def __init__(self, encoder, n_classes, final_bias=0., chs=256, n_anchors=9, flatten=True, chip_size=(256,256), n_bands=3):
# chs - channels for top down layers in FPN
super().__init__()
self.n_classes,self.flatten = n_classes,flatten
self.chip_size = chip_size
# Fetch the sizes of various activation layers of the backbone
sfs_szs = model_sizes(encoder, size=self.chip_size)
hooks = hook_outputs(encoder)
self.encoder = encoder
self.c5top5 = conv2d(sfs_szs[-1][1], chs, ks=1, bias=True)
self.c5top6 = conv2d(sfs_szs[-1][1], chs, stride=2, bias=True)
self.p6top7 = nn.Sequential(nn.ReLU(), conv2d(chs, chs, stride=2, bias=True))
self.merges = nn.ModuleList([LateralUpsampleMerge(chs, szs[1], hook)
for szs,hook in zip(sfs_szs[-2:-4:-1], hooks[-2:-4:-1])])
self.smoothers = nn.ModuleList([conv2d(chs, chs, 3, bias=True) for _ in range(3)])
self.classifier = self._head_subnet(n_classes, n_anchors, final_bias, chs=chs)
self.box_regressor = self._head_subnet(4, n_anchors, 0., chs=chs)
# Create a dummy x to be passed through the model and fetch the sizes
x_dummy = torch.rand(n_bands,self.chip_size[0],self.chip_size[1]).unsqueeze(0)
p_states = self._create_p_states(x_dummy)
self.sizes = [[p.size(2), p.size(3)] for p in p_states]
def __init__(self, hooks, nc):
super(Hcolumns, self).__init__()
self.hooks = hooks
self.n = len(self.hooks)
self.factorization = None
if nc is not None:
self.factorization = nn.ModuleList()
for i in range(self.n):
self.factorization.append(nn.Sequential(
conv2d(nc[i], nc[-1], 3, padding=1, bias=True),
conv2d(nc[-1], nc[-1], 3, padding=1, bias=True)))
def __init__(self, body, out_dims, fpn_dim=256, emb_type='conv'):
super().__init__()
self.ifn, self.bu, self.td, self.fusion, ch = build_fpn(body, fpn_dim, bu_in_lateral=True,
out_dim=out_dims['object'])
self.embedding = resolve_embedding(emb_type, out_dims.keys(), ch[-1])
head = {key: nn.Sequential(nnlayers.conv_layer(fpn_dim, fpn_dim), conv2d(fpn_dim, fn, ks=1, bias=True))
for key, fn in out_dims.items()}
self.head = nn.ModuleDict(head)
def __init__(self, in_dim, out_dims):
super().__init__()
d = {
key: nn.Sequential(conv_layer(in_dim, in_dim),
conv2d(in_dim, fn, ks=1, bias=True))
for key, fn in out_dims.items()
}
self.heads = nn.ModuleDict(d)
def _head_subnet(self,
n_classes,
n_anchors,
final_bias=0.,
n_conv=4,
chs=256):
layers = [conv_layer(chs, chs, bias=True) for _ in range(n_conv)]
layers += [conv2d(chs, n_classes * n_anchors, bias=True)]
layers[-1].bias.data.zero_().add_(final_bias)
layers[-1].weight.data.fill_(0)
return nn.Sequential(*layers)
def __init__(self, instructor, fpn, fpn_dim=512):
classifier = nn.Sequential(nnlayers.conv_layer(fpn_dim, fpn_dim), conv2d(fpn_dim, 1, ks=1, bias=True))
super().__init__(instructor, fpn, classifier)
def __init__(self, ch, ch_lat, hook):
super().__init__()
self.hook = hook
self.conv_lat = conv2d(ch_lat, ch, ks=1, bias=True)
