def __init__(self, in_channels, channels, kernel_size, stride=(1, 1), padding=(0, 0),
dilation=(1, 1), groups=1, bias=True,
radix=2, reduction_factor=4,
rectify=False, rectify_avg=False, norm_layer=None, num_splits=1,
dropblock_prob=0.0, **kwargs):
super(SplAtConv2d, self).__init__()
padding = _pair(padding)
self.rectify = rectify and (padding[0] > 0 or padding[1] > 0)
self.rectify_avg = rectify_avg
inter_channels = max(in_channels * radix // reduction_factor, 32)
self.radix = radix
self.cardinality = groups
self.channels = channels
self.dropblock_prob = dropblock_prob
if self.rectify:
from rfconv import RFConv2d
self.conv = RFConv2d(in_channels, channels * radix, kernel_size, stride, padding, dilation,
groups=groups * radix, bias=bias, average_mode=rectify_avg, **kwargs)
else:
self.conv = Conv2d(in_channels, channels * radix, kernel_size, stride, padding, dilation,
groups=groups * radix, bias=bias, **kwargs)
self.use_bn = norm_layer is not None
if self.use_bn:
self.bn0 = get_norm(norm_layer, channels * radix, num_splits)
self.relu = ReLU(inplace=True)
self.fc1 = Conv2d(channels, inter_channels, 1, groups=self.cardinality)
if self.use_bn:
self.bn1 = get_norm(norm_layer, inter_channels, num_splits)
self.fc2 = Conv2d(inter_channels, channels * radix, 1, groups=self.cardinality)
self.rsoftmax = rSoftMax(radix, groups)
def __init__(self,
inplanes,
planes,
bn_norm,
num_splits,
with_ibn=False,
with_se=False,
stride=1,
downsample=None,
reduction=16):
super(Bottleneck, self).__init__()
self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)
if with_ibn:
self.bn1 = IBN(planes, bn_norm, num_splits)
else:
self.bn1 = get_norm(bn_norm, planes, num_splits)
self.conv2 = nn.Conv2d(planes,
planes,
kernel_size=3,
stride=stride,
padding=1,
bias=False)
self.bn2 = get_norm(bn_norm, planes, num_splits)
self.conv3 = nn.Conv2d(planes,
planes * self.expansion,
kernel_size=1,
bias=False)
self.bn3 = get_norm(bn_norm, planes * self.expansion, num_splits)
self.relu = nn.ReLU(inplace=True)
if with_se:
self.se = SELayer(planes * self.expansion, reduction)
else:
self.se = nn.Identity()
self.downsample = downsample
self.stride = stride
def _make_layer(self,
block,
planes,
blocks,
stride=1,
bn_norm="BN",
num_splits=1,
with_ibn=False,
with_se=False):
downsample = None
if stride != 1 or self.inplanes != planes * block.expansion:
downsample = nn.Sequential(
nn.Conv2d(self.inplanes,
planes * block.expansion,
kernel_size=1,
stride=stride,
bias=False),
get_norm(bn_norm, planes * block.expansion, num_splits),
)
layers = []
if planes == 512:
with_ibn = False
layers.append(
block(self.inplanes, planes, bn_norm, num_splits, with_ibn,
with_se, stride, downsample))
self.inplanes = planes * block.expansion
for i in range(1, blocks):
layers.append(
block(self.inplanes, planes, bn_norm, num_splits, with_ibn,
with_se))
return nn.Sequential(*layers)
def __init__(self, last_stride, bn_norm, num_splits, with_ibn, with_se,
with_nl, block, layers, non_layers):
self.inplanes = 64
super().__init__()
self.conv1 = nn.Conv2d(3,
64,
kernel_size=7,
stride=2,
padding=3,
bias=False)
self.bn1 = get_norm(bn_norm, 64, num_splits)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(block, 64, layers[0], 1, bn_norm,
num_splits, with_ibn, with_se)
self.layer2 = self._make_layer(block, 128, layers[1], 2, bn_norm,
num_splits, with_ibn, with_se)
self.layer3 = self._make_layer(block, 256, layers[2], 2, bn_norm,
num_splits, with_ibn, with_se)
self.layer4 = self._make_layer(block,
512,
layers[3],
last_stride,
bn_norm,
num_splits,
with_se=with_se)
self.random_init()
if with_nl:
