def __init__(self,
embed_dims=256,
feedforward_channels=1024,
num_fcs=2,
act_cfg=dict(type='ReLU', inplace=True),
ffn_drop=0.,
dropout_layer=None,
add_identity=True,
init_cfg=None,
**kwargs):
super(FFN, self).__init__(init_cfg)
assert num_fcs >= 2, 'num_fcs should be no less ' \
f'than 2. got {num_fcs}.'
self.embed_dims = embed_dims
self.feedforward_channels = feedforward_channels
self.num_fcs = num_fcs
self.act_cfg = act_cfg
self.activate = build_activation_layer(act_cfg)
layers = []
in_channels = embed_dims
for _ in range(num_fcs - 1):
layers.append(
Sequential(Linear(in_channels, feedforward_channels),
self.activate, nn.Dropout(ffn_drop)))
in_channels = feedforward_channels
layers.append(Linear(feedforward_channels, embed_dims))
layers.append(nn.Dropout(ffn_drop))
self.layers = Sequential(*layers)
self.dropout_layer = build_dropout(
dropout_layer) if dropout_layer else torch.nn.Identity()
self.add_identity = add_identity
def __init__(self,
embed_dims,
feedforward_channels,
num_fcs=2,
act_cfg=dict(type='ReLU', inplace=True),
dropout=0.,
add_residual=True,
init_cfg=None):
super(FFN, self).__init__(init_cfg)
assert num_fcs >= 2, 'num_fcs should be no less ' \
f'than 2. got {num_fcs}.'
self.embed_dims = embed_dims
self.feedforward_channels = feedforward_channels
self.num_fcs = num_fcs
self.act_cfg = act_cfg
self.dropout = dropout
self.activate = build_activation_layer(act_cfg)
layers = []
in_channels = embed_dims
for _ in range(num_fcs - 1):
layers.append(
nn.Sequential(
Linear(in_channels, feedforward_channels), self.activate,
nn.Dropout(dropout)))
in_channels = feedforward_channels
layers.append(Linear(feedforward_channels, embed_dims))
self.layers = nn.Sequential(*layers)
self.dropout = nn.Dropout(dropout)
self.add_residual = add_residual
def _init_layers(self):
"""Initialize classification branch and regression branch of head."""
fc_cls = Linear(self.embed_dims, self.cls_out_channels)
reg_branch = []
for _ in range(self.num_reg_fcs):
reg_branch.append(Linear(self.embed_dims, self.embed_dims))
reg_branch.append(nn.ReLU())
reg_branch.append(Linear(self.embed_dims, 4))
reg_branch = nn.Sequential(*reg_branch)
def _get_clones(module, N):
return nn.ModuleList([copy.deepcopy(module) for i in range(N)])
# last reg_branch is used to generate proposal from
# encode feature map when as_two_stage is True.
num_pred = (self.transformer.decoder.num_layers + 1) if \
self.as_two_stage else self.transformer.decoder.num_layers
if self.with_box_refine:
self.cls_branches = _get_clones(fc_cls, num_pred)
self.reg_branches = _get_clones(reg_branch, num_pred)
else:
self.cls_branches = nn.ModuleList(
[fc_cls for _ in range(num_pred)])
self.reg_branches = nn.ModuleList(
[reg_branch for _ in range(num_pred)])
if not self.as_two_stage:
self.query_embedding = nn.Embedding(self.num_query,
self.embed_dims * 2)
def __init__(self,
num_convs=0,
num_fcs=2,
fc_out_channels=1024,
downsample_factor=2,
init_cfg=dict(type='Xavier',
override=[
dict(name='fcs'),
dict(type='Constant',
val=0.001,
name='fc_logits')
]),
*arg,
