File size: 6,229 Bytes
fa84113
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
""" PyTorch Soft-NMS

This code was adapted from a PR for detectron2 submitted by https://github.com/alekseynp
https://github.com/facebookresearch/detectron2/pull/1183/files

Detectron2 is licensed Apache 2.0, Copyright Facebook Inc.
"""
import torch
from typing import List


def pairwise_iou(boxes1, boxes2) -> torch.Tensor:
    """
    Given two lists of boxes of size N and M,
    compute the IoU (intersection over union)
    between __all__ N x M pairs of boxes.
    The box order must be (xmin, ymin, xmax, ymax).
    Args:
        boxes1,boxes2 (Boxes): two `Boxes`. Contains N & M boxes, respectively.
    Returns:
        Tensor: IoU, sized [N,M].
    """
    area1 = (boxes1[:, 2] - boxes1[:, 0]) * (boxes1[:, 3] - boxes1[:, 1])  # [N,]
    area2 = (boxes2[:, 2] - boxes2[:, 0]) * (boxes2[:, 3] - boxes2[:, 1])  # [M,]

    width_height = torch.min(boxes1[:, None, 2:], boxes2[:, 2:]) - torch.max(
        boxes1[:, None, :2], boxes2[:, :2]
    )  # [N,M,2]

    width_height.clamp_(min=0)  # [N,M,2]
    inter = width_height.prod(dim=2)  # [N,M]

    # handle empty boxes
    iou = torch.where(
        inter > 0,
        inter / (area1[:, None] + area2 - inter),
        torch.zeros(1, dtype=inter.dtype, device=inter.device),
    )
    return iou


def soft_nms(
    boxes,
    scores,
    method_gaussian: bool = True,
    sigma: float = 0.5,
    iou_threshold: float = .5,
    score_threshold: float = 0.005
):
    """
    Soft non-max suppression algorithm.

    Implementation of [Soft-NMS -- Improving Object Detection With One Line of Codec]
    (https://arxiv.org/abs/1704.04503)

    Args:
        boxes_remain (Tensor[N, ?]):
           boxes where NMS will be performed
           if Boxes, in (x1, y1, x2, y2) format
           if RotatedBoxes, in (x_ctr, y_ctr, width, height, angle_degrees) format
        scores_remain (Tensor[N]):
           scores for each one of the boxes
        method_gaussian (bool): use gaussian method if True, otherwise linear        
        sigma (float):
           parameter for Gaussian penalty function
        iou_threshold (float):
           iou threshold for applying linear decay. Nt from the paper
           re-used as threshold for standard "hard" nms
        score_threshold (float):
           boxes with scores below this threshold are pruned at each iteration.
           Dramatically reduces computation time. Authors use values in [10e-4, 10e-2]

    Returns:
        tuple(Tensor, Tensor):
            [0]: int64 tensor with the indices of the elements that have been kept
            by Soft NMS, sorted in decreasing order of scores
            [1]: float tensor with the re-scored scores of the elements that were kept
    """
    device = boxes.device
    boxes_remain = boxes.clone()
    scores_remain = scores.clone()
    num_elem = scores_remain.size()[0]
    idxs = torch.arange(num_elem)
    idxs_out = torch.zeros(num_elem, dtype=torch.int64, device=device)
    scores_out = torch.zeros(num_elem, dtype=torch.float32, device=device)
    count: int = 0

    while scores_remain.numel() > 0:
        top_idx = torch.argmax(scores_remain)
        idxs_out[count] = idxs[top_idx]
        scores_out[count] = scores_remain[top_idx]
        count += 1

        top_box = boxes_remain[top_idx]
        ious = pairwise_iou(top_box.unsqueeze(0), boxes_remain)[0]

        if method_gaussian:
            decay = torch.exp(-torch.pow(ious, 2) / sigma)
        else:
            decay = torch.ones_like(ious)
            decay_mask = ious > iou_threshold
            decay[decay_mask] = 1 - ious[decay_mask]

        scores_remain *= decay
        keep = scores_remain > score_threshold
        keep[top_idx] = torch.tensor(False, device=device)

        boxes_remain = boxes_remain[keep]
        scores_remain = scores_remain[keep]
        idxs = idxs[keep]

    return idxs_out[:count], scores_out[:count]


def batched_soft_nms(
    boxes, scores, idxs,
    method_gaussian: bool = True,
    sigma: float = 0.5,
    iou_threshold: float = .5,
    score_threshold: float = 0.001):

    """
    Performs soft non-maximum suppression in a batched fashion.

    Each index value correspond to a category, and NMS
    will not be applied between elements of different categories.

    Args:
        boxes (Tensor[N, 4]):
           boxes where NMS will be performed. They
           are expected to be in (x1, y1, x2, y2) format
        scores (Tensor[N]):
           scores for each one of the boxes
        idxs (Tensor[N]):
           indices of the categories for each one of the boxes.
        method (str):
           one of ['gaussian', 'linear', 'hard']
           see paper for details. users encouraged not to use "hard", as this is the
           same nms available elsewhere in detectron2
        sigma (float):
           parameter for Gaussian penalty function
        iou_threshold (float):
           iou threshold for applying linear decay. Nt from the paper
           re-used as threshold for standard "hard" nms
        score_threshold (float):
           boxes with scores below this threshold are pruned at each iteration.
           Dramatically reduces computation time. Authors use values in [10e-4, 10e-2]
    Returns:
        tuple(Tensor, Tensor):
            [0]: int64 tensor with the indices of the elements that have been kept
            by Soft NMS, sorted in decreasing order of scores
            [1]: float tensor with the re-scored scores of the elements that were kept
    """
    if boxes.numel() == 0:
        return (
            torch.empty((0,), dtype=torch.int64, device=boxes.device),
            torch.empty((0,), dtype=torch.float32, device=scores.device),
        )
    # strategy: in order to perform NMS independently per class.
    # we add an offset to all the boxes. The offset is dependent
    # only on the class idx, and is large enough so that boxes
    # from different classes do not overlap
    max_coordinate = boxes.max()
    offsets = idxs.to(boxes) * (max_coordinate + 1)
    boxes_for_nms = boxes + offsets[:, None]
    return soft_nms(
        boxes_for_nms, scores, method_gaussian=method_gaussian, sigma=sigma,
        iou_threshold=iou_threshold, score_threshold=score_threshold
    )