mmdetectiion中anchor的计算

mmdetection通过指定anchor_scales，anchor_ratios,anchor_strides来生成anchor。

anchor_scales是通过scales_per_octave和octave_base_scales计算得到的。计算过程位于：mmdetection/mmdetection/mmdet/models/anchor_heads/retina_head.py

octave_scales = np.array([2**(i / scales_per_octave) for i in range(scales_per_octave)])

anchor_scales = octave_scales * octave_base_scale

anchor_strides取决于特征图位于第几个strides,我们可以调整anchor_scales.

相关函数：

mmdet/models/anchor_heads/anchor_head.py

mmdet/core/anchor/anchor_generator.py

通过计算，在每个层次的特征图上都生成anchor.

（1）我们网络最后需要的anchors是N*4的tensor,其中N是每个cell对应的anchor的数目，4是该anchor在原图中对应的左上角坐标和右下角坐标，我们根据这个坐标去对应的特征图中裁剪特征。

为了得到一个feature_map中每个cell的左上和右下角坐标（get_anchors），我们可以先得到cell(0,0)所对应的anchor的左上角和右下角坐标(gen_base_anchors)，然后根据偏移量依次偏移得到一个feature_map上所有cell的左上和右下角坐标。

(2)为了得到cell（0,0）对应的anchors的左上角坐标和右下角坐标，我们先计算cell(0,0)所对应anchors的中心点和长宽。

假设该feature_map相对于原图的stride=8(即下采样8倍），那么cell(0,0)所对应的区域就是原图的[0:8,0:8]。这样，cell(0,0)的中心点为（3.5,3.5）。

anchors的长宽由该feature_map的stride(作为anchors_strides),指定的anchors_scales和anchor_ratios构成。其中

w = anchor_stride
h = anchor_strife

h_ratios = torch.sqrt(anchor_ratios)
w_ratios = 1 / h_ratios

ws = (w * w_ratios* anchor_scales)
hs = (h * h_ratios * anchor_scales)

(3)因此，当我们想要引入自己聚类得到的anchors时，首先根据该anchors的长宽决定它在哪一层的feature_map中，然后通过计算得到对应的anchors_scales和anchor_ratios.

import torch


class AnchorGenerator(object):
    """
    Examples:
        >>> from mmdet.core import AnchorGenerator
        >>> self = AnchorGenerator(9, [1.], [1.])
        >>> all_anchors = self.grid_anchors((2, 2), device='cpu')
        >>> print(all_anchors)
        tensor([[ 0.,  0.,  8.,  8.],
                [16.,  0., 24.,  8.],
                [ 0., 16.,  8., 24.],
                [16., 16., 24., 24.]])
    """

    def __init__(self, base_size, scales, ratios, scale_major=True, ctr=None):
        self.base_size = base_size
        self.scales = torch.Tensor(scales)
        self.ratios = torch.Tensor(ratios)
        self.scale_major = scale_major
        self.ctr = ctr
        self.base_anchors = self.gen_base_anchors()

    @property
    def num_base_anchors(self):
        return self.base_anchors.size(0)

    def gen_base_anchors(self):
        w = self.base_size
        h = self.base_size
        if self.ctr is None:
            x_ctr = 0.5 * (w - 1)
            y_ctr = 0.5 * (h - 1)
        else:
            x_ctr, y_ctr = self.ctr

        h_ratios = torch.sqrt(self.ratios)
        w_ratios = 1 / h_ratios
        if self.scale_major:
            ws = (w * w_ratios[:, None] * self.scales[None, :]).view(-1)
            hs = (h * h_ratios[:, None] * self.scales[None, :]).view(-1)
        else:
            ws = (w * self.scales[:, None] * w_ratios[None, :]).view(-1)
            hs = (h * self.scales[:, None] * h_ratios[None, :]).view(-1)

        # yapf: disable
        base_anchors = torch.stack(
            [
                x_ctr - 0.5 * (ws - 1), y_ctr - 0.5 * (hs - 1),
                x_ctr + 0.5 * (ws - 1), y_ctr + 0.5 * (hs - 1)
            ],
            dim=-1).round()
        # yapf: enable

        return base_anchors

    def _meshgrid(self, x, y, row_major=True):
        xx = x.repeat(len(y))
        yy = y.view(-1, 1).repeat(1, len(x)).view(-1)
        if row_major:
            return xx, yy
        else:
            return yy, xx

    def grid_anchors(self, featmap_size, stride=16, device='cuda'):
        base_anchors = self.base_anchors.to(device)

        feat_h, feat_w = featmap_size
        shift_x = torch.arange(0, feat_w, device=device) * stride
        shift_y = torch.arange(0, feat_h, device=device) * stride
        shift_xx, shift_yy = self._meshgrid(shift_x, shift_y)
        shifts = torch.stack([shift_xx, shift_yy, shift_xx, shift_yy], dim=-1)
        shifts = shifts.type_as(base_anchors)
        # first feat_w elements correspond to the first row of shifts
        # add A anchors (1, A, 4) to K shifts (K, 1, 4) to get
        # shifted anchors (K, A, 4), reshape to (K*A, 4)

        all_anchors = base_anchors[None, :, :] + shifts[:, None, :]
        all_anchors = all_anchors.view(-1, 4)
        # first A rows correspond to A anchors of (0, 0) in feature map,
        # then (0, 1), (0, 2), ...
        return all_anchors

    def valid_flags(self, featmap_size, valid_size, device='cuda'):
        feat_h, feat_w = featmap_size
        valid_h, valid_w = valid_size
        assert valid_h <= feat_h and valid_w <= feat_w
        valid_x = torch.zeros(feat_w, dtype=torch.uint8, device=device)
        valid_y = torch.zeros(feat_h, dtype=torch.uint8, device=device)
        valid_x[:valid_w] = 1
        valid_y[:valid_h] = 1
        valid_xx, valid_yy = self._meshgrid(valid_x, valid_y)
        valid = valid_xx & valid_yy
        valid = valid[:,
                      None].expand(valid.size(0),
                                   self.num_base_anchors).contiguous().view(-1)
        return valid


if __name__ == '__main__':
    self = AnchorGenerator(8,[8], [0.5])
    all_anchors = self.grid_anchors((5, 5), device='cpu')
    print(all_anchors)

另一种处理anchor的办法：

我们首先通过聚类生成自己的anchor,然后在每个层次的特征图上都指定使用我们聚类生成的anchor。这样，也不用像前面那种方法，根据anchor_stride,anchor_ratio等去计算每层生成的anchor的实际大小等。