self._build_nonlocal(layers, non_layers, bn_norm, num_splits)
else:
self.NL_1_idx = self.NL_2_idx = self.NL_3_idx = self.NL_4_idx = []
def _build_pool_reduce(pool_layer, bn_norm, num_splits, input_dim=2048, reduce_dim=256):
pool_reduce = nn.Sequential(
pool_layer,
nn.Conv2d(input_dim, reduce_dim, 1, bias=False),
get_norm(bn_norm, reduce_dim, num_splits),
nn.ReLU(True),
)
pool_reduce.apply(weights_init_kaiming)
return pool_reduce
def __init__(self,
inplanes,
planes,
bn_norm,
num_splits,
with_ibn=False,
stride=1,
downsample=None,
radix=1,
cardinality=1,
bottleneck_width=64,
avd=False,
avd_first=False,
dilation=1,
is_first=False,
rectified_conv=False,
rectify_avg=False,
dropblock_prob=0.0,
last_gamma=False):
super(Bottleneck, self).__init__()
group_width = int(planes * (bottleneck_width / 64.)) * cardinality
self.conv1 = nn.Conv2d(inplanes,
group_width,
kernel_size=1,
bias=False)
if with_ibn:
self.bn1 = IBN(group_width, bn_norm, num_splits)
else:
self.bn1 = get_norm(bn_norm, group_width, num_splits)
self.dropblock_prob = dropblock_prob
self.radix = radix
self.avd = avd and (stride > 1 or is_first)
self.avd_first = avd_first
if self.avd:
self.avd_layer = nn.AvgPool2d(3, stride, padding=1)
stride = 1
if radix > 1:
self.conv2 = SplAtConv2d(group_width,
group_width,
kernel_size=3,
stride=stride,
padding=dilation,
dilation=dilation,
groups=cardinality,
bias=False,
radix=radix,
rectify=rectified_conv,
rectify_avg=rectify_avg,
norm_layer=bn_norm,
num_splits=num_splits,
dropblock_prob=dropblock_prob)
elif rectified_conv:
from rfconv import RFConv2d
self.conv2 = RFConv2d(group_width,
group_width,
kernel_size=3,
stride=stride,
padding=dilation,
dilation=dilation,
groups=cardinality,
bias=False,
average_mode=rectify_avg)
self.bn2 = get_norm(bn_norm, group_width, num_splits)
else:
self.conv2 = nn.Conv2d(group_width,
group_width,
kernel_size=3,
stride=stride,
padding=dilation,
dilation=dilation,
groups=cardinality,
bias=False)
self.bn2 = get_norm(bn_norm, group_width, num_splits)
self.conv3 = nn.Conv2d(group_width,
planes * 4,
kernel_size=1,
bias=False)
self.bn3 = get_norm(bn_norm, planes * 4, num_splits)
if last_gamma:
from torch.nn.init import zeros_
zeros_(self.bn3.weight)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
self.dilation = dilation
self.stride = stride
def _make_layer(self,
block,
planes,
blocks,
stride=1,
bn_norm="BN",
num_splits=1,
with_ibn=False,
dilation=1,
dropblock_prob=0.0,
is_first=True):
downsample = None
if stride != 1 or self.inplanes != planes * block.expansion:
down_layers = []
if self.avg_down:
if dilation == 1:
down_layers.append(
nn.AvgPool2d(kernel_size=stride,
stride=stride,
ceil_mode=True,
count_include_pad=False))
else:
down_layers.append(
nn.AvgPool2d(kernel_size=1,
stride=1,
ceil_mode=True,
count_include_pad=False))
down_layers.append(
nn.Conv2d(self.inplanes,
planes * block.expansion,
kernel_size=1,
stride=1,
bias=False))
else:
down_layers.append(
nn.Conv2d(self.inplanes,
planes * block.expansion,
kernel_size=1,
stride=stride,
bias=False))
down_layers.append(
get_norm(bn_norm, planes * block.expansion, num_splits))
downsample = nn.Sequential(*down_layers)
layers = []
if planes == 512:
with_ibn = False
if dilation == 1 or dilation == 2:
layers.append(
block(self.inplanes,
planes,
bn_norm,
num_splits,
with_ibn,
stride,
downsample=downsample,
radix=self.radix,
cardinality=self.cardinality,
bottleneck_width=self.bottleneck_width,
avd=self.avd,
avd_first=self.avd_first,
dilation=1,
is_first=is_first,
rectified_conv=self.rectified_conv,
rectify_avg=self.rectify_avg,
dropblock_prob=dropblock_prob,