**kwarg):
super(CoarseMaskHead, self).__init__(*arg,
num_convs=num_convs,
upsample_cfg=dict(type=None),
init_cfg=None,
**kwarg)
self.init_cfg = init_cfg
self.num_fcs = num_fcs
assert self.num_fcs > 0
self.fc_out_channels = fc_out_channels
self.downsample_factor = downsample_factor
assert self.downsample_factor >= 1
# remove conv_logit
delattr(self, 'conv_logits')
if downsample_factor > 1:
downsample_in_channels = (self.conv_out_channels if
self.num_convs > 0 else self.in_channels)
self.downsample_conv = ConvModule(downsample_in_channels,
self.conv_out_channels,
kernel_size=downsample_factor,
stride=downsample_factor,
padding=0,
conv_cfg=self.conv_cfg,
norm_cfg=self.norm_cfg)
else:
self.downsample_conv = None
self.output_size = (self.roi_feat_size[0] // downsample_factor,
self.roi_feat_size[1] // downsample_factor)
self.output_area = self.output_size[0] * self.output_size[1]
last_layer_dim = self.conv_out_channels * self.output_area
self.fcs = ModuleList()
for i in range(num_fcs):
fc_in_channels = (last_layer_dim
if i == 0 else self.fc_out_channels)
self.fcs.append(Linear(fc_in_channels, self.fc_out_channels))
last_layer_dim = self.fc_out_channels
output_channels = self.num_classes * self.output_area
self.fc_logits = Linear(last_layer_dim, output_channels)
def __init__(self,
num_convs=4,
num_fcs=2,
roi_feat_size=14,
in_channels=256,
conv_out_channels=256,
fc_out_channels=1024,
num_classes=80,
loss_iou=dict(type='MSELoss', loss_weight=0.5),
init_cfg=[
dict(type='Kaiming', override=dict(name='convs')),
dict(type='Caffe2Xavier', override=dict(name='fcs')),
dict(
type='Normal',
std=0.01,
override=dict(name='fc_mask_iou'))
]):
super(MaskIoUHead, self).__init__(init_cfg)
self.in_channels = in_channels
self.conv_out_channels = conv_out_channels
self.fc_out_channels = fc_out_channels
self.num_classes = num_classes
self.fp16_enabled = False
self.convs = nn.ModuleList()
for i in range(num_convs):
if i == 0:
# concatenation of mask feature and mask prediction
in_channels = self.in_channels + 1
else:
in_channels = self.conv_out_channels
stride = 2 if i == num_convs - 1 else 1
self.convs.append(
Conv2d(
in_channels,
self.conv_out_channels,
3,
stride=stride,
padding=1))
roi_feat_size = _pair(roi_feat_size)
pooled_area = (roi_feat_size[0] // 2) * (roi_feat_size[1] // 2)
self.fcs = nn.ModuleList()
for i in range(num_fcs):
in_channels = (
self.conv_out_channels *
pooled_area if i == 0 else self.fc_out_channels)
self.fcs.append(Linear(in_channels, self.fc_out_channels))
self.fc_mask_iou = Linear(self.fc_out_channels, self.num_classes)
self.relu = nn.ReLU()
self.max_pool = MaxPool2d(2, 2)
self.loss_iou = build_loss(loss_iou)
def __init__(self,
in_features,
hidden_features=None,
out_features=None,
act_cfg=dict(type='GELU'),
drop=0.):
super(Mlp, self).__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
self.fc1 = Linear(in_features, hidden_features)
self.act = build_activation_layer(act_cfg)
self.fc2 = Linear(hidden_features, out_features)
self.drop = nn.Dropout(drop)
def _init_layers(self):
"""Initialize layers of the transformer head."""
