def __init__(self,
input_shape: ShapeSpec,
*,
conv_dims: List[int],
fc_dims: List[int],
conv_norm="",
activation="ReLU"):
"""
NOTE: this interface is experimental.
Args:
input_shape (ShapeSpec): shape of the input feature.
conv_dims (list[int]): the output dimensions of the conv layers
fc_dims (list[int]): the output dimensions of the fc layers
conv_norm (str or callable): normalization for the conv layers.
See :func:`detectron2.layers.get_norm` for supported types.
"""
super().__init__()
assert len(conv_dims) + len(fc_dims) > 0
self._output_size = (input_shape.channels, input_shape.height,
input_shape.width)
self.conv_norm_relus = []
for k, conv_dim in enumerate(conv_dims):
conv = Conv2d(
self._output_size[0],
conv_dim,
kernel_size=3,
padding=1,
bias=not conv_norm,
norm=get_norm(conv_norm, conv_dim),
activation=get_activation(activation),
)
self.add_module("conv{}".format(k + 1), conv)
self.conv_norm_relus.append(conv)
self._output_size = (conv_dim, self._output_size[1],
self._output_size[2])
self.fcs = []
for k, fc_dim in enumerate(fc_dims):
if k == 0:
self.add_module("flatten", nn.Flatten())
fc = nn.Linear(int(np.prod(self._output_size)), fc_dim)
self.add_module("fc{}".format(k + 1), fc)
self.add_module("fc_activation{}".format(k + 1),
get_activation(activation))
self.fcs.append(fc)
self._output_size = fc_dim
for layer in self.conv_norm_relus:
weight_init.c2_msra_fill(layer)
for layer in self.fcs:
weight_init.c2_xavier_fill(layer)
def __init__(self,
in_channels,
out_channels,
*,
stride=1,
norm="BN",
activation="ReLU",
attention=""):
"""
Args:
in_channels (int): Number of input channels.
out_channels (int): Number of output channels.
stride (int): Stride for the first conv.
norm (str or callable): normalization for all conv layers.
See :func:`layers.get_norm` for supported format.
activation(str):activation to use.
"""
super().__init__(in_channels, out_channels, stride)
self.activation = get_activation(activation)
self.attention = CBAM(out_channels) if attention == "CBAM" else None
if in_channels != out_channels:
self.shortcut = Conv2d(
in_channels,
out_channels,
kernel_size=1,
stride=stride,
bias=False,
norm=get_norm(norm, out_channels),
)
else:
self.shortcut = None
self.conv1 = Conv2d(
in_channels,
out_channels,
kernel_size=3,
stride=stride,
padding=1,
bias=False,
norm=get_norm(norm, out_channels),
activation=self.activation,
)
self.conv2 = Conv2d(
out_channels,
out_channels,
kernel_size=3,
stride=1,
padding=1,
bias=False,
norm=get_norm(norm, out_channels),
)
for layer in [self.conv1, self.conv2, self.shortcut]:
if layer is not None: # shortcut can be None
weight_init.c2_msra_fill(layer)
def __init__(self,
in_channels,
out_channels,
in_feature="res5",
activation="ReLU"):
super().__init__()
self.num_levels = 2
self.in_feature = in_feature
self.p6 = nn.Conv2d(in_channels, out_channels, 3, 2, 1)
self.p7 = nn.Conv2d(out_channels, out_channels, 3, 2, 1)
for module in [self.p6, self.p7]:
weight_init.c2_xavier_fill(module)
self.activation = get_activation(activation)
def __init__(self, input_shape: ShapeSpec, *, num_classes, conv_dims, conv_norm="", **kwargs):
"""
NOTE: this interface is experimental.
Args:
input_shape (ShapeSpec): shape of the input feature
num_classes (int): the number of foreground classes (i.e. background is not
included). 1 if using class agnostic prediction.
conv_dims (list[int]): a list of N>0 integers representing the output dimensions
of N-1 conv layers and the last upsample layer.
conv_norm (str or callable): normalization for the conv layers.
See :func:`detectron2.layers.get_norm` for supported types.
"""
super().__init__(**kwargs)
assert len(conv_dims) >= 1, "conv_dims have to be non-empty!"
self.conv_norm_relus = []
activation = get_activation(activation)
cur_channels = input_shape.channels
for k, conv_dim in enumerate(conv_dims[:-1]):
conv = Conv2d(
cur_channels,
conv_dim,
kernel_size=3,
stride=1,
padding=1,
bias=not conv_norm,
norm=get_norm(conv_norm, conv_dim),
activation=activation,
)
self.add_module("mask_fcn{}".format(k + 1), conv)
self.conv_norm_relus.append(conv)
cur_channels = conv_dim
self.deconv = ConvTranspose2d(
cur_channels, conv_dims[-1], kernel_size=2, stride=2, padding=0
)
self.add_module("deconv_activation", activation)
cur_channels = conv_dims[-1]
self.predictor = Conv2d(cur_channels, num_classes, kernel_size=1, stride=1, padding=0)
for layer in self.conv_norm_relus + [self.deconv]:
weight_init.c2_msra_fill(layer)
# use normal distribution initialization for mask prediction layer
nn.init.normal_(self.predictor.weight, std=0.001)
if self.predictor.bias is not None:
nn.init.constant_(self.predictor.bias, 0)
def __init__(self, input_shape, *, num_keypoints, conv_dims, **kwargs):
"""
NOTE: this interface is experimental.
