def tf_model_pth(checkpoint_path, pth_output_path):
model = FlowNetSD.FlowNetSD(batchNorm=False).eval()
state_dict = model.state_dict()
with open(checkpoint_path) as f:
ckptFileName = f.readline().split('"')[1]
reader = tf.train.NewCheckpointReader(ckptFileName)
pth_keys = state_dict.keys()
keys = sorted(reader.get_variable_to_shape_map().keys())
for pth_key in pth_keys:
pth_keySplits = pth_key.split('.')
key_pre = 'FlowNetSD/'
if pth_keySplits[0][:8] == 'upsample':
key = key_pre + 'upsample_flow' + pth_keySplits[0][
-6] + 'to' + pth_keySplits[0][-1] + '/' + pth_keySplits[-1]
elif pth_keySplits[0][:5] == 'inter':
key = key_pre + 'interconv' + pth_keySplits[0][
-1] + '/' + pth_keySplits[-1]
else:
key = key_pre + pth_keySplits[0] + '/' + pth_keySplits[-1]
if pth_keySplits[-1] == 'weight':
tensor = reader.get_tensor(key + 's')
state_dict[pth_key] = torch.from_numpy(tensor).permute(
[3, 2, 0, 1])
else:
tensor = reader.get_tensor(key + 'es')
state_dict[pth_key] = torch.from_numpy(tensor)
torch.save(model.state_dict(), pth_output_path)
def __init__(self, args, batchNorm=False, div_flow=20.):
super(FlowNet2, self).__init__()
self.batchNorm = batchNorm
self.div_flow = div_flow
self.rgb_max = args.rgb_max
self.args = args
self.channelnorm = ChannelNorm()
self.pad = Padding()
# First Block (FlowNetC)
self.flownetc = FlowNetC.FlowNetC(args, batchNorm=self.batchNorm)
self.upsample1 = nn.Upsample(scale_factor=4, mode='bilinear')
if args.fp16:
self.resample1 = nn.Sequential(tofp32(), Resample2d(), tofp16())
else:
self.resample1 = Resample2d()
# Block (FlowNetS1)
self.flownets_1 = FlowNetS.FlowNetS(args, batchNorm=self.batchNorm)
self.upsample2 = nn.Upsample(scale_factor=4, mode='bilinear')
if args.fp16:
self.resample2 = nn.Sequential(tofp32(), Resample2d(), tofp16())
else:
self.resample2 = Resample2d()
# Block (FlowNetS2)
self.flownets_2 = FlowNetS.FlowNetS(args, batchNorm=self.batchNorm)
# Block (FlowNetSD)
self.flownets_d = FlowNetSD.FlowNetSD(args, batchNorm=self.batchNorm)
self.upsample3 = nn.Upsample(scale_factor=4, mode='nearest')
self.upsample4 = nn.Upsample(scale_factor=4, mode='nearest')
if args.fp16:
self.resample3 = nn.Sequential(tofp32(), Resample2d(), tofp16())
else:
self.resample3 = Resample2d()
if args.fp16:
self.resample4 = nn.Sequential(tofp32(), Resample2d(), tofp16())
else:
self.resample4 = Resample2d()
# Block (FLowNetFusion)
self.flownetfusion = FlowNetFusion.FlowNetFusion(
args, batchNorm=self.batchNorm)
for m in self.modules():
if isinstance(m, nn.Conv2d):
if m.bias is not None:
init.uniform(m.bias)
init.xavier_uniform(m.weight)
if isinstance(m, nn.ConvTranspose2d):
if m.bias is not None:
init.uniform(m.bias)
init.xavier_uniform(m.weight)
def __init__(self, rgb_max=1, batchNorm=False, div_flow=20.):
super(FlowNet2, self).__init__()
self.batchNorm = batchNorm
self.div_flow = div_flow
self.rgb_max = rgb_max
self.channelnorm = ChannelNorm()
# First Block (FlowNetC)
self.flownetc = FlowNetC.FlowNetC(batchNorm=self.batchNorm)
self.upsample1 = nn.Upsample(scale_factor=4, mode='bilinear')
self.resample1 = Resample2d()
# Block (FlowNetS1)
self.flownets_1 = FlowNetS.FlowNetS(batchNorm=self.batchNorm)
self.upsample2 = nn.Upsample(scale_factor=4, mode='bilinear')
self.resample2 = Resample2d()
# Block (FlowNetS2)
self.flownets_2 = FlowNetS.FlowNetS(batchNorm=self.batchNorm)
# Block (FlowNetSD)
self.flownets_d = FlowNetSD.FlowNetSD(batchNorm=self.batchNorm)
self.upsample3 = nn.Upsample(scale_factor=4, mode='nearest')
self.upsample4 = nn.Upsample(scale_factor=4, mode='nearest')
self.resample3 = Resample2d()
self.resample4 = Resample2d()
# Block (FLowNetFusion)
self.flownetfusion = FlowNetFusion.FlowNetFusion(
batchNorm=self.batchNorm)
def __init__(self, batchNorm=False, div_flow=20.):
super(FlowNet2, self).__init__()
self.batchNorm = batchNorm
self.div_flow = div_flow
self.rgb_max = 255.
