Python mmdet.ops.nms() Examples
The following are 30
code examples of mmdet.ops.nms().
You can vote up the ones you like or vote down the ones you don't like,
and go to the original project or source file by following the links above each example.
You may also want to check out all available functions/classes of the module
mmdet.ops
, or try the search function
.
.
Example #1
| Source File: merge_augs.py From kaggle-kuzushiji-recognition with MIT License | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #2
| Source File: merge_augs.py From Cascade-RPN with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #3
| Source File: merge_augs.py From FoveaBox with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #4
| Source File: merge_augs.py From Feature-Selective-Anchor-Free-Module-for-Single-Shot-Object-Detection with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #5
| Source File: merge_augs.py From Libra_R-CNN with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #6
| Source File: merge_augs.py From Reasoning-RCNN with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #7
| Source File: merge_augs.py From mmdetection with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): list of image info dict where each dict has:
'img_shape', 'scale_factor', 'flip', and my also contain
'filename', 'ori_shape', 'pad_shape', and 'img_norm_cfg'.
For details on the values of these keys see
`mmdet/datasets/pipelines/formatting.py:Collect`.
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
flip_direction = img_info['flip_direction']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip,
flip_direction)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #8
| Source File: merge_augs.py From IoU-Uniform-R-CNN with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): list of image info dict where each dict has:
'img_shape', 'scale_factor', 'flip', and my also contain
'filename', 'ori_shape', 'pad_shape', and 'img_norm_cfg'.
For details on the values of these keys see
`mmdet/datasets/pipelines/formatting.py:Collect`.
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #9
| Source File: merge_augs.py From kaggle-imaterialist with MIT License | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #10
| Source File: merge_augs.py From RDSNet with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #11
| Source File: merge_augs.py From Grid-R-CNN with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #12
| Source File: merge_augs.py From hrnet with MIT License | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #13
| Source File: merge_augs.py From mmdetection-annotated with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #14
| Source File: merge_augs.py From AugFPN with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #15
| Source File: merge_augs_rotate.py From AerialDetection with Apache License 2.0 | 5 votes |
|
def merge_rotate_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
# rotation mapping
angle = img_info['angle']
if angle != 0:
# TODO: check the angle
_proposals[:, :4] = bbox_rotate_mapping(_proposals[:, :4], img_shape, -angle)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #16
| Source File: merge_augs.py From AerialDetection with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #17
| Source File: merge_augs.py From GCNet with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #18
| Source File: merge_augs.py From ttfnet with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): list of image info dict where each dict has:
'img_shape', 'scale_factor', 'flip', and my also contain
'filename', 'ori_shape', 'pad_shape', and 'img_norm_cfg'.
For details on the values of these keys see
`mmdet/datasets/pipelines/formatting.py:Collect`.
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #19
| Source File: make_submission.py From kaggle-kuzushiji-recognition with MIT License | 5 votes |
|
def post_process(preds, num_classes=4787, iou_thr=0.3, score_thr=0.3):
ret = []
for pred in tqdm(preds):
bboxes = np.vstack(pred)
labels = np.concatenate([[i] * len(bb) for i, bb in enumerate(pred)])
# nms
_, inds = nms(bboxes, iou_thr)
bboxes, labels = bboxes[inds], labels[inds]
# score filtering
inds = bboxes[:, 4] > score_thr
bboxes, labels = bboxes[inds], labels[inds]
#
ret.append([bboxes[labels == i] for i in range(num_classes)])
return ret
Example #20
| Source File: merge_augs.py From mmdetection_with_SENet154 with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #21
| Source File: merge_augs.py From CenterNet with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #22
| Source File: merge_augs.py From PolarMask with Apache License 2.0 | 5 votes |
|
def merge_aug_proposals(aug_proposals, img_metas, rpn_test_cfg):
"""Merge augmented proposals (multiscale, flip, etc.)
