            if self.size_divisibility > 0
            else self.backbone.size_divisibility
        )
        images = ImageList.from_tensors(images, size_divisibility)

        features = self.backbone(images.tensor)

        losses = {}
        if "sem_seg" in batched_inputs[0]:
            targets = [x["sem_seg"].to(self.device) for x in batched_inputs]
            targets = ImageList.from_tensors(
                targets, size_divisibility, self.sem_seg_head.ignore_value
            ).tensor
            if "sem_seg_weights" in batched_inputs[0]:
                # The default D2 DatasetMapper may not contain "sem_seg_weights"
                # Avoid error in testing when default DatasetMapper is used.
                weights = [x["sem_seg_weights"].to(self.device) for x in batched_inputs]
                weights = ImageList.from_tensors(weights, size_divisibility).tensor
            else:
                weights = None
        else:
            targets = None
            weights = None
        sem_seg_results, sem_seg_losses = self.sem_seg_head(features, targets, weights)
        losses.update(sem_seg_losses)

        if "center" in batched_inputs[0] and "offset" in batched_inputs[0]:
            center_targets = [x["center"].to(self.device) for x in batched_inputs]
            center_targets = ImageList.from_tensors(
                center_targets, size_divisibility
            ).tensor.unsqueeze(1)
            center_weights = [x["center_weights"].to(self.device) for x in batched_inputs]
            center_weights = ImageList.from_tensors(center_weights, size_divisibility).tensor

            offset_targets = [x["offset"].to(self.device) for x in batched_inputs]
            offset_targets = ImageList.from_tensors(offset_targets, size_divisibility).tensor
            offset_weights = [x["offset_weights"].to(self.device) for x in batched_inputs]
            offset_weights = ImageList.from_tensors(offset_weights, size_divisibility).tensor
        else:
            center_targets = None
            center_weights = None

            offset_targets = None
            offset_weights = None

        center_results, offset_results, center_losses, offset_losses = self.ins_embed_head(
            features, center_targets, center_weights, offset_targets, offset_weights
        )
        losses.update(center_losses)
        losses.update(offset_losses)

        if self.training:
            return losses

        if self.benchmark_network_speed:
            return []

        processed_results = []
        for sem_seg_result, center_result, offset_result, input_per_image, image_size in zip(
            sem_seg_results, center_results, offset_results, batched_inputs, images.image_sizes
        ):
            height = input_per_image.get("height")
            width = input_per_image.get("width")
            r = sem_seg_postprocess(sem_seg_result, image_size, height, width)
            c = sem_seg_postprocess(center_result, image_size, height, width)
            o = sem_seg_postprocess(offset_result, image_size, height, width)
            # Post-processing to get panoptic segmentation.
            panoptic_image, _ = get_panoptic_segmentation(
                r.argmax(dim=0, keepdim=True),
                c,
                o,
                thing_ids=self.meta.thing_dataset_id_to_contiguous_id.values(),
                label_divisor=self.meta.label_divisor,
                stuff_area=self.stuff_area,
                void_label=-1,
                threshold=self.threshold,
                nms_kernel=self.nms_kernel,
                top_k=self.top_k,
            )
            # For semantic segmentation evaluation.
            processed_results.append({"sem_seg": r})
            panoptic_image = panoptic_image.squeeze(0)
            semantic_prob = F.softmax(r, dim=0)
            # For panoptic segmentation evaluation.
            processed_results[-1]["panoptic_seg"] = (panoptic_image, None)
            # For instance segmentation evaluation.
            if self.predict_instances:
                instances = []
                panoptic_image_cpu = panoptic_image.cpu().numpy()
                for panoptic_label in np.unique(panoptic_image_cpu):
                    if panoptic_label == -1:
                        continue
                    pred_class = panoptic_label // self.meta.label_divisor
                    isthing = pred_class in list(
                        self.meta.thing_dataset_id_to_contiguous_id.values()
                    )
                    # Get instance segmentation results.
                    if isthing:
                        instance = Instances((height, width))
                        # Evaluation code takes continuous id starting from 0
                        instance.pred_classes = torch.tensor(
                            [pred_class], device=panoptic_image.device
                        )
                        mask = panoptic_image == panoptic_label
                        instance.pred_masks = mask.unsqueeze(0)
                        # Average semantic probability
                        sem_scores = semantic_prob[pred_class, ...]
                        sem_scores = torch.mean(sem_scores[mask])
                        # Center point probability
                        mask_indices = torch.nonzero(mask).float()
                        center_y, center_x = (
                            torch.mean(mask_indices[:, 0]),
                            torch.mean(mask_indices[:, 1]),
                        )
                        center_scores = c[0, int(center_y.item()), int(center_x.item())]
                        # Confidence score is semantic prob * center prob.
                        instance.scores = torch.tensor(
                            [sem_scores * center_scores], device=panoptic_image.device
                        )
                        # Get bounding boxes
                        instance.pred_boxes = BitMasks(instance.pred_masks).get_bounding_boxes()
                        instances.append(instance)
                if len(instances) > 0:
                    processed_results[-1]["instances"] = Instances.cat(instances)

