parta2_rpn_head.py 13.4 KB
Newer Older
dingchang's avatar
dingchang committed
1
# Copyright (c) OpenMMLab. All rights reserved.
wuyuefeng's avatar
wuyuefeng committed
2
3
import numpy as np
import torch
4
from mmcv.runner import force_fp32
wuyuefeng's avatar
wuyuefeng committed
5

zhangwenwei's avatar
zhangwenwei committed
6
from mmdet3d.core import limit_period, xywhr2xyxyr
7
from mmdet3d.core.post_processing import nms_bev, nms_normal_bev
wuyuefeng's avatar
wuyuefeng committed
8
from mmdet.models import HEADS
zhangwenwei's avatar
zhangwenwei committed
9
from .anchor3d_head import Anchor3DHead
wuyuefeng's avatar
wuyuefeng committed
10
11


12
@HEADS.register_module()
zhangwenwei's avatar
zhangwenwei committed
13
class PartA2RPNHead(Anchor3DHead):
zhangwenwei's avatar
zhangwenwei committed
14
    """RPN head for PartA2.
zhangwenwei's avatar
zhangwenwei committed
15
16
17
18
19
20
21
22
23
24
25

    Note:
        The main difference between the PartA2 RPN head and the Anchor3DHead
        lies in their output during inference. PartA2 RPN head further returns
        the original classification score for the second stage since the bbox
        head in RoI head does not do classification task.

        Different from RPN heads in 2D detectors, this RPN head does
        multi-class classification task and uses FocalLoss like the SECOND and
        PointPillars do. But this head uses class agnostic nms rather than
        multi-class nms.
wuyuefeng's avatar
wuyuefeng committed
26
27

    Args:
zhangwenwei's avatar
zhangwenwei committed
28
        num_classes (int): Number of classes.
wuyuefeng's avatar
wuyuefeng committed
29
        in_channels (int): Number of channels in the input feature map.
wangtai's avatar
wangtai committed
30
31
        train_cfg (dict): Train configs.
        test_cfg (dict): Test configs.
wuyuefeng's avatar
wuyuefeng committed
32
33
34
35
36
37
38
39
40
41
        feat_channels (int): Number of channels of the feature map.
        use_direction_classifier (bool): Whether to add a direction classifier.
        anchor_generator(dict): Config dict of anchor generator.
        assigner_per_size (bool): Whether to do assignment for each separate
            anchor size.
        assign_per_class (bool): Whether to do assignment for each class.
        diff_rad_by_sin (bool): Whether to change the difference into sin
            difference for box regression loss.
        dir_offset (float | int): The offset of BEV rotation angles
            (TODO: may be moved into box coder)
wuyuefeng's avatar
wuyuefeng committed
42
43
44
        dir_limit_offset (float | int): The limited range of BEV
            rotation angles. (TODO: may be moved into box coder)
        bbox_coder (dict): Config dict of box coders.
wuyuefeng's avatar
wuyuefeng committed
45
46
47
        loss_cls (dict): Config of classification loss.
        loss_bbox (dict): Config of localization loss.
        loss_dir (dict): Config of direction classifier loss.
zhangwenwei's avatar
zhangwenwei committed
48
    """
wuyuefeng's avatar
wuyuefeng committed
49
50

    def __init__(self,
zhangwenwei's avatar
zhangwenwei committed
51
                 num_classes,
wuyuefeng's avatar
wuyuefeng committed
52
53
54
55
56
57
58
59
60
                 in_channels,
                 train_cfg,
                 test_cfg,
                 feat_channels=256,
                 use_direction_classifier=True,
                 anchor_generator=dict(
                     type='Anchor3DRangeGenerator',
                     range=[0, -39.68, -1.78, 69.12, 39.68, -1.78],
                     strides=[2],
61
                     sizes=[[3.9, 1.6, 1.56]],
wuyuefeng's avatar
wuyuefeng committed
62
63
64
65
66
67
                     rotations=[0, 1.57],
                     custom_values=[],
                     reshape_out=False),
                 assigner_per_size=False,
                 assign_per_class=False,
                 diff_rad_by_sin=True,
68
69
                 dir_offset=-np.pi / 2,
                 dir_limit_offset=0,
wuyuefeng's avatar
wuyuefeng committed
70
71
72
73
74
75
76
                 bbox_coder=dict(type='DeltaXYZWLHRBBoxCoder'),
                 loss_cls=dict(
                     type='CrossEntropyLoss',
                     use_sigmoid=True,
                     loss_weight=1.0),
                 loss_bbox=dict(
                     type='SmoothL1Loss', beta=1.0 / 9.0, loss_weight=2.0),
77
78
                 loss_dir=dict(type='CrossEntropyLoss', loss_weight=0.2),
                 init_cfg=None):
zhangwenwei's avatar
zhangwenwei committed
79
        super().__init__(num_classes, in_channels, train_cfg, test_cfg,
wuyuefeng's avatar
wuyuefeng committed
80
                         feat_channels, use_direction_classifier,
zhangwenwei's avatar
zhangwenwei committed
81
82
                         anchor_generator, assigner_per_size, assign_per_class,
                         diff_rad_by_sin, dir_offset, dir_limit_offset,
83
                         bbox_coder, loss_cls, loss_bbox, loss_dir, init_cfg)
wuyuefeng's avatar
wuyuefeng committed
84

