def __init__(self,
num_refinement_stages=1,
num_channels=128,
num_heatmaps=17,
mode='bilinear'):
super().__init__()
self.model = nn.Sequential(
conv(3, 32, stride=2, bias=False),
conv_dw(32, 64),
conv_dw(64, 128, stride=2),
conv_dw(128, 128),
conv_dw(128, 256, stride=2),
conv_dw(256, 256),
conv_dw(256, 512), # conv4_2
conv_dw(512, 512, dilation=2, padding=2),
conv_dw(512, 512),
conv_dw(512, 512),
conv_dw(512, 512),
conv_dw(512, 512), # conv5_5
)
self.cpm = nn.Sequential(
conv(512, 256),
conv(256, 128),
)
self.initial_stage = InitialStage(num_channels, num_heatmaps)
self.refinement_stages = nn.ModuleList()
for idx in range(num_refinement_stages):
self.refinement_stages.append(
RefinementStage(num_channels + num_heatmaps, num_channels,
num_heatmaps, mode))
def __init__(self, in_channels, out_channels, num_heatmaps, num_pafs):
super().__init__()
self.trunk = nn.Sequential(
RefinementStageBlock(in_channels, out_channels),
RefinementStageBlock(out_channels, out_channels),
RefinementStageBlock(out_channels, out_channels),
RefinementStageBlock(out_channels, out_channels),
RefinementStageBlock(out_channels, out_channels))
self.heatmaps = nn.Sequential(
conv(out_channels,
out_channels,
kernel_size=1,
padding=0,
bn=False),
conv(out_channels,
num_heatmaps,
kernel_size=1,
padding=0,
bn=False,
relu=False))
self.pafs = nn.Sequential(
conv(out_channels,
out_channels,
kernel_size=1,
padding=0,
bn=False),
conv(out_channels,
num_pafs,
kernel_size=1,
padding=0,
bn=False,
relu=False))
def __init__(self, in_channels, out_channels):
super().__init__()
self.initial = conv(in_channels, out_channels, kernel_size=1, padding=0, bn=False)
self.trunk = nn.Sequential(
conv(out_channels, out_channels),
conv(out_channels, out_channels, dilation=2, padding=2)
)
def __init__(self, in_channels, out_channels, num_heatmaps, num_pafs):
super().__init__()
self.trunk = nn.Sequential( #----------------------------------------------------In OpenPose five 7 * 7,
RefinementStageBlock(
in_channels, out_channels
), #---------------------------here replaced with five time [3*3 , 3*3 , 1*1]
RefinementStageBlock(out_channels, out_channels),
RefinementStageBlock(out_channels, out_channels),
RefinementStageBlock(out_channels, out_channels),
RefinementStageBlock(out_channels, out_channels))
self.heatmaps = nn.Sequential(
conv(out_channels,
out_channels,
kernel_size=1,
padding=0,
bn=False),
conv(out_channels,
num_heatmaps,
kernel_size=1,
padding=0,
bn=False,
relu=False))
self.pafs = nn.Sequential(
conv(out_channels,
out_channels,
kernel_size=1,
padding=0,
bn=False),
conv(out_channels,
num_pafs,
kernel_size=1,
padding=0,
bn=False,
relu=False))
def __init__(self,
in_channels,
out_channels,
num_heatmaps,
mode='bilinear'):
super().__init__()
self.trunk = nn.Sequential(
UShapedContextBlock(in_channels, mode),
RefinementStageBlock(in_channels, out_channels),
RefinementStageBlock(out_channels, out_channels),
RefinementStageBlock(out_channels, out_channels),
)
self.heatmaps = nn.Sequential(
conv(out_channels,
out_channels,
kernel_size=1,
padding=0,
bn=False),
conv(out_channels,
num_heatmaps,
kernel_size=1,
padding=0,
bn=False,
relu=False))
def __init__(self, in_channels, out_channels):
super().__init__()
self.align = conv(in_channels, out_channels, kernel_size=1, padding=0, bn=False)
self.trunk = nn.Sequential(
conv_dw_no_bn(out_channels, out_channels),
conv_dw_no_bn(out_channels, out_channels),
conv_dw_no_bn(out_channels, out_channels)
)
self.conv = conv(out_channels, out_channels, bn=False)
def __init__(self, in_channels, out_channels):#--------512,128
super().__init__()
