File size: 6,378 Bytes
beb7843
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import torch
import torch.nn as nn
import torch.nn.functional as F
from .matcher import SimOTA
from utils.box_ops import get_ious
from utils.distributed_utils import get_world_size, is_dist_avail_and_initialized


class SigmoidFocalLoss(object):
    def __init__(self, alpha=0.25, gamma=2.0, reduction='none'):
        self.alpha = alpha
        self.gamma = gamma
        self.reduction = reduction

    def __call__(self, logits, targets):      
        p = torch.sigmoid(logits)
        ce_loss = F.binary_cross_entropy_with_logits(input=logits, 
                                                        target=targets, 
                                                        reduction="none")
        p_t = p * targets + (1.0 - p) * (1.0 - targets)
        loss = ce_loss * ((1.0 - p_t) ** self.gamma)

        if self.alpha >= 0:
            alpha_t = self.alpha * targets + (1.0 - self.alpha) * (1.0 - targets)
            loss = alpha_t * loss

        if self.reduction == "mean":
            loss = loss.mean()

        elif self.reduction == "sum":
            loss = loss.sum()

        return loss


class Criterion(object):
    def __init__(self, args, img_size, num_classes=3, multi_hot=False):
        self.num_classes = num_classes
        self.img_size = img_size
        self.loss_conf_weight = args.loss_conf_weight
        self.loss_cls_weight = args.loss_cls_weight
        self.loss_reg_weight = args.loss_reg_weight
        self.focal_loss = args.focal_loss
        self.multi_hot = multi_hot

        # loss
        self.obj_lossf = nn.BCEWithLogitsLoss(reduction='none')
        self.cls_lossf = nn.BCEWithLogitsLoss(reduction='none')
            
        # matcher
        self.matcher = SimOTA(
            num_classes=num_classes,
            center_sampling_radius=args.center_sampling_radius,
            topk_candidate=args.topk_candicate
            )

    def __call__(self, outputs, targets):        
        """
            outputs['pred_conf']: List(Tensor) [B, M, 1]
            outputs['pred_cls']: List(Tensor) [B, M, C]
            outputs['pred_box']: List(Tensor) [B, M, 4]
            outputs['strides']: List(Int) [8, 16, 32] output stride
            targets: (List) [dict{'boxes': [...], 
                                 'labels': [...], 
                                 'orig_size': ...}, ...]
        """
        bs = outputs['pred_cls'][0].shape[0]
        device = outputs['pred_cls'][0].device
        fpn_strides = outputs['strides']
        anchors = outputs['anchors']
        # preds: [B, M, C]
        conf_preds = torch.cat(outputs['pred_conf'], dim=1)
        cls_preds = torch.cat(outputs['pred_cls'], dim=1)
        box_preds = torch.cat(outputs['pred_box'], dim=1)

        # label assignment
        cls_targets = []
        box_targets = []
        conf_targets = []
        fg_masks = []

        for batch_idx in range(bs):
            tgt_labels = targets[batch_idx]["labels"].to(device)
            tgt_bboxes = targets[batch_idx]["boxes"].to(device)

            # denormalize tgt_bbox
            tgt_bboxes *= self.img_size

            # check target
            if len(tgt_labels) == 0 or tgt_bboxes.max().item() == 0.:
                num_anchors = sum([ab.shape[0] for ab in anchors])
                # There is no valid gt
                cls_target = conf_preds.new_zeros((0, self.num_classes))
                box_target = conf_preds.new_zeros((0, 4))
                conf_target = conf_preds.new_zeros((num_anchors, 1))
                fg_mask = conf_preds.new_zeros(num_anchors).bool()
            else:
                (
                    gt_matched_classes,
                    fg_mask,
                    pred_ious_this_matching,
                    matched_gt_inds,
                    num_fg_img,
                ) = self.matcher(
                    fpn_strides = fpn_strides,
                    anchors = anchors,
                    pred_conf = conf_preds[batch_idx],
                    pred_cls = cls_preds[batch_idx], 
                    pred_box = box_preds[batch_idx],
                    tgt_labels = tgt_labels,
                    tgt_bboxes = tgt_bboxes,
                    )

                conf_target = fg_mask.unsqueeze(-1)
                box_target = tgt_bboxes[matched_gt_inds]
                if self.multi_hot:
                    cls_target = gt_matched_classes.float()
                else:
                    cls_target = F.one_hot(gt_matched_classes.long(), self.num_classes)
                cls_target = cls_target * pred_ious_this_matching.unsqueeze(-1)

            cls_targets.append(cls_target)
            box_targets.append(box_target)
            conf_targets.append(conf_target)
            fg_masks.append(fg_mask)

        cls_targets = torch.cat(cls_targets, 0)
        box_targets = torch.cat(box_targets, 0)
        conf_targets = torch.cat(conf_targets, 0)
        fg_masks = torch.cat(fg_masks, 0)
        num_foregrounds = fg_masks.sum()

        if is_dist_avail_and_initialized():
            torch.distributed.all_reduce(num_foregrounds)
        num_foregrounds = (num_foregrounds / get_world_size()).clamp(1.0)

        # conf loss
        loss_conf = self.obj_lossf(conf_preds.view(-1, 1), conf_targets.float())
        loss_conf = loss_conf.sum() / num_foregrounds
        
        # cls loss
        matched_cls_preds = cls_preds.view(-1, self.num_classes)[fg_masks]
        loss_cls = self.cls_lossf(matched_cls_preds, cls_targets)
        loss_cls = loss_cls.sum() / num_foregrounds

        # box loss
        matched_box_preds = box_preds.view(-1, 4)[fg_masks]
        ious = get_ious(matched_box_preds,
                        box_targets,
                        box_mode="xyxy",
                        iou_type='giou')
        loss_box = (1.0 - ious).sum() / num_foregrounds

        # total loss
        losses = self.loss_conf_weight * loss_conf + \
                 self.loss_cls_weight * loss_cls + \
                 self.loss_reg_weight * loss_box

        loss_dict = dict(
                loss_conf = loss_conf,
                loss_cls = loss_cls,
                loss_box = loss_box,
                losses = losses
        )

        return loss_dict


def build_criterion(args, img_size, num_classes, multi_hot=False):
    criterion = Criterion(args, img_size, num_classes, multi_hot)
    
    return criterion