Args:
aug_proposals (list[Tensor]): proposals from different testing
schemes, shape (n, 5). Note that they are not rescaled to the
original image size.
img_metas (list[dict]): image info including "shape_scale" and "flip".
rpn_test_cfg (dict): rpn test config.
Returns:
Tensor: shape (n, 4), proposals corresponding to original image scale.
"""
recovered_proposals = []
for proposals, img_info in zip(aug_proposals, img_metas):
img_shape = img_info['img_shape']
scale_factor = img_info['scale_factor']
flip = img_info['flip']
_proposals = proposals.clone()
_proposals[:, :4] = bbox_mapping_back(_proposals[:, :4], img_shape,
scale_factor, flip)
recovered_proposals.append(_proposals)
aug_proposals = torch.cat(recovered_proposals, dim=0)
merged_proposals, _ = nms(aug_proposals, rpn_test_cfg.nms_thr)
scores = merged_proposals[:, 4]
_, order = scores.sort(0, descending=True)
num = min(rpn_test_cfg.max_num, merged_proposals.shape[0])
order = order[:num]
merged_proposals = merged_proposals[order, :]
return merged_proposals
Example #23
| Source File: rpn_head.py From ttfnet with Apache License 2.0 | 4 votes |
|
def get_bboxes_single(self,
cls_scores,
bbox_preds,
mlvl_anchors,
img_shape,
scale_factor,
cfg,
rescale=False):
mlvl_proposals = []
for idx in range(len(cls_scores)):
rpn_cls_score = cls_scores[idx]
rpn_bbox_pred = bbox_preds[idx]
assert rpn_cls_score.size()[-2:] == rpn_bbox_pred.size()[-2:]
anchors = mlvl_anchors[idx]
rpn_cls_score = rpn_cls_score.permute(1, 2, 0)
if self.use_sigmoid_cls:
rpn_cls_score = rpn_cls_score.reshape(-1)
scores = rpn_cls_score.sigmoid()
else:
rpn_cls_score = rpn_cls_score.reshape(-1, 2)
scores = rpn_cls_score.softmax(dim=1)[:, 1]
rpn_bbox_pred = rpn_bbox_pred.permute(1, 2, 0).reshape(-1, 4)
if cfg.nms_pre > 0 and scores.shape[0] > cfg.nms_pre:
_, topk_inds = scores.topk(cfg.nms_pre)
rpn_bbox_pred = rpn_bbox_pred[topk_inds, :]
anchors = anchors[topk_inds, :]
scores = scores[topk_inds]
proposals = delta2bbox(anchors, rpn_bbox_pred, self.target_means,
self.target_stds, img_shape)
if cfg.min_bbox_size > 0:
w = proposals[:, 2] - proposals[:, 0] + 1
h = proposals[:, 3] - proposals[:, 1] + 1
valid_inds = torch.nonzero((w >= cfg.min_bbox_size) &
(h >= cfg.min_bbox_size)).squeeze()
proposals = proposals[valid_inds, :]
scores = scores[valid_inds]
proposals = torch.cat([proposals, scores.unsqueeze(-1)], dim=-1)
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.nms_post, :]
mlvl_proposals.append(proposals)
proposals = torch.cat(mlvl_proposals, 0)
if cfg.nms_across_levels:
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.max_num, :]
else:
scores = proposals[:, 4]
num = min(cfg.max_num, proposals.shape[0])
_, topk_inds = scores.topk(num)
proposals = proposals[topk_inds, :]
return proposals
Example #24
| Source File: rpn_head.py From hrnet with MIT License | 4 votes |
|
def get_bboxes_single(self,
cls_scores,
bbox_preds,
mlvl_anchors,
img_shape,
scale_factor,
cfg,
rescale=False):
mlvl_proposals = []
for idx in range(len(cls_scores)):