self.align = conv(in_channels, out_channels, kernel_size=1, padding=0, bn=False)
self.trunk = nn.Sequential(
conv_dw_no_bn(out_channels, out_channels), #kernel_size=3, padding=1, stride=1, dilation=1
conv_dw_no_bn(out_channels, out_channels), #kernel_size=3, padding=1, stride=1, dilation=1
conv_dw_no_bn(out_channels, out_channels) #kernel_size=3, padding=1, stride=1, dilation=1
)
self.conv = conv(out_channels, out_channels, bn=False)
def __init__(self, in_channels, out_channels, ratio, should_align=False):
super().__init__()
self.should_align = should_align
self.bottleneck = nn.Sequential(
conv(in_channels, in_channels // ratio, kernel_size=1, padding=0),
conv(in_channels // ratio, in_channels // ratio),
conv(in_channels // ratio, out_channels, kernel_size=1, padding=0)
)
if self.should_align:
self.align = conv(in_channels, out_channels, kernel_size=1, padding=0)
def __init__(self, in_channels, mid_channels, out_channels):
super().__init__()
self.trunk = nn.Sequential(
RefinementStageBlock(in_channels, mid_channels),
RefinementStageBlock(mid_channels, mid_channels)
)
self.feature_maps = nn.Sequential(
conv(mid_channels, mid_channels, kernel_size=1, padding=0, bn=False),
conv(mid_channels, out_channels, kernel_size=1, padding=0, bn=False, relu=False)
)
def __init__(self, in_channels, cfg):
super().__init__()
self.align = conv(in_channels,
int(cfg[0]),
kernel_size=1,
padding=0,
bn=False)
# self.select = Selection(cfg[0])
self.trunk = nn.Sequential(conv_dw_no_bn(int(cfg[0]), int(cfg[2])),
conv_dw_no_bn(int(cfg[2]), int(cfg[4])),
conv_dw_no_bn(int(cfg[4]), int(cfg[6])))
self.conv = conv(int(cfg[6]), int(cfg[7]), bn=False)
def __init__(self, num_channels, num_heatmaps):
super().__init__()
self.trunk = nn.Sequential(conv(num_channels, num_channels),
conv(num_channels, num_channels),
conv(num_channels, num_channels))
self.heatmaps = nn.Sequential(
conv(num_channels, 512, kernel_size=1, padding=0, bn=False),
conv(512,
num_heatmaps,
kernel_size=1,
padding=0,
bn=False,
relu=False))
def __init__(self,
num_refinement_stages=1,
num_channels=128,
num_heatmaps=1 + num_keys,
num_pafs=2 * len(BODY_PARTS_PAF_IDS)):
super().__init__()
self.model = nn.Sequential(
conv(3, 32, stride=2, bias=False),
conv_dw(32, 64),
conv_dw(64, 128, stride=2),
conv_dw(128, 128),
conv_dw(128, 256, stride=2),
conv_dw(256, 256),
conv_dw(256, 512), # conv4_2
conv_dw(512, 512, dilation=2, padding=2),
conv_dw(512, 512),
conv_dw(512, 512),
conv_dw(512, 512),
conv_dw(512, 512) # conv5_5
)
self.cpm = Cpm(512, num_channels)
self.initial_stage = InitialStage(num_channels, num_heatmaps, num_pafs)
self.refinement_stages = nn.ModuleList()
for idx in range(num_refinement_stages):
self.refinement_stages.append(
RefinementStage(num_channels + num_heatmaps + num_pafs,
num_channels, num_heatmaps, num_pafs))
def __init__(self, num_refinement_stages=1, num_channels=128, num_heatmaps=19, num_pafs=38):#--------------why 38
super().__init__()
self.model = nn.Sequential(#---------------------------------------------------MobileNet-----------------------------------first
conv( 3, 32, stride=2, bias=False),
conv_dw( 32, 64),#---------------------------------------------dw means depthwise convolution
conv_dw( 64, 128, stride=2),
conv_dw(128, 128),
conv_dw(128, 256, stride=2),
conv_dw(256, 256),
conv_dw(256, 512), # Stride of this conv4_2 removed from MobileNetv1 to preserve the receptive field. conv4_2 means kernel 4 * 4, padding 2
conv_dw(512, 512, dilation=2, padding=2),
conv_dw(512, 512),
conv_dw(512, 512),
conv_dw(512, 512),
conv_dw(512, 512) # conv5_5 The last layer used from MobileNet is this one.