        return processed_results


@SEM_SEG_HEADS_REGISTRY.register()
class PanopticDeepLabSemSegHead(DeepLabV3PlusHead):
    """
    A semantic segmentation head described in :paper:`Panoptic-DeepLab`.
    """

    @configurable
    def __init__(
        self,
        input_shape: Dict[str, ShapeSpec],
        *,
        decoder_channels: List[int],
        norm: Union[str, Callable],
        head_channels: int,
        loss_weight: float,
        loss_type: str,
        loss_top_k: float,
        ignore_value: int,
        num_classes: int,
        **kwargs,
    ):
        """
        NOTE: this interface is experimental.

        Args:
            input_shape (ShapeSpec): shape of the input feature
            decoder_channels (list[int]): a list of output channels of each
                decoder stage. It should have the same length as "input_shape"
                (each element in "input_shape" corresponds to one decoder stage).
            norm (str or callable): normalization for all conv layers.
            head_channels (int): the output channels of extra convolutions
                between decoder and predictor.
            loss_weight (float): loss weight.
            loss_top_k: (float): setting the top k% hardest pixels for
                "hard_pixel_mining" loss.
            loss_type, ignore_value, num_classes: the same as the base class.
        """
        super().__init__(
            input_shape,
            decoder_channels=decoder_channels,
            norm=norm,
            ignore_value=ignore_value,
            **kwargs,
        )
        assert self.decoder_only

        self.loss_weight = loss_weight
        use_bias = norm == ""
        # `head` is additional transform before predictor
        if self.use_depthwise_separable_conv:
            # We use a single 5x5 DepthwiseSeparableConv2d to replace
            # 2 3x3 Conv2d since they have the same receptive field.
            self.head = DepthwiseSeparableConv2d(
                decoder_channels[0],
                head_channels,
                kernel_size=5,
                padding=2,
                norm1=norm,
                activation1=F.relu,
                norm2=norm,
                activation2=F.relu,
            )
        else:
            self.head = nn.Sequential(
                Conv2d(
                    decoder_channels[0],
                    decoder_channels[0],
                    kernel_size=3,
                    padding=1,
                    bias=use_bias,
                    norm=get_norm(norm, decoder_channels[0]),
                    activation=F.relu,
                ),
                Conv2d(
                    decoder_channels[0],
                    head_channels,
                    kernel_size=3,
                    padding=1,
                    bias=use_bias,
                    norm=get_norm(norm, head_channels),
                    activation=F.relu,
                ),
            )
            weight_init.c2_xavier_fill(self.head[0])
            weight_init.c2_xavier_fill(self.head[1])
        self.predictor = Conv2d(head_channels, num_classes, kernel_size=1)
        nn.init.normal_(self.predictor.weight, 0, 0.001)
        nn.init.constant_(self.predictor.bias, 0)

        if loss_type == "cross_entropy":
            self.loss = nn.CrossEntropyLoss(reduction="mean", ignore_index=ignore_value)
        elif loss_type == "hard_pixel_mining":
            self.loss = DeepLabCE(ignore_label=ignore_value, top_k_percent_pixels=loss_top_k)
        else:
            raise ValueError("Unexpected loss type: %s" % loss_type)