last_gamma=self.last_gamma))
elif dilation == 4:
layers.append(
block(self.inplanes,
planes,
bn_norm,
num_splits,
with_ibn,
stride,
downsample=downsample,
radix=self.radix,
cardinality=self.cardinality,
bottleneck_width=self.bottleneck_width,
avd=self.avd,
avd_first=self.avd_first,
dilation=2,
is_first=is_first,
rectified_conv=self.rectified_conv,
rectify_avg=self.rectify_avg,
dropblock_prob=dropblock_prob,
last_gamma=self.last_gamma))
else:
raise RuntimeError("=> unknown dilation size: {}".format(dilation))
self.inplanes = planes * block.expansion
for i in range(1, blocks):
layers.append(
block(self.inplanes,
planes,
bn_norm,
num_splits,
with_ibn,
radix=self.radix,
cardinality=self.cardinality,
bottleneck_width=self.bottleneck_width,
avd=self.avd,
avd_first=self.avd_first,
dilation=dilation,
rectified_conv=self.rectified_conv,
rectify_avg=self.rectify_avg,
dropblock_prob=dropblock_prob,
last_gamma=self.last_gamma))
return nn.Sequential(*layers)
def __init__(self,
last_stride,
bn_norm,
num_splits,
with_ibn,
with_nl,
block,
layers,
non_layers,
radix=1,
groups=1,
bottleneck_width=64,
dilated=False,
dilation=1,
deep_stem=False,
stem_width=64,
avg_down=False,
rectified_conv=False,
rectify_avg=False,
avd=False,
avd_first=False,
final_drop=0.0,
dropblock_prob=0,
last_gamma=False):
self.cardinality = groups
self.bottleneck_width = bottleneck_width
# ResNet-D params
self.inplanes = stem_width * 2 if deep_stem else 64
self.avg_down = avg_down
self.last_gamma = last_gamma
# ResNeSt params
self.radix = radix
self.avd = avd
self.avd_first = avd_first
super().__init__()
self.rectified_conv = rectified_conv
self.rectify_avg = rectify_avg
if rectified_conv:
from rfconv import RFConv2d
conv_layer = RFConv2d
else:
conv_layer = nn.Conv2d
conv_kwargs = {'average_mode': rectify_avg} if rectified_conv else {}
if deep_stem:
self.conv1 = nn.Sequential(
conv_layer(3,
stem_width,
kernel_size=3,
stride=2,
padding=1,
bias=False,
**conv_kwargs),
get_norm(bn_norm, stem_width, num_splits),
nn.ReLU(inplace=True),
conv_layer(stem_width,
stem_width,
kernel_size=3,
stride=1,
padding=1,
bias=False,
**conv_kwargs),
get_norm(bn_norm, stem_width, num_splits),
nn.ReLU(inplace=True),
conv_layer(stem_width,
stem_width * 2,
kernel_size=3,
stride=1,
padding=1,
bias=False,
**conv_kwargs),
)
else:
self.conv1 = conv_layer(3,
64,
kernel_size=7,
stride=2,
padding=3,
bias=False,
**conv_kwargs)
self.bn1 = get_norm(bn_norm, self.inplanes, num_splits)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(block,
64,
layers[0],
1,
bn_norm,
num_splits,
with_ibn=with_ibn,
is_first=False)
self.layer2 = self._make_layer(block,
128,
layers[1],
2,
bn_norm,
num_splits,
with_ibn=with_ibn)
if dilated or dilation == 4:
self.layer3 = self._make_layer(block,
256,
layers[2],
1,
bn_norm,
num_splits,
with_ibn=with_ibn,
dilation=2,
dropblock_prob=dropblock_prob)
self.layer4 = self._make_layer(block,
512,
layers[3],
1,
bn_norm,
num_splits,
with_ibn=with_ibn,
dilation=4,
dropblock_prob=dropblock_prob)
elif dilation == 2:
self.layer3 = self._make_layer(block,
256,
layers[2],
2,
bn_norm,
num_splits,
with_ibn=with_ibn,
dilation=1,
dropblock_prob=dropblock_prob)
self.layer4 = self._make_layer(block,
512,
layers[3],
1,
bn_norm,
num_splits,
with_ibn=with_ibn,
dilation=2,
dropblock_prob=dropblock_prob)
else:
self.layer3 = self._make_layer(block,
256,
layers[2],
2,
bn_norm,
num_splits,
with_ibn=with_ibn,
dropblock_prob=dropblock_prob)
self.layer4 = self._make_layer(block,
512,
layers[3],
last_stride,
bn_norm,
num_splits,
with_ibn=with_ibn,
dropblock_prob=dropblock_prob)
for m in self.modules():