)
self.cpm = Cpm(512, num_channels)#-------------------------------------------------What is this CPM?--------------second
self.initial_stage = InitialStage(num_channels, num_heatmaps, num_pafs)#--------------------------------------------third
self.refinement_stages = nn.ModuleList()#------------------------------------------------------------------------fourth
for idx in range(num_refinement_stages):
self.refinement_stages.append(RefinementStage(num_channels + num_heatmaps + num_pafs, num_channels,
num_heatmaps, num_pafs))
def __init__(self, num_refinement_stages=1, num_channels=128, num_heatmaps=19, num_pafs=38,
is_convertible_by_mo=False):
super().__init__()
self.is_convertible_by_mo = is_convertible_by_mo
self.model = nn.Sequential(
conv( 3, 32, stride=2, bias=False),
conv_dw( 32, 64),
conv_dw( 64, 128, stride=2),
conv_dw(128, 128),
conv_dw(128, 256, stride=2),
conv_dw(256, 256),
conv_dw(256, 512), # conv4_2
conv_dw(512, 512, dilation=2, padding=2),
conv_dw(512, 512),
conv_dw(512, 512),
conv_dw(512, 512),
conv_dw(512, 512) # conv5_5
)
self.cpm = Cpm(512, num_channels)
self.initial_stage = InitialStage(num_channels, num_heatmaps, num_pafs)
self.refinement_stages = nn.ModuleList()
for idx in range(num_refinement_stages):
self.refinement_stages.append(RefinementStage(num_channels + num_heatmaps + num_pafs, num_channels,
num_heatmaps, num_pafs))
self.Pose3D = Pose3D(128, num_2d_heatmaps=57)
if self.is_convertible_by_mo:
self.fake_conv_heatmaps = nn.Conv2d(num_heatmaps, num_heatmaps, kernel_size=1, bias=False)
self.fake_conv_heatmaps.weight = nn.Parameter(torch.zeros(num_heatmaps, num_heatmaps, 1, 1))
self.fake_conv_pafs = nn.Conv2d(num_pafs, num_pafs, kernel_size=1, bias=False)
self.fake_conv_pafs.weight = nn.Parameter(torch.zeros(num_pafs, num_pafs, 1, 1))
def __init__(self, in_channels):
super().__init__()
self.encoder1 = nn.Sequential(
conv(in_channels, in_channels * 2, stride=2),
conv(in_channels * 2, in_channels * 2),
)
self.encoder2 = nn.Sequential(
conv(in_channels * 2, in_channels * 2, stride=2),
conv(in_channels * 2, in_channels * 2),
)
self.decoder2 = nn.Sequential(
conv(in_channels * 2 + in_channels * 2, in_channels * 2),
conv(in_channels * 2, in_channels * 2),
)
self.decoder1 = nn.Sequential(conv(in_channels * 3, in_channels * 2),
conv(in_channels * 2, in_channels))
def __init__(self, in_channels, mode='bilinear'):
super().__init__()
self.mode_interpolation = mode
self.encoder1 = nn.Sequential(
conv(in_channels, in_channels * 2, stride=2),
conv(in_channels * 2, in_channels * 2),
)
self.encoder2 = nn.Sequential(
conv(in_channels * 2, in_channels * 2, stride=2),
conv(in_channels * 2, in_channels * 2),
)
self.decoder2 = nn.Sequential(
conv(in_channels * 2 + in_channels * 2, in_channels * 2),
conv(in_channels * 2, in_channels * 2),
)
self.decoder1 = nn.Sequential(conv(in_channels * 3, in_channels * 2),
conv(in_channels * 2, in_channels))
def __init__(self,
num_refinement_stages=1,
num_channels=128,
num_heatmaps=19,
num_pafs=38):
super().__init__()
self.ghost0 = conv(3, 32, stride=2, bias=False)
self.ghost1 = conv(3, 32, stride=2, bias=False)
self.ghost2 = conv(3, 32, stride=2, bias=False)
self.phase0 = conv_dw(32, 64)
# self.phase1 = nn.Sequential(
# conv_dw(32, 64),
# )
self.model = nn.Sequential(
conv_dw(64, 128, stride=2),
conv_dw(128, 128),
conv_dw(128, 256, stride=2),
conv_dw(256, 256),
conv_dw(256, 512), # conv4_2