    @classmethod
    def from_config(cls, cfg, input_shape):
        ret = super().from_config(cfg, input_shape)
        ret["head_channels"] = cfg.MODEL.SEM_SEG_HEAD.HEAD_CHANNELS
        ret["loss_top_k"] = cfg.MODEL.SEM_SEG_HEAD.LOSS_TOP_K
        return ret

    def forward(self, features, targets=None, weights=None):
        """
        Returns:
            In training, returns (None, dict of losses)
            In inference, returns (CxHxW logits, {})
        """
        y = self.layers(features)
        if self.training:
            return None, self.losses(y, targets, weights)
        else:
            y = F.interpolate(
                y, scale_factor=self.common_stride, mode="bilinear", align_corners=False
            )
            return y, {}

    def layers(self, features):
        assert self.decoder_only
        y = super().layers(features)
        y = self.head(y)
        y = self.predictor(y)
        return y

    def losses(self, predictions, targets, weights=None):
        predictions = F.interpolate(
            predictions, scale_factor=self.common_stride, mode="bilinear", align_corners=False
        )
        loss = self.loss(predictions, targets, weights)
        losses = {"loss_sem_seg": loss * self.loss_weight}
        return losses


def build_ins_embed_branch(cfg, input_shape):
    """
    Build a instance embedding branch from `cfg.MODEL.INS_EMBED_HEAD.NAME`.
    """
    name = cfg.MODEL.INS_EMBED_HEAD.NAME
    return INS_EMBED_BRANCHES_REGISTRY.get(name)(cfg, input_shape)


@INS_EMBED_BRANCHES_REGISTRY.register()
class PanopticDeepLabInsEmbedHead(DeepLabV3PlusHead):
    """
    A instance embedding head described in :paper:`Panoptic-DeepLab`.
    """

    @configurable
    def __init__(
        self,
        input_shape: Dict[str, ShapeSpec],
        *,
        decoder_channels: List[int],
        norm: Union[str, Callable],
        head_channels: int,
        center_loss_weight: float,
        offset_loss_weight: float,
        **kwargs,
    ):
        """
        NOTE: this interface is experimental.

        Args:
            input_shape (ShapeSpec): shape of the input feature
            decoder_channels (list[int]): a list of output channels of each
                decoder stage. It should have the same length as "input_shape"
                (each element in "input_shape" corresponds to one decoder stage).
            norm (str or callable): normalization for all conv layers.
            head_channels (int): the output channels of extra convolutions
                between decoder and predictor.
            center_loss_weight (float): loss weight for center point prediction.
            offset_loss_weight (float): loss weight for center offset prediction.
        """
        super().__init__(input_shape, decoder_channels=decoder_channels, norm=norm, **kwargs)
        assert self.decoder_only

        self.center_loss_weight = center_loss_weight
        self.offset_loss_weight = offset_loss_weight
        use_bias = norm == ""
        # center prediction
        # `head` is additional transform before predictor
        self.center_head = nn.Sequential(
            Conv2d(
                decoder_channels[0],
                decoder_channels[0],
                kernel_size=3,
                padding=1,
                bias=use_bias,
                norm=get_norm(norm, decoder_channels[0]),
                activation=F.relu,
            ),
            Conv2d(
                decoder_channels[0],
                head_channels,
                kernel_size=3,
                padding=1,
                bias=use_bias,
                norm=get_norm(norm, head_channels),
                activation=F.relu,
            ),
        )
        weight_init.c2_xavier_fill(self.center_head[0])
        weight_init.c2_xavier_fill(self.center_head[1])
        self.center_predictor = Conv2d(head_channels, 1, kernel_size=1)
        nn.init.normal_(self.center_predictor.weight, 0, 0.001)
        nn.init.constant_(self.center_predictor.bias, 0)

        # offset prediction
        # `head` is additional transform before predictor
        if self.use_depthwise_separable_conv:
            # We use a single 5x5 DepthwiseSeparableConv2d to replace
            # 2 3x3 Conv2d since they have the same receptive field.
            self.offset_head = DepthwiseSeparableConv2d(
                decoder_channels[0],
                head_channels,
                kernel_size=5,
                padding=2,
                norm1=norm,
                activation1=F.relu,
                norm2=norm,
                activation2=F.relu,
            )
        else:
            self.offset_head = nn.Sequential(
                Conv2d(
                    decoder_channels[0],
                    decoder_channels[0],
                    kernel_size=3,
                    padding=1,
                    bias=use_bias,
                    norm=get_norm(norm, decoder_channels[0]),
                    activation=F.relu,
                ),
                Conv2d(
                    decoder_channels[0],
                    head_channels,
                    kernel_size=3,
                    padding=1,
                    bias=use_bias,
                    norm=get_norm(norm, head_channels),
                    activation=F.relu,
                ),
            )
            weight_init.c2_xavier_fill(self.offset_head[0])
            weight_init.c2_xavier_fill(self.offset_head[1])
        self.offset_predictor = Conv2d(head_channels, 2, kernel_size=1)
        nn.init.normal_(self.offset_predictor.weight, 0, 0.001)
        nn.init.constant_(self.offset_predictor.bias, 0)

        self.center_loss = nn.MSELoss(reduction="none")
        self.offset_loss = nn.L1Loss(reduction="none")