85
    @force_fp32(apply_to=('cls_scores', 'bbox_preds', 'dir_cls_preds'))
zhangwenwei's avatar
zhangwenwei committed
86
87
88
89
90
91
92
93
    def loss(self,
             cls_scores,
             bbox_preds,
             dir_cls_preds,
             gt_bboxes,
             gt_labels,
             input_metas,
             gt_bboxes_ignore=None):
94
95
96
97
98
99
100
        """Calculate losses.

        Args:
            cls_scores (list[torch.Tensor]): Multi-level class scores.
            bbox_preds (list[torch.Tensor]): Multi-level bbox predictions.
            dir_cls_preds (list[torch.Tensor]): Multi-level direction
                class predictions.
101
            gt_bboxes (list[:obj:`BaseInstance3DBoxes`]): Ground truth boxes
102
                of each sample.
wangtai's avatar
wangtai committed
103
104
            gt_labels (list[torch.Tensor]): Labels of each sample.
            input_metas (list[dict]): Point cloud and image's meta info.
105
            gt_bboxes_ignore (list[torch.Tensor]): Specify
106
107
108
                which bounding.

        Returns:
109
            dict[str, list[torch.Tensor]]: Classification, bbox, and
zhangwenwei's avatar
zhangwenwei committed
110
                direction losses of each level.
111
112
113

                - loss_rpn_cls (list[torch.Tensor]): Classification losses.
                - loss_rpn_bbox (list[torch.Tensor]): Box regression losses.
114
                - loss_rpn_dir (list[torch.Tensor]): Direction classification
115
116
                    losses.
        """
zhangwenwei's avatar
zhangwenwei committed
117
118
119
120
121
122
123
124
125
        loss_dict = super().loss(cls_scores, bbox_preds, dir_cls_preds,
                                 gt_bboxes, gt_labels, input_metas,
                                 gt_bboxes_ignore)
        # change the loss key names to avoid conflict
        return dict(
            loss_rpn_cls=loss_dict['loss_cls'],
            loss_rpn_bbox=loss_dict['loss_bbox'],
            loss_rpn_dir=loss_dict['loss_dir'])

wuyuefeng's avatar
wuyuefeng committed
126
127
128
129
130
131
132
133
    def get_bboxes_single(self,
                          cls_scores,
                          bbox_preds,
                          dir_cls_preds,
                          mlvl_anchors,
                          input_meta,
                          cfg,
                          rescale=False):
wuyuefeng's avatar
wuyuefeng committed
134
135
136
        """Get bboxes of single branch.

        Args:
liyinhao's avatar
liyinhao committed
137
138
139
140
141
            cls_scores (torch.Tensor): Class score in single batch.
            bbox_preds (torch.Tensor): Bbox prediction in single batch.
            dir_cls_preds (torch.Tensor): Predictions of direction class
                in single batch.
            mlvl_anchors (List[torch.Tensor]): Multi-level anchors
wuyuefeng's avatar
wuyuefeng committed
142
143
                in single batch.
            input_meta (list[dict]): Contain pcd and img's meta info.
144
            cfg (:obj:`ConfigDict`): Training or testing config.
liyinhao's avatar
liyinhao committed
145
            rescale (list[torch.Tensor]): whether th rescale bbox.
wuyuefeng's avatar
wuyuefeng committed
146
147