conv_dw(512, 512, dilation=2, padding=2),
conv_dw(512, 512),
conv_dw(512, 512),
conv_dw(512, 512),
conv_dw(512, 512) # conv5_5
)
self.cpm = Cpm(512, num_channels)
self.initial_stage = InitialStage(num_channels, num_heatmaps, num_pafs)
self.refinement_stages = nn.ModuleList()
for idx in range(num_refinement_stages):
self.refinement_stages.append(
RefinementStage(num_channels + num_heatmaps + num_pafs,
num_channels, num_heatmaps, num_pafs))
def __init__(self, num_refinement_stages=1, num_channels=128, num_heatmaps=19, num_pafs=38):
super().__init__()
self.model = nn.Sequential(
conv( 3, 32, stride=2, bias=False),
conv_dw( 32, 64),
conv_dw( 64, 128, stride=2),
conv_dw(128, 128),
conv_dw(128, 256, stride=2),
conv_dw(256, 256),
conv_dw(256, 512), # conv4_2
conv_dw(512, 512, dilation=2, padding=2),
conv_dw(512, 512),
conv_dw(512, 512),
conv_dw(512, 512),
conv_dw(512, 512) # conv5_5
)
self.cpm = Cpm(512, num_channels)
self.initial_stage = InitialStage(num_channels, num_heatmaps, num_pafs)
def __init__(self, num_channels, num_heatmaps, num_pafs):
super().__init__()
self.trunk = nn.Sequential(
conv(num_channels, num_channels, bn=False),
conv(num_channels, num_channels, bn=False),
#conv(num_channels, num_channels, bn=False)#-----------------------------------remove one convolution layer
)
self.heatmaps = nn.Sequential(
conv(num_channels, 512, kernel_size=1, padding=0, bn=False),
conv(512, num_heatmaps, kernel_size=1, padding=0, bn=False, relu=False)
)
self.pafs = nn.Sequential(
conv(num_channels, 512, kernel_size=1, padding=0, bn=False),
conv(512, num_pafs, kernel_size=1, padding=0, bn=False, relu=False)
)
def __init__(self,
cfg,
num_refinement_stages=1,
num_channels=128,
num_heatmaps=15,
num_pafs=26):
super().__init__()
layers = []
layers += [conv(3, int(cfg[0][0]), stride=2, bias=False)]
for i in range(int((len(cfg[0]) - 1) / 2)):
if (i == 1 or i == 3):
layers += [
conv_dw(int(cfg[0][i * 2]),
int(cfg[0][i * 2 + 2]),
stride=2)
]
elif (i == 6):
layers += [
conv_dw(int(cfg[0][i * 2]),
int(cfg[0][i * 2 + 2]),
dilation=2,
padding=2)
]
else:
layers += [conv_dw(int(cfg[0][i * 2]), int(cfg[0][i * 2 + 2]))]
self.model = nn.Sequential(*layers)
self.cpm = Cpm_new(int(cfg[0][-1]), cfg[1])
self.initial_stage = InitialStage_new(int(cfg[1][-1]), num_channels,
num_heatmaps, num_pafs)
self.refinement_stages = nn.ModuleList()
for idx in range(num_refinement_stages):
self.refinement_stages.append(
RefinementStage(
int(cfg[1][-1]) + num_heatmaps + num_pafs, num_channels,
num_heatmaps, num_pafs))
def __init__(self, num_channels, num_heatmaps, num_pafs):
super().__init__()
self.trunk = nn.Sequential( #----------------------------------------Original OPENPOSE contained two copies of these conv() layers, for PAF and HMs
conv(
num_channels, num_channels, bn=False
), #--------------------Here just 1 in the original lightweight openpose
conv(num_channels, num_channels, bn=False),
conv(num_channels, num_channels, bn=False))
#--------------------------------------------------------------------Then, give the op to both paf and HM branches
self.heatmaps = nn.Sequential(
conv(num_channels, 512, kernel_size=1, padding=0, bn=False),
conv(512,
num_heatmaps,
kernel_size=1,
padding=0,
bn=False,
relu=False))
self.pafs = nn.Sequential(
conv(num_channels, 512, kernel_size=1, padding=0, bn=False),
conv(512, num_pafs, kernel_size=1, padding=0, bn=False,
relu=False))
