roi_pool.py 2.73 KB
Newer Older
1
2
from typing import List, Union

3
import torch
4
from torch import nn, Tensor
5
from torch.jit.annotations import BroadcastingList2
6
from torch.nn.modules.utils import _pair
7
from torchvision.extension import _assert_has_ops
8

9
from ._utils import convert_boxes_to_roi_format, check_roi_boxes_shape
10
11


12
13
def roi_pool(
    input: Tensor,
14
    boxes: Union[Tensor, List[Tensor]],
15
16
17
    output_size: BroadcastingList2[int],
    spatial_scale: float = 1.0,
) -> Tensor:
18
19
20
    """
    Performs Region of Interest (RoI) Pool operator described in Fast R-CNN

21
    Args:
22
23
        input (Tensor[N, C, H, W]): The input tensor, i.e. a batch with ``N`` elements. Each element
            contains ``C`` feature maps of dimensions ``H x W``.
24
        boxes (Tensor[K, 5] or List[Tensor[L, 4]]): the box coordinates in (x1, y1, x2, y2)
25
26
            format where the regions will be taken from.
            The coordinate must satisfy ``0 <= x1 < x2`` and ``0 <= y1 < y2``.
27
28
29
30
            If a single Tensor is passed, then the first column should
            contain the index of the corresponding element in the batch, i.e. a number in ``[0, N - 1]``.
            If a list of Tensors is passed, then each Tensor will correspond to the boxes for an element i
            in the batch.
31
32
        output_size (int or Tuple[int, int]): the size of the output after the cropping
            is performed, as (height, width)
33
34
35
36
        spatial_scale (float): a scaling factor that maps the box coordinates to
            the input coordinates. For example, if your boxes are defined on the scale
            of a 224x224 image and your input is a 112x112 feature map (resulting from a 0.5x scaling of
            the original image), you'll want to set this to 0.5. Default: 1.0
37
38

    Returns:
39
        Tensor[K, C, output_size[0], output_size[1]]: The pooled RoIs.
40
    """
41
    _assert_has_ops()
42
    check_roi_boxes_shape(boxes)
43
    rois = boxes
44
    output_size = _pair(output_size)
45
46
    if not isinstance(rois, torch.Tensor):
        rois = convert_boxes_to_roi_format(rois)
47
    output, _ = torch.ops.torchvision.roi_pool(input, rois, spatial_scale, output_size[0], output_size[1])
48
    return output
49
50
51
52


class RoIPool(nn.Module):
    """
53
    See :func:`roi_pool`.
54
    """
55

56
    def __init__(self, output_size: BroadcastingList2[int], spatial_scale: float):
57
        super().__init__()
58
59
60
        self.output_size = output_size
        self.spatial_scale = spatial_scale

61
    def forward(self, input: Tensor, rois: Tensor) -> Tensor:
62
63
        return roi_pool(input, rois, self.output_size, self.spatial_scale)

64
    def __repr__(self) -> str:
65
66
67
68
        tmpstr = self.__class__.__name__ + "("
        tmpstr += "output_size=" + str(self.output_size)
        tmpstr += ", spatial_scale=" + str(self.spatial_scale)
        tmpstr += ")"
69
        return tmpstr