Args:
input_shape (ShapeSpec): shape of the input feature
conv_dims: an iterable of output channel counts for each conv in the head
e.g. (512, 512, 512) for three convs outputting 512 channels.
"""
super().__init__(num_keypoints=num_keypoints, **kwargs)
# default up_scale to 2.0 (this can be made an option)
up_scale = 2.0
in_channels = input_shape.channels
activation = get_activation(activation)
for idx, layer_channels in enumerate(conv_dims, 1):
module = Conv2d(in_channels,
layer_channels,
3,
stride=1,
padding=1)
self.add_module("conv_fcn{}".format(idx), module)
self.add_module("conv_fcn_activation{}".format(idx), activation)
in_channels = layer_channels
deconv_kernel = 4
self.score_lowres = ConvTranspose2d(in_channels,
num_keypoints,
deconv_kernel,
stride=2,
padding=deconv_kernel // 2 - 1)
self.up_scale = up_scale
for name, param in self.named_parameters():
if "bias" in name:
nn.init.constant_(param, 0)
elif "weight" in name:
# Caffe2 implementation uses MSRAFill, which in fact
# corresponds to kaiming_normal_ in PyTorch
nn.init.kaiming_normal_(param,
mode="fan_out",
nonlinearity="relu")
def __init__(self,
in_channels=3,
out_channels=64,
norm="BN",
activation="ReLU"):
"""
Args:
norm (str or callable): norm after the first conv layer.
See :func:`layers.get_norm` for supported format.
"""
super().__init__(in_channels, out_channels, 4)
self.in_channels = in_channels
self.activation = get_activation(activation)
self.conv1 = Conv2d(
in_channels,
out_channels,
kernel_size=7,
stride=2,
padding=3,
bias=False,
norm=get_norm(norm, out_channels),
activation=self.activation,
)
weight_init.c2_msra_fill(self.conv1)
def __init__(
self,
in_channels,
out_channels,
*,
bottleneck_channels,
stride=1,
num_groups=1,
norm="BN",
activation="ReLU",
attention="",
stride_in_1x1=False,
dilation=1,
):
"""
Args:
bottleneck_channels (int): number of output channels for the 3x3
"bottleneck" conv layers.
num_groups (int): number of groups for the 3x3 conv layer.
norm (str or callable): normalization for all conv layers.
See :func:`layers.get_norm` for supported format.
activation(str): activation to use.
stride_in_1x1 (bool): when stride>1, whether to put stride in the
first 1x1 convolution or the bottleneck 3x3 convolution.
dilation (int): the dilation rate of the 3x3 conv layer.
"""
super().__init__(in_channels, out_channels, stride)
self.activation = get_activation(activation)
self.attention = CBAM(out_channels) if attention == "CBAM" else None
if in_channels != out_channels:
self.shortcut = Conv2d(
in_channels,
out_channels,
kernel_size=1,
stride=stride,
bias=False,
norm=get_norm(norm, out_channels),
)
else:
self.shortcut = None
# The original MSRA ResNet models have stride in the first 1x1 conv
# The subsequent fb.torch.resnet and Caffe2 ResNe[X]t implementations have
# stride in the 3x3 conv
stride_1x1, stride_3x3 = (stride, 1) if stride_in_1x1 else (1, stride)
self.conv1 = Conv2d(
in_channels,
bottleneck_channels,
kernel_size=1,
stride=stride_1x1,
bias=False,
norm=get_norm(norm, bottleneck_channels),
activation=self.activation,
)
self.conv2 = Conv2d(
bottleneck_channels,
bottleneck_channels,
kernel_size=3,
stride=stride_3x3,
padding=1 * dilation,
bias=False,
groups=num_groups,
dilation=dilation,
norm=get_norm(norm, bottleneck_channels),
activation=self.activation,
)
self.conv3 = Conv2d(
bottleneck_channels,
out_channels,
kernel_size=1,
bias=False,
norm=get_norm(norm, out_channels),
)
for layer in [self.conv1, self.conv2, self.conv3, self.shortcut]:
if layer is not None: # shortcut can be None
weight_init.c2_msra_fill(layer)
def __init__(
self,
*,
input_shape: List[ShapeSpec],
num_classes,
num_anchors,
conv_dims: List[int],
norm="",
activation="ReLU",
prior_prob=0.01,
):
"""
NOTE: this interface is experimental.
Args:
input_shape (List[ShapeSpec]): input shape
num_classes (int): number of classes. Used to label background proposals.
num_anchors (int): number of generated anchors
conv_dims (List[int]): dimensions for each convolution layer
norm (str or callable):
Normalization for conv layers except for the two output layers.
