def loss(self,
cls_scores,
bbox_preds,
gt_bboxes,
gt_labels,
img_metas,
cfg,
gt_bboxes_ignore=None):
featmap_sizes = [featmap.size()[-2:] for featmap in cls_scores]
assert len(featmap_sizes) == len(self.anchor_generators)
device = cls_scores[0].device
anchor_list, valid_flag_list = self.get_anchors(featmap_sizes,
img_metas,
device=device)
label_channels = self.cls_out_channels if self.use_sigmoid_cls else 1
cls_reg_targets = anchor_target_rotated(
anchor_list,
valid_flag_list,
gt_bboxes,
img_metas,
self.target_means,
self.target_stds,
cfg,
bbox_preds=bbox_preds,
gt_bboxes_ignore_list=gt_bboxes_ignore,
gt_labels_list=gt_labels,
label_channels=label_channels,
sampling=self.sampling)
if cls_reg_targets is None:
return None
(labels_list, label_weights_list, bbox_targets_list, bbox_weights_list,
num_total_pos, num_total_neg) = cls_reg_targets
num_total_samples = (num_total_pos +
num_total_neg if self.sampling else num_total_pos)
anchor_list, valid_flag_list = self.get_anchors(featmap_sizes,
img_metas,
device=device)
nB = len(anchor_list)
nL = len(anchor_list[0])
anchors = [[0] * nB for _ in range(nL)]
for nb in range(nB):
for nl in range(nL):
anchors[nl][nb] = anchor_list[nb][nl]
anchor_list = [torch.stack(_anchors) for _anchors in anchors]
losses_cls, losses_bbox = multi_apply(
self.loss_single,
cls_scores,
bbox_preds,
labels_list,
label_weights_list,
bbox_targets_list,
bbox_weights_list,
anchor_list,
num_total_samples=num_total_samples,
cfg=cfg)
return dict(loss_cls=losses_cls, loss_bbox=losses_bbox)
def loss(self,
fam_cls_scores,
fam_bbox_preds,
refine_anchors,
odm_cls_scores,
odm_bbox_preds,
gt_bboxes,
gt_labels,
img_metas,
cfg,
gt_bboxes_ignore=None):
featmap_sizes = [featmap.size()[-2:] for featmap in odm_cls_scores]
assert len(featmap_sizes) == len(self.anchor_generators)
device = odm_cls_scores[0].device
anchors_list, valid_flag_list = self.get_init_anchors(featmap_sizes,
img_metas,
device=device)
# Feature Alignment Module tagets
label_channels = self.cls_out_channels if self.use_sigmoid_cls else 1
cls_reg_targets = anchor_target_rotated(
anchors_list,
valid_flag_list,
gt_bboxes,
img_metas,
self.target_means,
self.target_stds,
cfg.fam_cfg,
bbox_preds=fam_bbox_preds,
gt_bboxes_ignore_list=gt_bboxes_ignore,
gt_labels_list=gt_labels,
label_channels=label_channels,
sampling=self.sampling)
if cls_reg_targets is None:
return None
(labels_list, label_weights_list, bbox_targets_list, bbox_weights_list,
num_total_pos, num_total_neg) = cls_reg_targets
num_total_samples = (num_total_pos +
num_total_neg if self.sampling else num_total_pos)
losses_fam_cls, losses_fam_bbox = multi_apply(
self.loss_fam_single,
fam_cls_scores,
fam_bbox_preds,
labels_list,
label_weights_list,
bbox_targets_list,
bbox_weights_list,
num_total_samples=num_total_samples,
cfg=cfg.fam_cfg)
# Object Detection Module targets
refine_anchors_list, valid_flag_list = self.get_refine_anchors(
featmap_sizes, refine_anchors, img_metas, device=device)
label_channels = self.cls_out_channels if self.use_sigmoid_cls else 1
cls_reg_targets = anchor_target_rotated(
refine_anchors_list,
valid_flag_list,
gt_bboxes,
img_metas,
self.target_means,
