def setSource(self, Is_org):
with torch.no_grad():
IsList = []
for i in range(len(self.scaleList)):
IsList.append(
self.ResizeMinSize(Is_org,
int(self.minSize * self.scaleList[i])))
self.Is = IsList[len(self.scaleList) // 2]
if self.use_cuda:
self.IsTensor = self.toTensor(self.Is).unsqueeze(0).cuda()
else:
self.IsTensor = self.toTensor(self.Is).unsqueeze(0)
self.featsMultiScale = []
self.WMultiScale = []
self.HMultiScale = []
for i in range(len(self.scaleList)):
if self.use_cuda:
feat = F.normalize(
self.net(self.preproc(IsList[i]).unsqueeze(0).cuda()))
else:
feat = F.normalize(
self.net(self.preproc(IsList[i]).unsqueeze(0)))
Ws, Hs = outil.getWHTensor(feat)
self.featsMultiScale.append(feat.contiguous().view(1024, -1))
self.WMultiScale.append(Ws)
self.HMultiScale.append(Hs)
self.featsMultiScale = torch.cat(self.featsMultiScale, dim=1)
self.WMultiScale = torch.cat(self.WMultiScale)
self.HMultiScale = torch.cat(self.HMultiScale)
def setTarget(self, It_org):
with torch.no_grad():
self.It = self.ResizeMaxSize(It_org, self.minSize)
self.ItTensor = self.toTensor(self.It).unsqueeze(0).cuda()
self.featt = F.normalize(
self.net(self.preproc(self.It).unsqueeze(0).cuda()))
self.Wt, self.Ht = outil.getWHTensor(self.featt)
def setPair(self, Is_org, It_org):
with torch.no_grad():
IsList = []
for i in range(len(self.scaleList)):
IsList.append(
self.ResizeMinSize(Is_org,
int(self.minSize * self.scaleList[i])))
self.Is = IsList[len(self.scaleList) // 2]
self.IsTensor = self.toTensor(self.Is).unsqueeze(0).cuda()
featsMultiScale = []
WMultiScale = []
HMultiScale = []
for i in range(len(self.scaleList)):
feat = F.normalize(
self.net(self.preproc(IsList[i]).unsqueeze(0).cuda()))
Ws, Hs = outil.getWHTensor(feat)
featsMultiScale.append(feat.contiguous().view(1024, -1))
WMultiScale.append(Ws)
HMultiScale.append(Hs)
torch.cuda.empty_cache()
featsMultiScale = torch.cat(featsMultiScale, dim=1)
WMultiScale = torch.cat(WMultiScale)
HMultiScale = torch.cat(HMultiScale)
torch.cuda.empty_cache()
self.It = self.ResizeMinSize(It_org, self.minSize)
self.ItTensor = self.toTensor(self.It).unsqueeze(0).cuda()
featt = F.normalize(
self.net(self.preproc(self.It).unsqueeze(0).cuda()))
Wt, Ht = outil.getWHTensor(featt)
WtInt, HtInt = outil.getWHTensor_Int(featt)
self.W2, self.H2 = featt.size()[2], featt.size()[3]
featt = featt.contiguous().view(1024, -1)
index1, index2 = outil.mutualMatching(featsMultiScale, featt)
self.W1MutualMatch = WMultiScale[index1]
self.H1MutualMatch = HMultiScale[index1]
self.W2MutualMatch = Wt[index2]
self.H2MutualMatch = Ht[index2]
self.W2MutualMatchInt = WtInt[index2]
self.H2MutualMatchInt = HtInt[index2]