See :func:`detectron2.layers.get_norm` for supported types.
prior_prob (float): Prior weight for computing bias
"""
super().__init__()
activation = get_activation(activation)
if norm == "BN" or norm == "SyncBN":
logger.warning(
"Shared norm does not work well for BN, SyncBN, expect poor results"
)
cls_subnet = []
bbox_subnet = []
for in_channels, out_channels in zip([input_shape[0].channels] +
conv_dims, conv_dims):
cls_subnet.append(
nn.Conv2d(in_channels,
out_channels,
kernel_size=3,
stride=1,
padding=1))
if norm:
cls_subnet.append(get_norm(norm, out_channels))
cls_subnet.append(activation)
bbox_subnet.append(
nn.Conv2d(in_channels,
out_channels,
kernel_size=3,
stride=1,
padding=1))
if norm:
bbox_subnet.append(get_norm(norm, out_channels))
bbox_subnet.append(activation)
self.cls_subnet = nn.Sequential(*cls_subnet)
self.bbox_subnet = nn.Sequential(*bbox_subnet)
self.cls_score = nn.Conv2d(conv_dims[-1],
num_anchors * num_classes,
kernel_size=3,
stride=1,
padding=1)
self.bbox_pred = nn.Conv2d(conv_dims[-1],
num_anchors * 4,
kernel_size=3,
stride=1,
padding=1)
# Initialization
for modules in [
self.cls_subnet, self.bbox_subnet, self.cls_score,
self.bbox_pred
]:
for layer in modules.modules():
if isinstance(layer, nn.Conv2d):
torch.nn.init.normal_(layer.weight, mean=0, std=0.01)
torch.nn.init.constant_(layer.bias, 0)
# Use prior in model initialization to improve stability
bias_value = -(math.log((1 - prior_prob) / prior_prob))
torch.nn.init.constant_(self.cls_score.bias, bias_value)
def __init__(self, cfg, input_shape: List[ShapeSpec]):
"""
Arguments:
in_channels (int): number of channels of the input feature
"""
super().__init__()
# TODO: Implement the sigmoid version first.
self.num_classes = cfg.MODEL.FCOS.NUM_CLASSES
self.fpn_strides = cfg.MODEL.FCOS.FPN_STRIDES
head_configs = {"cls": (cfg.MODEL.FCOS.NUM_CLS_CONVS,
cfg.MODEL.FCOS.USE_DEFORMABLE),
"bbox": (cfg.MODEL.FCOS.NUM_BOX_CONVS,
cfg.MODEL.FCOS.USE_DEFORMABLE),
"share": (cfg.MODEL.FCOS.NUM_SHARE_CONVS,
False)}
norm = None if cfg.MODEL.FCOS.NORM == "none" else cfg.MODEL.FCOS.NORM
activation = get_activation(cfg.MODEL.FCOS.ACTIVATION)
self.num_levels = len(input_shape)
in_channels = [s.channels for s in input_shape]
assert len(set(in_channels)) == 1, "Each level must have the same channel!"
in_channels = in_channels[0]
self.in_channels_to_top_module = in_channels
for head in head_configs:
tower = []
num_convs, use_deformable = head_configs[head]
for i in range(num_convs):
# TODO: comment deformable
# if use_deformable and i == num_convs - 1:
# conv_func = DFConv2d
# else:
conv_func = nn.Conv2d
tower.append(conv_func(
in_channels, in_channels,
kernel_size=3, stride=1,
padding=1, bias=True
))
if norm == "GN":
tower.append(nn.GroupNorm(32, in_channels))
# TODO: NaiveGN
# elif norm == "NaiveGN":
# tower.append(NaiveGroupNorm(32, in_channels))
elif norm == "BN":
tower.append(ModuleListDial([
nn.BatchNorm2d(in_channels) for _ in range(self.num_levels)
]))
elif norm == "SyncBN":
tower.append(ModuleListDial([
NaiveSyncBatchNorm(in_channels) for _ in range(self.num_levels)
]))
tower.append(activation)
self.add_module('{}_tower'.format(head),
nn.Sequential(*tower))
self.cls_logits = nn.Conv2d(
in_channels, self.num_classes,
kernel_size=3, stride=1,
padding=1
)
self.bbox_pred = nn.Conv2d(
in_channels, 4, kernel_size=3,
stride=1, padding=1
)
self.ctrness = nn.Conv2d(
in_channels, 1, kernel_size=3,
stride=1, padding=1
)
if cfg.MODEL.FCOS.USE_SCALE:
self.scales = nn.ModuleList([Scale(init_value=1.0) for _ in range(self.num_levels)])
else:
self.scales = None
for modules in [
self.cls_tower, self.bbox_tower,
self.share_tower, self.cls_logits,
self.bbox_pred, self.ctrness
]:
for l in modules.modules():
if isinstance(l, nn.Conv2d):
torch.nn.init.normal_(l.weight, std=0.01)
torch.nn.init.constant_(l.bias, 0)
# initialize the bias for focal loss
prior_prob = cfg.MODEL.FCOS.PRIOR_PROB
bias_value = -math.log((1 - prior_prob) / prior_prob)
torch.nn.init.constant_(self.cls_logits.bias, bias_value)