    @classmethod
    def from_config(cls, cfg, input_shape):
        if cfg.INPUT.CROP.ENABLED:
            assert cfg.INPUT.CROP.TYPE == "absolute"
            train_size = cfg.INPUT.CROP.SIZE
        else:
            train_size = None
        decoder_channels = [cfg.MODEL.INS_EMBED_HEAD.CONVS_DIM] * (
            len(cfg.MODEL.INS_EMBED_HEAD.IN_FEATURES) - 1
        ) + [cfg.MODEL.INS_EMBED_HEAD.ASPP_CHANNELS]
        ret = dict(
            input_shape={
                k: v for k, v in input_shape.items() if k in cfg.MODEL.INS_EMBED_HEAD.IN_FEATURES
            },
            project_channels=cfg.MODEL.INS_EMBED_HEAD.PROJECT_CHANNELS,
            aspp_dilations=cfg.MODEL.INS_EMBED_HEAD.ASPP_DILATIONS,
            aspp_dropout=cfg.MODEL.INS_EMBED_HEAD.ASPP_DROPOUT,
            decoder_channels=decoder_channels,
            common_stride=cfg.MODEL.INS_EMBED_HEAD.COMMON_STRIDE,
            norm=cfg.MODEL.INS_EMBED_HEAD.NORM,
            train_size=train_size,
            head_channels=cfg.MODEL.INS_EMBED_HEAD.HEAD_CHANNELS,
            center_loss_weight=cfg.MODEL.INS_EMBED_HEAD.CENTER_LOSS_WEIGHT,
            offset_loss_weight=cfg.MODEL.INS_EMBED_HEAD.OFFSET_LOSS_WEIGHT,
            use_depthwise_separable_conv=cfg.MODEL.SEM_SEG_HEAD.USE_DEPTHWISE_SEPARABLE_CONV,
        )
        return ret

    def forward(
        self,
        features,
        center_targets=None,
        center_weights=None,
        offset_targets=None,
        offset_weights=None,
    ):
        """
        Returns:
            In training, returns (None, dict of losses)
            In inference, returns (CxHxW logits, {})
        """
        center, offset = self.layers(features)
        if self.training:
            return (
                None,
                None,
                self.center_losses(center, center_targets, center_weights),
                self.offset_losses(offset, offset_targets, offset_weights),
            )
        else:
            center = F.interpolate(
                center, scale_factor=self.common_stride, mode="bilinear", align_corners=False
            )
            offset = (
                F.interpolate(
                    offset, scale_factor=self.common_stride, mode="bilinear", align_corners=False
                )
                * self.common_stride
            )
            return center, offset, {}, {}

    def layers(self, features):
        assert self.decoder_only
        y = super().layers(features)
        # center
        center = self.center_head(y)
        center = self.center_predictor(center)
        # offset
        offset = self.offset_head(y)
        offset = self.offset_predictor(offset)
        return center, offset

    def center_losses(self, predictions, targets, weights):
        predictions = F.interpolate(
            predictions, scale_factor=self.common_stride, mode="bilinear", align_corners=False
        )
        loss = self.center_loss(predictions, targets) * weights
        if weights.sum() > 0:
            loss = loss.sum() / weights.sum()
        else:
            loss = loss.sum() * 0
        losses = {"loss_center": loss * self.center_loss_weight}
        return losses

    def offset_losses(self, predictions, targets, weights):
        predictions = (
            F.interpolate(
                predictions, scale_factor=self.common_stride, mode="bilinear", align_corners=False
            )
            * self.common_stride
        )
        loss = self.offset_loss(predictions, targets) * weights
        if weights.sum() > 0:
            loss = loss.sum() / weights.sum()
        else:
            loss = loss.sum() * 0
        losses = {"loss_offset": loss * self.offset_loss_weight}
        return losses