        Returns:
zhangwenwei's avatar
zhangwenwei committed
148
            dict: Predictions of single batch containing the following keys:
149

zhangwenwei's avatar
zhangwenwei committed
150
                - boxes_3d (:obj:`BaseInstance3DBoxes`): Predicted 3d bboxes.
liyinhao's avatar
liyinhao committed
151
152
153
                - scores_3d (torch.Tensor): Score of each bbox.
                - labels_3d (torch.Tensor): Label of each bbox.
                - cls_preds (torch.Tensor): Class score of each bbox.
wuyuefeng's avatar
wuyuefeng committed
154
        """
wuyuefeng's avatar
wuyuefeng committed
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
        assert len(cls_scores) == len(bbox_preds) == len(mlvl_anchors)
        mlvl_bboxes = []
        mlvl_max_scores = []
        mlvl_label_pred = []
        mlvl_dir_scores = []
        mlvl_cls_score = []
        for cls_score, bbox_pred, dir_cls_pred, anchors in zip(
                cls_scores, bbox_preds, dir_cls_preds, mlvl_anchors):
            assert cls_score.size()[-2:] == bbox_pred.size()[-2:]
            assert cls_score.size()[-2:] == dir_cls_pred.size()[-2:]
            dir_cls_pred = dir_cls_pred.permute(1, 2, 0).reshape(-1, 2)
            dir_cls_score = torch.max(dir_cls_pred, dim=-1)[1]

            cls_score = cls_score.permute(1, 2,
                                          0).reshape(-1, self.num_classes)

            if self.use_sigmoid_cls:
                scores = cls_score.sigmoid()
            else:
                scores = cls_score.softmax(-1)
            bbox_pred = bbox_pred.permute(1, 2,
                                          0).reshape(-1, self.box_code_size)

            nms_pre = cfg.get('nms_pre', -1)
            if self.use_sigmoid_cls:
                max_scores, pred_labels = scores.max(dim=1)
            else:
                max_scores, pred_labels = scores[:, :-1].max(dim=1)
            # get topk
            if nms_pre > 0 and scores.shape[0] > nms_pre:
                topk_scores, topk_inds = max_scores.topk(nms_pre)
                anchors = anchors[topk_inds, :]
                bbox_pred = bbox_pred[topk_inds, :]
                max_scores = topk_scores
zhangwenwei's avatar
zhangwenwei committed
189
                cls_score = scores[topk_inds, :]
wuyuefeng's avatar
wuyuefeng committed
190
191
192
193
194
195
196
197
198
199
200
                dir_cls_score = dir_cls_score[topk_inds]
                pred_labels = pred_labels[topk_inds]

            bboxes = self.bbox_coder.decode(anchors, bbox_pred)
            mlvl_bboxes.append(bboxes)
            mlvl_max_scores.append(max_scores)
            mlvl_cls_score.append(cls_score)
            mlvl_label_pred.append(pred_labels)
            mlvl_dir_scores.append(dir_cls_score)

        mlvl_bboxes = torch.cat(mlvl_bboxes)
zhangwenwei's avatar
zhangwenwei committed
201
202
        mlvl_bboxes_for_nms = xywhr2xyxyr(input_meta['box_type_3d'](
            mlvl_bboxes, box_dim=self.box_code_size).bev)
wuyuefeng's avatar
wuyuefeng committed
203
204
205
        mlvl_max_scores = torch.cat(mlvl_max_scores)
        mlvl_label_pred = torch.cat(mlvl_label_pred)
        mlvl_dir_scores = torch.cat(mlvl_dir_scores)
zhangwenwei's avatar
zhangwenwei committed
206
207
        # shape [k, num_class] before sigmoid
        # PartA2 need to keep raw classification score
208
        # because the bbox head in the second stage does not have
zhangwenwei's avatar
zhangwenwei committed
209
210
211
        # classification branch,
        # roi head need this score as classification score
        mlvl_cls_score = torch.cat(mlvl_cls_score)
wuyuefeng's avatar
wuyuefeng committed
212
213
214
215
216

        score_thr = cfg.get('score_thr', 0)
        result = self.class_agnostic_nms(mlvl_bboxes, mlvl_bboxes_for_nms,
                                         mlvl_max_scores, mlvl_label_pred,
                                         mlvl_cls_score, mlvl_dir_scores,
217
218
                                         score_thr, cfg.nms_post, cfg,
                                         input_meta)
wuyuefeng's avatar
wuyuefeng committed
219
220
221
222
223

        return result

    def class_agnostic_nms(self, mlvl_bboxes, mlvl_bboxes_for_nms,
                           mlvl_max_scores, mlvl_label_pred, mlvl_cls_score,
224
225
                           mlvl_dir_scores, score_thr, max_num, cfg,
                           input_meta):
wuyuefeng's avatar
wuyuefeng committed
226
227
228
        """Class agnostic nms for single batch.

