def forward(self, img1, img2, objectPrevious):
Flow = None
Features = None
if objectPrevious is None:
Cost_Volume = F.leaky_relu(
input=correlation.FunctionCorrelation(
tensorFirst=img1, tensorSecond=img2),
negative_slope=.1)
Features = Cost_Volume
else:
Flow = self.Upflow(objectPrevious['flow'])
Features = self.Upfeature(objectPrevious['feature'])
Cost_Volume = F.leaky_relu(
input=correlation.FunctionCorrelation(
tensorFirst=img1,
tensorSecond=Warp(img2, Flow * self.scales)),
negative_slope=.1)
Features = torch.cat([Cost_Volume, img1, Flow, Features],
dim=1)
# end
Features = torch.cat([self.conv1(Features), Features], dim=1)
Features = torch.cat([self.conv2(Features), Features], dim=1)
Features = torch.cat([self.conv3(Features), Features], dim=1)
Features = torch.cat([self.conv4(Features), Features], dim=1)
Features = torch.cat([self.conv5(Features), Features], dim=1)
Flow = self.conv6(Features)
return {'flow': Flow, 'feature': Features}
def forward(self, tensorFirst, tensorSecond, objectPrevious):
tensorFlow = None
tensorFeat = None
if objectPrevious is None:
tensorFlow = None
tensorFeat = None
tensorVolume = torch.nn.functional.leaky_relu(input=correlation.FunctionCorrelation(tensorFirst=tensorFirst, tensorSecond=tensorSecond), negative_slope=0.1, inplace=False)
tensorFeat = torch.cat([ tensorVolume ], 1)
elif objectPrevious is not None:
tensorFlow = self.moduleUpflow(objectPrevious['tensorFlow'])
tensorFeat = self.moduleUpfeat(objectPrevious['tensorFeat'])
tensorVolume = torch.nn.functional.leaky_relu(input=correlation.FunctionCorrelation(tensorFirst=tensorFirst, tensorSecond=Backward(tensorInput=tensorSecond, tensorFlow=tensorFlow * self.dblBackward)), negative_slope=0.1, inplace=False)
tensorFeat = torch.cat([ tensorVolume, tensorFirst, tensorFlow, tensorFeat], 1)
# end
tensorFeat = torch.cat([ self.moduleOne(tensorFeat), tensorFeat], 1)
tensorFeat = torch.cat([ self.moduleTwo(tensorFeat), tensorFeat], 1)
tensorFeat = torch.cat([ self.moduleThr(tensorFeat), tensorFeat], 1)
tensorFeat = torch.cat([ self.moduleFou(tensorFeat), tensorFeat], 1)
tensorFeat = torch.cat([ self.moduleFiv(tensorFeat), tensorFeat], 1)
tensorFlow = self.moduleSix(tensorFeat)
return {
'tensorFlow': tensorFlow,
'tensorFeat': tensorFeat
}
def forward(self, tensorFirst, tensorSecond, tensorFeaturesFirst,
tensorFeaturesSecond, tensorFlow):
tensorFeaturesFirst = self.moduleFeat(tensorFeaturesFirst)
tensorFeaturesSecond = self.moduleFeat(tensorFeaturesSecond)
if tensorFlow is not None:
tensorFlow = self.moduleUpflow(tensorFlow)
if tensorFlow is not None:
tensorFeaturesSecond = warp(
tensorInput=tensorFeaturesSecond,
tensorFlow=tensorFlow * self.dblBackward)
if self.moduleUpcorr is None:
tensorCorrelation = torch.nn.functional.leaky_relu(
input=correlation.FunctionCorrelation(
tensorFirst=tensorFeaturesFirst,
tensorSecond=tensorFeaturesSecond,
intStride=1),
negative_slope=0.1,
inplace=False)
elif self.moduleUpcorr is not None:
tensorCorrelation = self.moduleUpcorr(
torch.nn.functional.leaky_relu(
input=correlation.FunctionCorrelation(
tensorFirst=tensorFeaturesFirst,
tensorSecond=tensorFeaturesSecond,
intStride=2),
negative_slope=0.1,
inplace=False))
return (tensorFlow if tensorFlow is not None else
0.0) + self.moduleMain(tensorCorrelation)
def forward(self, tensorFirst, tensorSecond, objectPrevious):
try:
from correlation import correlation # the custom cost volume layer
except:
sys.path.insert(0, './correlation')
import correlation # you should consider upgrading python
tensorFlow = None
tensorFeat = None
if objectPrevious is None:
tensorFlow = None
tensorFeat = None
tensorVolume = torch.nn.functional.leaky_relu(
input=correlation.FunctionCorrelation(
tensorFirst=tensorFirst,
tensorSecond=tensorSecond),
negative_slope=0.1,
inplace=False)
tensorFeat = torch.cat([tensorVolume], 1)
elif objectPrevious is not None:
tensorFlow = self.moduleUpflow(
objectPrevious['tensorFlow'])
tensorFeat = self.moduleUpfeat(
objectPrevious['tensorFeat'])
tensorVolume = torch.nn.functional.leaky_relu(
input=correlation.FunctionCorrelation(
tensorFirst=tensorFirst,
tensorSecond=Backward(tensorInput=tensorSecond,
tensorFlow=tensorFlow *
self.dblBackward)),
negative_slope=0.1,
inplace=False)
tensorFeat = torch.cat(
[tensorVolume, tensorFirst, tensorFlow, tensorFeat], 1)
# end
tensorFeat = torch.cat(
[self.moduleOne(tensorFeat), tensorFeat], 1)
tensorFeat = torch.cat(
[self.moduleTwo(tensorFeat), tensorFeat], 1)
tensorFeat = torch.cat(
[self.moduleThr(tensorFeat), tensorFeat], 1)
tensorFeat = torch.cat(
[self.moduleFou(tensorFeat), tensorFeat], 1)
tensorFeat = torch.cat(
[self.moduleFiv(tensorFeat), tensorFeat], 1)
tensorFlow = self.moduleSix(tensorFeat)
return {'tensorFlow': tensorFlow, 'tensorFeat': tensorFeat}
def forward(self, tenFirst, tenSecond, tenFeaturesFirst, tenFeaturesSecond, tenFlow): tenFeaturesFirst = self.netFeat(tenFeaturesFirst) tenFeaturesSecond = self.netFeat(tenFeaturesSecond) if tenFlow is not None: tenFlow = self.netUpflow(tenFlow) # end if tenFlow is not None: tenFeaturesSecond = backwarp(tenInput=tenFeaturesSecond, tenFlow=tenFlow * self.fltBackwarp) # end if self.netUpcorr is None: tenCorrelation = torch.nn.functional.leaky_relu(input=correlation.FunctionCorrelation(tenFirst=tenFeaturesFirst, tenSecond=tenFeaturesSecond, intStride=1), negative_slope=0.1, inplace=False) elif self.netUpcorr is not None: tenCorrelation = self.netUpcorr(torch.nn.functional.leaky_relu(input=correlation.FunctionCorrelation(tenFirst=tenFeaturesFirst, tenSecond=tenFeaturesSecond, intStride=2), negative_slope=0.1, inplace=False)) # end return (tenFlow if tenFlow is not None else 0.0) + self.netMain(tenCorrelation)
