File size: 6,794 Bytes
f549064
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Copyright (c) OpenMMLab. All rights reserved.
from typing import List, Tuple

import numpy as np
import torch
from torch import Tensor

from mmdet.models.losses import SmoothL1Loss
from mmdet.models.task_modules.samplers import SamplingResult
from mmdet.registry import MODELS
from mmdet.structures import SampleList
from mmdet.structures.bbox import bbox2roi
from mmdet.utils import InstanceList
from ..utils.misc import unpack_gt_instances
from .standard_roi_head import StandardRoIHead

EPS = 1e-15


@MODELS.register_module()
class DynamicRoIHead(StandardRoIHead):
    """RoI head for `Dynamic R-CNN <https://arxiv.org/abs/2004.06002>`_."""

    def __init__(self, **kwargs) -> None:
        super().__init__(**kwargs)
        assert isinstance(self.bbox_head.loss_bbox, SmoothL1Loss)
        # the IoU history of the past `update_iter_interval` iterations
        self.iou_history = []
        # the beta history of the past `update_iter_interval` iterations
        self.beta_history = []

    def loss(self, x: Tuple[Tensor], rpn_results_list: InstanceList,
             batch_data_samples: SampleList) -> dict:
        """Forward function for training.

        Args:
            x (tuple[Tensor]): List of multi-level img features.
            rpn_results_list (list[:obj:`InstanceData`]): List of region
                proposals.
            batch_data_samples (list[:obj:`DetDataSample`]): The batch
                data samples. It usually includes information such
                as `gt_instance` or `gt_panoptic_seg` or `gt_sem_seg`.

        Returns:
            dict[str, Tensor]: a dictionary of loss components
        """
        assert len(rpn_results_list) == len(batch_data_samples)
        outputs = unpack_gt_instances(batch_data_samples)
        batch_gt_instances, batch_gt_instances_ignore, _ = outputs

        # assign gts and sample proposals
        num_imgs = len(batch_data_samples)
        sampling_results = []
        cur_iou = []
        for i in range(num_imgs):
            # rename rpn_results.bboxes to rpn_results.priors
            rpn_results = rpn_results_list[i]
            rpn_results.priors = rpn_results.pop('bboxes')

            assign_result = self.bbox_assigner.assign(
                rpn_results, batch_gt_instances[i],
                batch_gt_instances_ignore[i])
            sampling_result = self.bbox_sampler.sample(
                assign_result,
                rpn_results,
                batch_gt_instances[i],
                feats=[lvl_feat[i][None] for lvl_feat in x])
            # record the `iou_topk`-th largest IoU in an image
            iou_topk = min(self.train_cfg.dynamic_rcnn.iou_topk,
                           len(assign_result.max_overlaps))
            ious, _ = torch.topk(assign_result.max_overlaps, iou_topk)
            cur_iou.append(ious[-1].item())
            sampling_results.append(sampling_result)
        # average the current IoUs over images
        cur_iou = np.mean(cur_iou)
        self.iou_history.append(cur_iou)

        losses = dict()
        # bbox head forward and loss
        if self.with_bbox:
            bbox_results = self.bbox_loss(x, sampling_results)
            losses.update(bbox_results['loss_bbox'])

        # mask head forward and loss
        if self.with_mask:
            mask_results = self.mask_loss(x, sampling_results,
                                          bbox_results['bbox_feats'],
                                          batch_gt_instances)
            losses.update(mask_results['loss_mask'])

        # update IoU threshold and SmoothL1 beta
        update_iter_interval = self.train_cfg.dynamic_rcnn.update_iter_interval
        if len(self.iou_history) % update_iter_interval == 0:
            new_iou_thr, new_beta = self.update_hyperparameters()

        return losses

    def bbox_loss(self, x: Tuple[Tensor],
                  sampling_results: List[SamplingResult]) -> dict:
        """Perform forward propagation and loss calculation of the bbox head on
        the features of the upstream network.

        Args:
            x (tuple[Tensor]): List of multi-level img features.
            sampling_results (list["obj:`SamplingResult`]): Sampling results.

        Returns:
            dict[str, Tensor]: Usually returns a dictionary with keys:

                - `cls_score` (Tensor): Classification scores.
                - `bbox_pred` (Tensor): Box energies / deltas.
                - `bbox_feats` (Tensor): Extract bbox RoI features.
                - `loss_bbox` (dict): A dictionary of bbox loss components.
        """
        rois = bbox2roi([res.priors for res in sampling_results])
        bbox_results = self._bbox_forward(x, rois)

        bbox_loss_and_target = self.bbox_head.loss_and_target(
            cls_score=bbox_results['cls_score'],
            bbox_pred=bbox_results['bbox_pred'],
            rois=rois,
            sampling_results=sampling_results,
            rcnn_train_cfg=self.train_cfg)
        bbox_results.update(loss_bbox=bbox_loss_and_target['loss_bbox'])

        # record the `beta_topk`-th smallest target
        # `bbox_targets[2]` and `bbox_targets[3]` stand for bbox_targets
        # and bbox_weights, respectively
        bbox_targets = bbox_loss_and_target['bbox_targets']
        pos_inds = bbox_targets[3][:, 0].nonzero().squeeze(1)
        num_pos = len(pos_inds)
        num_imgs = len(sampling_results)
        if num_pos > 0:
            cur_target = bbox_targets[2][pos_inds, :2].abs().mean(dim=1)
            beta_topk = min(self.train_cfg.dynamic_rcnn.beta_topk * num_imgs,
                            num_pos)
            cur_target = torch.kthvalue(cur_target, beta_topk)[0].item()
            self.beta_history.append(cur_target)

        return bbox_results

    def update_hyperparameters(self):
        """Update hyperparameters like IoU thresholds for assigner and beta for
        SmoothL1 loss based on the training statistics.

        Returns:
            tuple[float]: the updated ``iou_thr`` and ``beta``.
        """
        new_iou_thr = max(self.train_cfg.dynamic_rcnn.initial_iou,
                          np.mean(self.iou_history))
        self.iou_history = []
        self.bbox_assigner.pos_iou_thr = new_iou_thr
        self.bbox_assigner.neg_iou_thr = new_iou_thr
        self.bbox_assigner.min_pos_iou = new_iou_thr
        if (not self.beta_history) or (np.median(self.beta_history) < EPS):
            # avoid 0 or too small value for new_beta
            new_beta = self.bbox_head.loss_bbox.beta
        else:
            new_beta = min(self.train_cfg.dynamic_rcnn.initial_beta,
                           np.median(self.beta_history))
        self.beta_history = []
        self.bbox_head.loss_bbox.beta = new_beta
        return new_iou_thr, new_beta