if isinstance(m, nn.Conv2d):
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
m.weight.data.normal_(0, math.sqrt(2. / n))
elif isinstance(m, nn.BatchNorm2d):
m.weight.data.fill_(1)
m.bias.data.zero_()
if with_nl:
self._build_nonlocal(layers, non_layers, bn_norm, num_splits)
else:
self.NL_1_idx = self.NL_2_idx = self.NL_3_idx = self.NL_4_idx = []
def __init__(self, cfg):
super().__init__()
self._cfg = cfg
assert len(cfg.MODEL.PIXEL_MEAN) == len(cfg.MODEL.PIXEL_STD)
self.register_buffer("pixel_mean", torch.Tensor(cfg.MODEL.PIXEL_MEAN).view(1, -1, 1, 1))
self.register_buffer("pixel_std", torch.Tensor(cfg.MODEL.PIXEL_STD).view(1, -1, 1, 1))
# backbone
bn_norm = cfg.MODEL.BACKBONE.NORM
num_splits = cfg.MODEL.BACKBONE.NORM_SPLIT
with_se = cfg.MODEL.BACKBONE.WITH_SE
backbone = build_backbone(cfg)
self.backbone = nn.Sequential(
backbone.conv1,
backbone.bn1,
backbone.relu,
backbone.maxpool,
backbone.layer1,
backbone.layer2,
backbone.layer3[0]
)
res_conv4 = nn.Sequential(*backbone.layer3[1:])
res_g_conv5 = backbone.layer4
res_p_conv5 = nn.Sequential(
Bottleneck(1024, 512, bn_norm, num_splits, False, with_se, downsample=nn.Sequential(
nn.Conv2d(1024, 2048, 1, bias=False), get_norm(bn_norm, 2048, num_splits))),
Bottleneck(2048, 512, bn_norm, num_splits, False, with_se),
Bottleneck(2048, 512, bn_norm, num_splits, False, with_se))
res_p_conv5.load_state_dict(backbone.layer4.state_dict())
pool_type = cfg.MODEL.HEADS.POOL_LAYER
if pool_type == 'avgpool': pool_layer = FastGlobalAvgPool2d()
elif pool_type == 'maxpool': pool_layer = nn.AdaptiveMaxPool2d(1)
elif pool_type == 'gempool': pool_layer = GeneralizedMeanPoolingP()
elif pool_type == "avgmaxpool": pool_layer = AdaptiveAvgMaxPool2d()
elif pool_type == "identity": pool_layer = nn.Identity()
else:
raise KeyError(f"{pool_type} is invalid, please choose from "
f"'avgpool', 'maxpool', 'gempool', 'avgmaxpool' and 'identity'.")
# head
in_feat = cfg.MODEL.HEADS.IN_FEAT
num_classes = cfg.MODEL.HEADS.NUM_CLASSES
# branch1
self.b1 = nn.Sequential(
copy.deepcopy(res_conv4), copy.deepcopy(res_g_conv5)
)
self.b1_pool = self._build_pool_reduce(pool_layer, bn_norm, num_splits, reduce_dim=in_feat)
self.b1_head = build_reid_heads(cfg, in_feat, num_classes, nn.Identity())
# branch2
self.b2 = nn.Sequential(
copy.deepcopy(res_conv4), copy.deepcopy(res_p_conv5)
)
self.b2_pool = self._build_pool_reduce(pool_layer, bn_norm, num_splits, reduce_dim=in_feat)
self.b2_head = build_reid_heads(cfg, in_feat, num_classes, nn.Identity())
self.b21_pool = self._build_pool_reduce(pool_layer, bn_norm, num_splits, reduce_dim=in_feat)
self.b21_head = build_reid_heads(cfg, in_feat, num_classes, nn.Identity())
self.b22_pool = self._build_pool_reduce(pool_layer, bn_norm, num_splits, reduce_dim=in_feat)
self.b22_head = build_reid_heads(cfg, in_feat, num_classes, nn.Identity())
# branch3
self.b3 = nn.Sequential(
copy.deepcopy(res_conv4), copy.deepcopy(res_p_conv5)
)
self.b3_pool = self._build_pool_reduce(pool_layer, bn_norm, num_splits, reduce_dim=in_feat)
self.b3_head = build_reid_heads(cfg, in_feat, num_classes, nn.Identity())
self.b31_pool = self._build_pool_reduce(pool_layer, bn_norm, num_splits, reduce_dim=in_feat)
self.b31_head = build_reid_heads(cfg, in_feat, num_classes, nn.Identity())
self.b32_pool = self._build_pool_reduce(pool_layer, bn_norm, num_splits, reduce_dim=in_feat)
self.b32_head = build_reid_heads(cfg, in_feat, num_classes, nn.Identity())
self.b33_pool = self._build_pool_reduce(pool_layer, bn_norm, num_splits, reduce_dim=in_feat)
self.b33_head = build_reid_heads(cfg, in_feat, num_classes, nn.Identity())
