def generate_anchors_global(feat_stride,
height,
width,
anchor_scales=(8, 16, 32),
anchor_ratios=(0.5, 1, 2)):
anchors = generate_anchors(base_size=feat_stride,
ratios=np.array(anchor_ratios),
scales=np.array(anchor_scales))
# Enumerate all shifts
shift_x = np.arange(0, width) * feat_stride
shift_y = np.arange(0, height) * feat_stride
shift_x, shift_y = np.meshgrid(shift_x, shift_y)
shifts = np.vstack((shift_x.ravel(), shift_y.ravel(), shift_x.ravel(),
shift_y.ravel())).transpose()
# Enumerate all shifted anchors:
#
# add A anchors (1, A, 4) to
# cell K shifts (K, 1, 4) to get
# shift anchors (K, A, 4)
# reshape to (A*K, 4) shifted anchors
A = anchors.shape[0]
K = shifts.shape[0]
anchors = anchors.reshape((1, A, 4)) + shifts.reshape((1, K, 4)).transpose(
(1, 0, 2))
anchors = anchors.reshape((K * A, 4)).astype(np.float32, copy=False)
return anchors
def __init__(self, feat_stride, scales, ratios):
super(ProposalLayer, self).__init__()
self._feat_stride = feat_stride
self._anchors = torch.from_numpy(
generate_anchors(scales=np.array(scales),
ratios=np.array(ratios))).float()
self._num_anchors = self._anchors.size(0)
def __init__(self, cfg, feat_stride, scales, ratios):
super(AnchorTargetLayer, self).__init__()
self.cfg = cfg
self._feat_stride = feat_stride
self._scales = scales
anchor_scales = scales
self._anchors = torch.from_numpy(
generate_anchors(base_size=cfg.FEAT_STRIDE[0],
scales=np.array(anchor_scales),
ratios=np.array(ratios))).float()
self._num_anchors = self._anchors.size(0)
# allow boxes to sit over the edge by a small amount
self._allowed_border = 0 # default is 0
def __init__(self, cfg):
super(AnchorTargetLayer, self).__init__()
self.cfg = cfg
self.scales = np.array(cfg.ANCHOR_SCALES)
self.ratios = np.array(cfg.ANCHOR_RATIOS)
self._feat_stride = cfg.FEAT_STRIDE[0]
base_anchors = generate_anchors(base_size=self._feat_stride,
scales=self.scales,
ratios=self.ratios)
self._anchors = torch.from_numpy(base_anchors).float()
self._num_anchors = self._anchors.size(0)
# allow boxes to sit over the edge by a small amount
self._allowed_border = 0 # default is 0
def generate_anchors_global(feat_strides,
heights,
widths,
anchor_scales=(8, 16, 32),
anchor_ratios=(0.5, 1, 2)):
assert len(feat_strides) == len(heights)
assert len(heights) == len(widths)
anchors_list = list()
for index in range(len(feat_strides)):
anchors = generate_anchors(base_size=feat_strides[index],
ratios=np.array(anchor_ratios),
scales=np.array([anchor_scales[index]]))
anchors_list.append(anchors)
num_anchors = np.asarray([anchors.shape[0] for anchors in anchors_list])
def global_anchors(height, width, feat_stride, anchors):
# Enumerate all shifts
shift_x = np.arange(0, width) * feat_stride
shift_y = np.arange(0, height) * feat_stride
shift_x, shift_y = np.meshgrid(shift_x, shift_y)
shifts = np.vstack((shift_x.ravel(), shift_y.ravel(), shift_x.ravel(),
shift_y.ravel())).transpose()
# Enumerate all shifted anchors:
#
# add A anchors (1, A, 4) to
# cell K shifts (K, 1, 4) to get
# shift anchors (K, A, 4)
# reshape to (A*K, 4) shifted anchors
A = anchors.shape[0]
K = shifts.shape[0]
anchors = anchors.reshape((1, A, 4)) + shifts.reshape(
(1, K, 4)).transpose((1, 0, 2))
anchors = anchors.reshape((K * A, 4)).astype(np.float32, copy=False)
return anchors
global_anchors_list = list()
for index in range(len(feat_strides)):
anchors = global_anchors(height=heights[index],
width=widths[index],
feat_stride=feat_strides[index],
anchors=anchors_list[index])
global_anchors_list.append(anchors)
anchors = np.concatenate(global_anchors_list, axis=0)
return anchors, num_anchors
def grid_anchors(self, featmap_size, stride=16, device='cuda'):
base_anchors = torch.from_numpy(generate_anchors(base_size=cfg.FEAT_STRIDE[0], scales=np.array(stride),
ratios=np.array(1.0))).float()
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