self.target_stds,
cfg.odm_cfg,
bbox_preds=odm_bbox_preds,
gt_bboxes_ignore_list=gt_bboxes_ignore,
gt_labels_list=gt_labels,
label_channels=label_channels,
sampling=self.sampling)
if cls_reg_targets is None:
return None
(labels_list, label_weights_list, bbox_targets_list, bbox_weights_list,
num_total_pos, num_total_neg) = cls_reg_targets
num_total_samples = (num_total_pos +
num_total_neg if self.sampling else num_total_pos)
losses_odm_cls, losses_odm_bbox = multi_apply(
self.loss_odm_single,
odm_cls_scores,
odm_bbox_preds,
labels_list,
label_weights_list,
bbox_targets_list,
bbox_weights_list,
refine_anchors,
num_total_samples=num_total_samples,
cfg=cfg.odm_cfg)
return dict(loss_fam_cls=losses_fam_cls,
loss_fam_bbox=losses_fam_bbox,
loss_odm_cls=losses_odm_cls,
loss_odm_bbox=losses_odm_bbox)
def loss(self,
ref_bbox_preds,
refine_anchors,
cls_scores,
bbox_preds,
gt_bboxes,
gt_labels,
img_metas,
cfg,
gt_bboxes_ignore=None):
featmap_sizes = [featmap.size()[-2:] for featmap in cls_scores]
assert len(featmap_sizes) == len(self.anchor_generators)
device = cls_scores[0].device
anchors_list, valid_flag_list = self.get_init_anchors(
featmap_sizes, img_metas, device=device)
# Anchor Refinement Module loss
label_channels = self.cls_out_channels if self.use_sigmoid_cls else 1
cls_reg_targets = anchor_target_rotated(
anchors_list,
valid_flag_list,
gt_bboxes,
img_metas,
self.target_means,
self.target_stds,
cfg.ref_cfg,
bbox_preds=ref_bbox_preds,
gt_bboxes_ignore_list=gt_bboxes_ignore,
gt_labels_list=gt_labels,
label_channels=label_channels,
sampling=self.sampling)
if cls_reg_targets is None:
return None
(labels_list, label_weights_list, bbox_targets_list, bbox_weights_list,
num_total_pos, num_total_neg) = cls_reg_targets
num_total_samples = (
num_total_pos + num_total_neg if self.sampling else num_total_pos)
losses_ref_bbox = []
for level in range(len(ref_bbox_preds)):
losses_ref_bbox.append(self.loss_ref_single(ref_bbox_preds[level],
bbox_targets_list[level],
bbox_weights_list[level],
num_total_samples=num_total_samples,
cfg=cfg.ref_cfg))
# Detection Module loss
refine_anchors_list, valid_flag_list = self.get_refine_anchors(
featmap_sizes, refine_anchors, img_metas, device=device)
label_channels = self.cls_out_channels if self.use_sigmoid_cls else 1
cls_reg_targets = anchor_target_rotated(
refine_anchors_list,
valid_flag_list,
gt_bboxes,
img_metas,
self.target_means,
self.target_stds,
cfg.odm_cfg,
bbox_preds=bbox_preds,
gt_bboxes_ignore_list=gt_bboxes_ignore,
gt_labels_list=gt_labels,
label_channels=label_channels,
sampling=self.sampling)
if cls_reg_targets is None:
return None
(labels_list, label_weights_list, bbox_targets_list, bbox_weights_list,
num_total_pos, num_total_neg) = cls_reg_targets
num_total_samples = (
num_total_pos + num_total_neg if self.sampling else num_total_pos)
losses_odm_cls, losses_odm_bbox = multi_apply(
self.loss_odm_single,
cls_scores,
bbox_preds,
labels_list,
label_weights_list,
bbox_targets_list,
bbox_weights_list,
num_total_samples=num_total_samples,
cfg=cfg.odm_cfg)
return dict(loss_ref_bbox=losses_ref_bbox,
loss_odm_cls=losses_odm_cls,
loss_odm_bbox=losses_odm_bbox)