        Args:
liyinhao's avatar
liyinhao committed
229
230
231
232
233
234
235
236
237
238
            mlvl_bboxes (torch.Tensor): Bboxes from Multi-level.
            mlvl_bboxes_for_nms (torch.Tensor): Bboxes for nms
                (bev or minmax boxes) from Multi-level.
            mlvl_max_scores (torch.Tensor): Max scores of Multi-level bbox.
            mlvl_label_pred (torch.Tensor): Class predictions
                of Multi-level bbox.
            mlvl_cls_score (torch.Tensor): Class scores of
                Multi-level bbox.
            mlvl_dir_scores (torch.Tensor): Direction scores of
                Multi-level bbox.
wuyuefeng's avatar
wuyuefeng committed
239
240
            score_thr (int): Score threshold.
            max_num (int): Max number of bboxes after nms.
241
            cfg (:obj:`ConfigDict`): Training or testing config.
wuyuefeng's avatar
wuyuefeng committed
242
243
244
245
            input_meta (dict): Contain pcd and img's meta info.

        Returns:
            dict: Predictions of single batch. Contain the keys:
246

zhangwenwei's avatar
zhangwenwei committed
247
                - boxes_3d (:obj:`BaseInstance3DBoxes`): Predicted 3d bboxes.
liyinhao's avatar
liyinhao committed
248
249
250
                - scores_3d (torch.Tensor): Score of each bbox.
                - labels_3d (torch.Tensor): Label of each bbox.
                - cls_preds (torch.Tensor): Class score of each bbox.
wuyuefeng's avatar
wuyuefeng committed
251
        """
wuyuefeng's avatar
wuyuefeng committed
252
253
254
255
256
257
258
259
260
        bboxes = []
        scores = []
        labels = []
        dir_scores = []
        cls_scores = []
        score_thr_inds = mlvl_max_scores > score_thr
        _scores = mlvl_max_scores[score_thr_inds]
        _bboxes_for_nms = mlvl_bboxes_for_nms[score_thr_inds, :]
        if cfg.use_rotate_nms:
261
            nms_func = nms_bev
wuyuefeng's avatar
wuyuefeng committed
262
        else:
263
            nms_func = nms_normal_bev
wuyuefeng's avatar
wuyuefeng committed
264
265
266
267
268
269
270
271
272
273
274
275
276
        selected = nms_func(_bboxes_for_nms, _scores, cfg.nms_thr)

        _mlvl_bboxes = mlvl_bboxes[score_thr_inds, :]
        _mlvl_dir_scores = mlvl_dir_scores[score_thr_inds]
        _mlvl_label_pred = mlvl_label_pred[score_thr_inds]
        _mlvl_cls_score = mlvl_cls_score[score_thr_inds]

        if len(selected) > 0:
            bboxes.append(_mlvl_bboxes[selected])
            scores.append(_scores[selected])
            labels.append(_mlvl_label_pred[selected])
            cls_scores.append(_mlvl_cls_score[selected])
            dir_scores.append(_mlvl_dir_scores[selected])
zhangwenwei's avatar
zhangwenwei committed
277
278
            dir_rot = limit_period(bboxes[-1][..., 6] - self.dir_offset,
                                   self.dir_limit_offset, np.pi)
wuyuefeng's avatar
wuyuefeng committed
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
            bboxes[-1][..., 6] = (
                dir_rot + self.dir_offset +
                np.pi * dir_scores[-1].to(bboxes[-1].dtype))

        if bboxes:
            bboxes = torch.cat(bboxes, dim=0)
            scores = torch.cat(scores, dim=0)
            cls_scores = torch.cat(cls_scores, dim=0)
            labels = torch.cat(labels, dim=0)
            if bboxes.shape[0] > max_num:
                _, inds = scores.sort(descending=True)
                inds = inds[:max_num]
                bboxes = bboxes[inds, :]
                labels = labels[inds]
                scores = scores[inds]
                cls_scores = cls_scores[inds]
295
296
            bboxes = input_meta['box_type_3d'](
                bboxes, box_dim=self.box_code_size)
wuyuefeng's avatar
wuyuefeng committed
297
            return dict(
zhangwenwei's avatar
zhangwenwei committed
298
299
300
                boxes_3d=bboxes,
                scores_3d=scores,
                labels_3d=labels,
wuyuefeng's avatar
wuyuefeng committed
301
                cls_preds=cls_scores  # raw scores [max_num, cls_num]
wuyuefeng's avatar
wuyuefeng committed
302
303
304
            )
        else:
            return dict(
305
306
307
                boxes_3d=input_meta['box_type_3d'](
                    mlvl_bboxes.new_zeros([0, self.box_code_size]),
                    box_dim=self.box_code_size),
zhangwenwei's avatar
zhangwenwei committed
308
309
                scores_3d=mlvl_bboxes.new_zeros([0]),
                labels_3d=mlvl_bboxes.new_zeros([0]),
wuyuefeng's avatar
wuyuefeng committed
310
                cls_preds=mlvl_bboxes.new_zeros([0, mlvl_cls_score.shape[-1]]))