rpn_cls_score = cls_scores[idx]
rpn_bbox_pred = bbox_preds[idx]
assert rpn_cls_score.size()[-2:] == rpn_bbox_pred.size()[-2:]
anchors = mlvl_anchors[idx]
rpn_cls_score = rpn_cls_score.permute(1, 2, 0)
if self.use_sigmoid_cls:
rpn_cls_score = rpn_cls_score.reshape(-1)
scores = rpn_cls_score.sigmoid()
else:
rpn_cls_score = rpn_cls_score.reshape(-1, 2)
scores = rpn_cls_score.softmax(dim=1)[:, 1]
rpn_bbox_pred = rpn_bbox_pred.permute(1, 2, 0).reshape(-1, 4)
if cfg.nms_pre > 0 and scores.shape[0] > cfg.nms_pre:
_, topk_inds = scores.topk(cfg.nms_pre)
rpn_bbox_pred = rpn_bbox_pred[topk_inds, :]
anchors = anchors[topk_inds, :]
scores = scores[topk_inds]
proposals = delta2bbox(anchors, rpn_bbox_pred, self.target_means,
self.target_stds, img_shape)
if cfg.min_bbox_size > 0:
w = proposals[:, 2] - proposals[:, 0] + 1
h = proposals[:, 3] - proposals[:, 1] + 1
valid_inds = torch.nonzero((w >= cfg.min_bbox_size) &
(h >= cfg.min_bbox_size)).squeeze()
proposals = proposals[valid_inds, :]
scores = scores[valid_inds]
proposals = torch.cat([proposals, scores.unsqueeze(-1)], dim=-1)
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.nms_post, :]
mlvl_proposals.append(proposals)
proposals = torch.cat(mlvl_proposals, 0)
if cfg.nms_across_levels:
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.max_num, :]
else:
scores = proposals[:, 4]
num = min(cfg.max_num, proposals.shape[0])
_, topk_inds = scores.topk(num)
proposals = proposals[topk_inds, :]
return proposals
Example #25
| Source File: rpn_head.py From kaggle-imaterialist with MIT License | 4 votes |
|
def get_bboxes_single(self,
cls_scores,
bbox_preds,
mlvl_anchors,
img_shape,
scale_factor,
cfg,
rescale=False):
mlvl_proposals = []
for idx in range(len(cls_scores)):
rpn_cls_score = cls_scores[idx]
rpn_bbox_pred = bbox_preds[idx]
assert rpn_cls_score.size()[-2:] == rpn_bbox_pred.size()[-2:]
anchors = mlvl_anchors[idx]
rpn_cls_score = rpn_cls_score.permute(1, 2, 0)
if self.use_sigmoid_cls:
rpn_cls_score = rpn_cls_score.reshape(-1)
scores = rpn_cls_score.sigmoid()
else:
rpn_cls_score = rpn_cls_score.reshape(-1, 2)
scores = rpn_cls_score.softmax(dim=1)[:, 1]
rpn_bbox_pred = rpn_bbox_pred.permute(1, 2, 0).reshape(-1, 4)
if cfg.nms_pre > 0 and scores.shape[0] > cfg.nms_pre:
_, topk_inds = scores.topk(cfg.nms_pre)
rpn_bbox_pred = rpn_bbox_pred[topk_inds, :]
anchors = anchors[topk_inds, :]
scores = scores[topk_inds]
proposals = delta2bbox(anchors, rpn_bbox_pred, self.target_means,
self.target_stds, img_shape)
if cfg.min_bbox_size > 0:
w = proposals[:, 2] - proposals[:, 0] + 1
h = proposals[:, 3] - proposals[:, 1] + 1
valid_inds = torch.nonzero((w >= cfg.min_bbox_size) &
(h >= cfg.min_bbox_size)).squeeze()
proposals = proposals[valid_inds, :]
scores = scores[valid_inds]
proposals = torch.cat([proposals, scores.unsqueeze(-1)], dim=-1)
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.nms_post, :]
mlvl_proposals.append(proposals)
proposals = torch.cat(mlvl_proposals, 0)
if cfg.nms_across_levels:
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.max_num, :]
else:
scores = proposals[:, 4]
num = min(cfg.max_num, proposals.shape[0])
_, topk_inds = scores.topk(num)
proposals = proposals[topk_inds, :]
return proposals
Example #26
| Source File: rpn_head.py From AugFPN with Apache License 2.0 | 4 votes |
|
def get_bboxes_single(self,
cls_scores,
bbox_preds,
mlvl_anchors,
img_shape,
scale_factor,
cfg,
rescale=False):
mlvl_proposals = []
for idx in range(len(cls_scores)):
rpn_cls_score = cls_scores[idx]
rpn_bbox_pred = bbox_preds[idx]
assert rpn_cls_score.size()[-2:] == rpn_bbox_pred.size()[-2:]
anchors = mlvl_anchors[idx]
rpn_cls_score = rpn_cls_score.permute(1, 2, 0)
if self.use_sigmoid_cls:
rpn_cls_score = rpn_cls_score.reshape(-1)
scores = rpn_cls_score.sigmoid()
else:
rpn_cls_score = rpn_cls_score.reshape(-1, 2)
scores = rpn_cls_score.softmax(dim=1)[:, 1]
rpn_bbox_pred = rpn_bbox_pred.permute(1, 2, 0).reshape(-1, 4)
if cfg.nms_pre > 0 and scores.shape[0] > cfg.nms_pre:
_, topk_inds = scores.topk(cfg.nms_pre)
rpn_bbox_pred = rpn_bbox_pred[topk_inds, :]
anchors = anchors[topk_inds, :]
scores = scores[topk_inds]
proposals = delta2bbox(anchors, rpn_bbox_pred, self.target_means,
self.target_stds, img_shape)
if cfg.min_bbox_size > 0:
w = proposals[:, 2] - proposals[:, 0] + 1
h = proposals[:, 3] - proposals[:, 1] + 1
valid_inds = torch.nonzero((w >= cfg.min_bbox_size) &
(h >= cfg.min_bbox_size)).squeeze()
proposals = proposals[valid_inds, :]
scores = scores[valid_inds]
proposals = torch.cat([proposals, scores.unsqueeze(-1)], dim=-1)
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.nms_post, :]
mlvl_proposals.append(proposals)
proposals = torch.cat(mlvl_proposals, 0)
if cfg.nms_across_levels:
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.max_num, :]
else:
scores = proposals[:, 4]
num = min(cfg.max_num, proposals.shape[0])
_, topk_inds = scores.topk(num)
proposals = proposals[topk_inds, :]
return proposals
Example #27
| Source File: rpn_head.py From Reasoning-RCNN with Apache License 2.0 | 4 votes |
|
def get_bboxes_single(self,
cls_scores,
bbox_preds,
mlvl_anchors,
img_shape,
scale_factor,
cfg,
rescale=False):
mlvl_proposals = []
for idx in range(len(cls_scores)):
rpn_cls_score = cls_scores[idx]
rpn_bbox_pred = bbox_preds[idx]
assert rpn_cls_score.size()[-2:] == rpn_bbox_pred.size()[-2:]
anchors = mlvl_anchors[idx]
rpn_cls_score = rpn_cls_score.permute(1, 2, 0)
if self.use_sigmoid_cls:
rpn_cls_score = rpn_cls_score.reshape(-1)
scores = rpn_cls_score.sigmoid()
else:
rpn_cls_score = rpn_cls_score.reshape(-1, 2)
scores = rpn_cls_score.softmax(dim=1)[:, 1]
rpn_bbox_pred = rpn_bbox_pred.permute(1, 2, 0).reshape(-1, 4)
if cfg.nms_pre > 0 and scores.shape[0] > cfg.nms_pre:
_, topk_inds = scores.topk(cfg.nms_pre)
rpn_bbox_pred = rpn_bbox_pred[topk_inds, :]
anchors = anchors[topk_inds, :]
scores = scores[topk_inds]
proposals = delta2bbox(anchors, rpn_bbox_pred, self.target_means,
self.target_stds, img_shape)
if cfg.min_bbox_size > 0:
w = proposals[:, 2] - proposals[:, 0] + 1
h = proposals[:, 3] - proposals[:, 1] + 1
valid_inds = torch.nonzero((w >= cfg.min_bbox_size) &
(h >= cfg.min_bbox_size)).squeeze()
proposals = proposals[valid_inds, :]
scores = scores[valid_inds]
proposals = torch.cat([proposals, scores.unsqueeze(-1)], dim=-1)
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.nms_post, :]
mlvl_proposals.append(proposals)
proposals = torch.cat(mlvl_proposals, 0)
if cfg.nms_across_levels:
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.max_num, :]
else:
scores = proposals[:, 4]
num = min(cfg.max_num, proposals.shape[0])
_, topk_inds = scores.topk(num)
proposals = proposals[topk_inds, :]
return proposals
Example #28
| Source File: rpn_head.py From Feature-Selective-Anchor-Free-Module-for-Single-Shot-Object-Detection with Apache License 2.0 | 4 votes |
|
def get_bboxes_single(self,
cls_scores,
bbox_preds,
mlvl_anchors,
img_shape,
scale_factor,
cfg,
rescale=False):
mlvl_proposals = []
for idx in range(len(cls_scores)):
rpn_cls_score = cls_scores[idx]
rpn_bbox_pred = bbox_preds[idx]
assert rpn_cls_score.size()[-2:] == rpn_bbox_pred.size()[-2:]
anchors = mlvl_anchors[idx]
rpn_cls_score = rpn_cls_score.permute(1, 2, 0)
if self.use_sigmoid_cls:
rpn_cls_score = rpn_cls_score.reshape(-1)
scores = rpn_cls_score.sigmoid()
else:
rpn_cls_score = rpn_cls_score.reshape(-1, 2)
scores = rpn_cls_score.softmax(dim=1)[:, 1]
rpn_bbox_pred = rpn_bbox_pred.permute(1, 2, 0).reshape(-1, 4)
if cfg.nms_pre > 0 and scores.shape[0] > cfg.nms_pre:
_, topk_inds = scores.topk(cfg.nms_pre)
rpn_bbox_pred = rpn_bbox_pred[topk_inds, :]
anchors = anchors[topk_inds, :]
scores = scores[topk_inds]
proposals = delta2bbox(anchors, rpn_bbox_pred, self.target_means,
self.target_stds, img_shape)
if cfg.min_bbox_size > 0:
w = proposals[:, 2] - proposals[:, 0] + 1
h = proposals[:, 3] - proposals[:, 1] + 1
valid_inds = torch.nonzero((w >= cfg.min_bbox_size) &
(h >= cfg.min_bbox_size)).squeeze()
proposals = proposals[valid_inds, :]
scores = scores[valid_inds]
proposals = torch.cat([proposals, scores.unsqueeze(-1)], dim=-1)
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.nms_post, :]
mlvl_proposals.append(proposals)
proposals = torch.cat(mlvl_proposals, 0)
if cfg.nms_across_levels:
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.max_num, :]
else:
scores = proposals[:, 4]
num = min(cfg.max_num, proposals.shape[0])
_, topk_inds = scores.topk(num)
proposals = proposals[topk_inds, :]
return proposals
Example #29
| Source File: rpn_head.py From CenterNet with Apache License 2.0 | 4 votes |
|
def get_bboxes_single(self,
cls_scores,
bbox_preds,
mlvl_anchors,
img_shape,
scale_factor,
cfg,
rescale=False):
mlvl_proposals = []
for idx in range(len(cls_scores)):
rpn_cls_score = cls_scores[idx]
rpn_bbox_pred = bbox_preds[idx]
assert rpn_cls_score.size()[-2:] == rpn_bbox_pred.size()[-2:]
anchors = mlvl_anchors[idx]
rpn_cls_score = rpn_cls_score.permute(1, 2, 0)
if self.use_sigmoid_cls:
rpn_cls_score = rpn_cls_score.reshape(-1)
scores = rpn_cls_score.sigmoid()
else:
rpn_cls_score = rpn_cls_score.reshape(-1, 2)
scores = rpn_cls_score.softmax(dim=1)[:, 1]
rpn_bbox_pred = rpn_bbox_pred.permute(1, 2, 0).reshape(-1, 4)
if cfg.nms_pre > 0 and scores.shape[0] > cfg.nms_pre:
_, topk_inds = scores.topk(cfg.nms_pre)
rpn_bbox_pred = rpn_bbox_pred[topk_inds, :]
anchors = anchors[topk_inds, :]
scores = scores[topk_inds]
proposals = delta2bbox(anchors, rpn_bbox_pred, self.target_means,
self.target_stds, img_shape)
if cfg.min_bbox_size > 0:
w = proposals[:, 2] - proposals[:, 0] + 1
h = proposals[:, 3] - proposals[:, 1] + 1
valid_inds = torch.nonzero((w >= cfg.min_bbox_size) &
(h >= cfg.min_bbox_size)).squeeze()
proposals = proposals[valid_inds, :]
scores = scores[valid_inds]
proposals = torch.cat([proposals, scores.unsqueeze(-1)], dim=-1)
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.nms_post, :]
mlvl_proposals.append(proposals)
proposals = torch.cat(mlvl_proposals, 0)
if cfg.nms_across_levels:
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.max_num, :]
else:
scores = proposals[:, 4]
num = min(cfg.max_num, proposals.shape[0])
_, topk_inds = scores.topk(num)
proposals = proposals[topk_inds, :]
return proposals
Example #30
| Source File: cascade_rpn_head.py From Cascade-RPN with Apache License 2.0 | 4 votes |
|
def get_bboxes_single(self,
cls_scores,
bbox_preds,
mlvl_anchors,
img_shape,
scale_factor,
cfg,
rescale=False):
mlvl_proposals = []
for idx in range(len(cls_scores)):
rpn_cls_score = cls_scores[idx]
rpn_bbox_pred = bbox_preds[idx]
assert rpn_cls_score.size()[-2:] == rpn_bbox_pred.size()[-2:]
anchors = mlvl_anchors[idx]
rpn_cls_score = rpn_cls_score.permute(1, 2, 0)
if self.use_sigmoid_cls:
rpn_cls_score = rpn_cls_score.reshape(-1)
scores = rpn_cls_score.sigmoid()
else:
rpn_cls_score = rpn_cls_score.reshape(-1, 2)
scores = rpn_cls_score.softmax(dim=1)[:, 1]
rpn_bbox_pred = rpn_bbox_pred.permute(1, 2, 0).reshape(-1, 4)
if cfg.nms_pre > 0 and scores.shape[0] > cfg.nms_pre:
_, topk_inds = scores.topk(cfg.nms_pre)
rpn_bbox_pred = rpn_bbox_pred[topk_inds, :]
anchors = anchors[topk_inds, :]
scores = scores[topk_inds]
proposals = delta2bbox(anchors, rpn_bbox_pred, self.target_means,
self.target_stds, img_shape)
if cfg.min_bbox_size > 0:
w = proposals[:, 2] - proposals[:, 0] + 1
h = proposals[:, 3] - proposals[:, 1] + 1
valid_inds = torch.nonzero((w >= cfg.min_bbox_size) &
(h >= cfg.min_bbox_size)).squeeze()
proposals = proposals[valid_inds, :]
scores = scores[valid_inds]
proposals = torch.cat([proposals, scores.unsqueeze(-1)], dim=-1)
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.nms_post, :]
mlvl_proposals.append(proposals)
proposals = torch.cat(mlvl_proposals, 0)
if cfg.nms_across_levels:
proposals, _ = nms(proposals, cfg.nms_thr)
proposals = proposals[:cfg.max_num, :]
else:
scores = proposals[:, 4]
num = min(cfg.max_num, proposals.shape[0])
_, topk_inds = scores.topk(num)
proposals = proposals[topk_inds, :]
return proposals
