def __init__(self, stride, scale, ratios):
super(proposal_layer, self).__init__()
self.anchors = torch.from_numpy(
generate_anchors(scales=numpy.array(scale),
ratios=numpy.array(numpy.array(ratios))))
self.anchor_num = self.anchors.shape[0]
self.stride = stride
def __init__(self, feat_stride, scales, ratios):
super(_SiamProposalLayer, 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, feat_stride, scales, ratios, classes, n_classes):
super(_RPNAggregationLayer, self).__init__()
self._feat_stride = feat_stride
self.classes = classes
self.n_classes = n_classes
anchor_scales = scales
self._anchors = torch.from_numpy(
generate_anchors(scales=np.array(anchor_scales),
ratios=np.array(ratios))).float()
self._num_anchors = self._anchors.size(0)
def __init__(self, stride, scale, ratios, allowed_border, name):
super(anchor_target_layer, self).__init__()
self.stride = stride
self.scale = scale
self.name = name
self.anchors = torch.from_numpy(
generate_anchors(scales=numpy.array(scale),
ratios=numpy.array(ratios))).float()
self.anchor_num = self.anchors.shape[0]
self.allowed_border = allowed_border
def __init__(self, feat_stride, scales, ratios):
super(_SiamAnchorTargetLayer, self).__init__()
self._feat_stride = feat_stride
self._scales = scales
anchor_scales = scales
self._anchors = torch.from_numpy(generate_anchors(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, feat_stride, scales, ratios, use_gpu=False):
super(_ProposalLayer, self).__init__()
self._feat_stride = feat_stride # [16]
self._anchors = torch.from_numpy(generate_anchors(scales=np.array(scales),
ratios=np.array(ratios))).float()
self._num_anchors = self._anchors.size(0)
self.use_gpu = use_gpu
if self.use_gpu:
self.nms_gpu = True
else:
self.nms_gpu = False
def __init__(self,
classes,
class_agnostic,
feat_name='efficientnet-b2',
feat_list=('conv3', 'conv4', 'conv5', 'conv6', 'conv7'),
pretrained=True,
D_bifpn=3,
W_bifpn=88):
super(EfficientDet, self).__init__(classes, class_agnostic, feat_name,
feat_list, pretrained)
self.ED_BIFPN = BIFPN(in_channels=self.FeatExt.get_list_features(),
out_channels=W_bifpn,
stack=D_bifpn,
num_outs=5)
self.ED_retinahead = retinaHeader(num_classes=self.n_classes,
in_channels=W_bifpn,
class_agnostic=self.class_agnostic)
self._anchor_scales = cfg.RCNN_COMMON.ANCHOR_SCALES
self._anchor_ratios = cfg.RCNN_COMMON.ANCHOR_RATIOS
self._feat_stride = cfg.RCNN_COMMON.FEAT_STRIDE
self._anchors = []
self._num_anchors = []
for i in self._feat_stride:
if isinstance(self._anchor_scales[0], list):
scales = self._anchor_scales[i]
else:
scales = self._anchor_scales
if isinstance(self._anchor_ratios[0], list):
ratios = self._anchor_ratios[i]
else:
ratios = self._anchor_ratios
anchor = torch.from_numpy(
generate_anchors(base_size=i,
scales=np.array(scales),
ratios=np.array(ratios))).float()
self._anchors.append(anchor)
self._num_anchors.append(anchor.size(0))
for m in self.modules():
if isinstance(m, nn.Conv2d):
weight_kaiming_init(m)
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
# self.freeze_bn()
self.iter_counter = 0