def __init__(self, feats=256, fc=(1,), score2prob=nn.Sigmoid()):
nn.Module.__init__(self)
self.meta = {
'mean': [0.485, 0.456, 0.406],
'std': [0.229, 0.224, 0.225],
'imageSize': [256, 128]
}
self.pair2bi = Pair2Bi()
self.pair2braid = Pair2Braid()
self.bi = osnet_x1_0(feats=feats)
self.bi.classifier = nn.Identity()
self.bi2braid = Bi2Braid()
self.braid = LinearMin2Block(feats, feats)
self.y = MinMaxY(feats, linear=True)
fc_blocks = []
channel_in = feats * 2
for i, sub_fc in enumerate(fc):
is_tail = (i + 1 == len(fc))
fc_blocks.append(FCBlock(channel_in, sub_fc, is_tail=is_tail))
channel_in = sub_fc
self.fc = nn.Sequential(*fc_blocks)
self.score2prob = score2prob
# initialize parameters
for m in [self.braid, self.fc]:
weights_init_kaiming(m)
self.correct_params()
def __init__(self, feats=512, num_classes=1000, **kwargs):
nn.Module.__init__(self)
self.meta = {
'mean': [0.485, 0.456, 0.406],
'std': [0.229, 0.224, 0.225],
'imageSize': [256, 128]
}
self.pair2bi = Pair2Bi()
self.bi = osnet_x1_0(feats=feats,
num_classes=num_classes)
self.dist = nn.CosineSimilarity(dim=1, eps=1e-6)
def __init__(self, feats=256, fc=(1,), score2prob=nn.Sigmoid()):
nn.Module.__init__(self)
self.meta = {
'mean': [0.485, 0.456, 0.406],
'std': [0.229, 0.224, 0.225],
'imageSize': [256, 128]
}
self.pair2bi = Pair2Bi()
self.pair2braid = Pair2Braid()
self.bi = osnet_x1_0(feats=feats)
self.bi.classifier = nn.Identity()
self.bi2braid = Bi2Braid()
self.braid = nn.Identity()
self.y = SumSquareY(feats, linear=True)
self.fc = nn.Identity()
# initialize parameters
for m in [self.braid, self.fc]:
weights_init_kaiming(m)
self.correct_params()