self.channelnorm = ChannelNorm()
# First Block (FlowNetC)
self.flownetc = FlowNetC.FlowNetC(batchNorm=self.batchNorm)
self.upsample1 = nn.Upsample(scale_factor=4, mode='bilinear')
self.resample1 = Resample2d()
# Block (FlowNetS1)
self.flownets_1 = FlowNetS.FlowNetS(batchNorm=self.batchNorm)
self.upsample2 = nn.Upsample(scale_factor=4, mode='bilinear')
self.resample2 = Resample2d()
# Block (FlowNetS2)
self.flownets_2 = FlowNetS.FlowNetS(batchNorm=self.batchNorm)
# Block (FlowNetSD)
self.flownets_d = FlowNetSD.FlowNetSD(batchNorm=self.batchNorm)
self.upsample3 = nn.Upsample(scale_factor=4, mode='nearest')
self.upsample4 = nn.Upsample(scale_factor=4, mode='nearest')
self.resample3 = Resample2d()
self.resample4 = Resample2d()
# Block (FLowNetFusion)
self.flownetfusion = FlowNetFusion.FlowNetFusion(
batchNorm=self.batchNorm)
for m in self.modules():
if isinstance(m, nn.Conv2d):
if m.bias is not None:
init.uniform_(m.bias)
init.xavier_uniform_(m.weight)
if isinstance(m, nn.ConvTranspose2d):
if m.bias is not None:
init.uniform_(m.bias)
init.xavier_uniform_(m.weight)
def test_flownet():
flownetSD = FlowNetSD.FlowNetSD().cuda().eval()
a = flownetSD.state_dict()
print(a)
def __init__(self, cfg, batchNorm=False, div_flow = 20.):
super(FlowNet2,self).__init__()
self.batchNorm = batchNorm
self.div_flow = div_flow
self.rgb_max = cfg.MODEL.auxiliary.optical_flow.rgb_max
self.cfg = cfg
self.channelnorm = ChannelNorm()
# First Block (FlowNetC)
self.flownetc = FlowNetC.FlowNetC(cfg, batchNorm=self.batchNorm)
self.upsample1 = nn.Upsample(scale_factor=4, mode='bilinear')
# import ipdb; ipdb.set_trace()
if cfg.MODEL.auxiliary.optical_flow.fp16:
self.resample1 = nn.Sequential(
tofp32(),
Resample2d(),
tofp16())
else:
self.resample1 = Resample2d()
# Block (FlowNetS1)
self.flownets_1 = FlowNetS.FlowNetS(cfg, batchNorm=self.batchNorm)
self.upsample2 = nn.Upsample(scale_factor=4, mode='bilinear')
if cfg.MODEL.auxiliary.optical_flow.fp16:
self.resample2 = nn.Sequential(
tofp32(),
Resample2d(),
tofp16())
else:
self.resample2 = Resample2d()
# Block (FlowNetS2)
self.flownets_2 = FlowNetS.FlowNetS(cfg, batchNorm=self.batchNorm)
# Block (FlowNetSD)
self.flownets_d = FlowNetSD.FlowNetSD(cfg, batchNorm=self.batchNorm)
self.upsample3 = nn.Upsample(scale_factor=4, mode='nearest')
self.upsample4 = nn.Upsample(scale_factor=4, mode='nearest')
if cfg.MODEL.auxiliary.optical_flow.fp16:
self.resample3 = nn.Sequential(
tofp32(),
Resample2d(),
tofp16())
else:
self.resample3 = Resample2d()
if cfg.MODEL.auxiliary.optical_flow.fp16:
self.resample4 = nn.Sequential(
tofp32(),
Resample2d(),
tofp16())
else:
self.resample4 = Resample2d()
# Block (FLowNetFusion)
self.flownetfusion = FlowNetFusion.FlowNetFusion(cfg, batchNorm=self.batchNorm)
for m in self.modules():
if isinstance(m, nn.Conv2d):
if m.bias is not None:
init.uniform_(m.bias)
init.xavier_uniform_(m.weight)
if isinstance(m, nn.ConvTranspose2d):
if m.bias is not None:
init.uniform_(m.bias)
init.xavier_uniform_(m.weight)