self.input_proj = Conv2d(
self.in_channels, self.embed_dims, kernel_size=1)
self.fc_cls = Linear(self.embed_dims, self.cls_out_channels)
self.reg_ffn = FFN(
self.embed_dims,
self.embed_dims,
self.num_fcs,
self.act_cfg,
dropout=0.0,
add_residual=False)
self.fc_reg = Linear(self.embed_dims, 4)
self.query_embedding = nn.Embedding(self.num_query, self.embed_dims)
def __init__(self, out_channels, norm_cfg=dict(type='BN')):
# Protect mutable default arguments
norm_cfg = cp.deepcopy(norm_cfg)
super().__init__()
self.out_channels = out_channels
self.global_pooling = nn.AdaptiveAvgPool2d((1, 1))
self.middle_path = nn.Sequential(
Linear(self.out_channels, self.out_channels),
build_norm_layer(dict(type='BN1d'), out_channels)[1],
build_activation_layer(dict(type='ReLU')),
Linear(self.out_channels, self.out_channels),
build_norm_layer(dict(type='BN1d'), out_channels)[1],
build_activation_layer(dict(type='ReLU')),
build_activation_layer(dict(type='Sigmoid')))
self.bottom_path = nn.Sequential(
ConvModule(
self.out_channels,
self.out_channels,
kernel_size=1,
stride=1,
padding=0,
norm_cfg=norm_cfg,
inplace=False),
DepthwiseSeparableConvModule(
self.out_channels,
1,
kernel_size=9,
stride=1,
padding=4,
norm_cfg=norm_cfg,
inplace=False), build_activation_layer(dict(type='Sigmoid')))
self.conv_bn_relu_prm_1 = ConvModule(
self.out_channels,
self.out_channels,
kernel_size=3,
stride=1,
padding=1,
norm_cfg=norm_cfg,
inplace=False)
def __init__(self,
in_channels=768,
out_channels=[96, 192, 384, 768],
readout_type='ignore',
patch_size=16,
init_cfg=None):
super(ReassembleBlocks, self).__init__(init_cfg)
assert readout_type in ['ignore', 'add', 'project']
self.readout_type = readout_type
self.patch_size = patch_size
self.projects = nn.ModuleList([
ConvModule(
in_channels=in_channels,
out_channels=out_channel,
kernel_size=1,
act_cfg=None,
) for out_channel in out_channels
])
self.resize_layers = nn.ModuleList([
nn.ConvTranspose2d(
in_channels=out_channels[0],
out_channels=out_channels[0],
kernel_size=4,
stride=4,
padding=0),
nn.ConvTranspose2d(
in_channels=out_channels[1],
out_channels=out_channels[1],
kernel_size=2,
stride=2,
padding=0),
nn.Identity(),
nn.Conv2d(
in_channels=out_channels[3],
out_channels=out_channels[3],
kernel_size=3,
stride=2,
padding=1)
])
if self.readout_type == 'project':
self.readout_projects = nn.ModuleList()
for _ in range(len(self.projects)):
self.readout_projects.append(
nn.Sequential(
Linear(2 * in_channels, in_channels),
build_activation_layer(dict(type='GELU'))))
def __init__(
self,
embed_dim,
num_heads,
dropout=0.0,
bias=True,
add_bias_kv=False,
add_zero_attn=False,
kdim=None,
vdim=None,
app_relation=True,
):
super(MultiheadAttention, self).__init__()
self.embed_dim = embed_dim
self.kdim = kdim if kdim is not None else embed_dim
self.vdim = vdim if vdim is not None else embed_dim
self._qkv_same_embed_dim = (self.kdim == embed_dim) and (self.vdim
== embed_dim)
self.num_heads = num_heads
self.dropout = dropout
self.head_dim = embed_dim // num_heads
assert (self.head_dim * num_heads == self.embed_dim
), 'embed_dim must be divisible by num_heads'
self.in_proj_weight = Parameter(torch.empty(3 * embed_dim, embed_dim))
if self._qkv_same_embed_dim is False:
self.q_proj_weight = Parameter(torch.Tensor(embed_dim, embed_dim))
self.k_proj_weight = Parameter(torch.Tensor(embed_dim, self.kdim))
self.v_proj_weight = Parameter(torch.Tensor(embed_dim, self.vdim))
if bias:
self.in_proj_bias = Parameter(torch.empty(3 * embed_dim))
else:
self.register_parameter('in_proj_bias', None)
self.out_proj = Linear(embed_dim, embed_dim, bias=bias)
if add_bias_kv:
self.bias_k = Parameter(torch.empty(1, 1, embed_dim))
self.bias_v = Parameter(torch.empty(1, 1, embed_dim))
else:
self.bias_k = self.bias_v = None
self.add_zero_attn = add_zero_attn
self.app_relation = app_relation
self._reset_parameters()
def __init__(self,
dim,
num_heads=8,
qkv_bias=False,
qk_scale=None,
attn_drop=0.,
proj_drop=0.):
super(Attention, self).__init__()
self.num_heads = num_heads
head_dim = dim // num_heads
self.scale = qk_scale or head_dim**-0.5
self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
self.attn_drop = nn.Dropout(attn_drop)
self.proj = Linear(dim, dim)
self.proj_drop = nn.Dropout(proj_drop)
