def forward(self, data_shot, data_query):
proto = self.encoder(data_shot)
if self.args.hyperbolic:
proto = self.e2p(proto)
if self.training:
proto = proto.reshape(self.args.shot, self.args.way, -1)
else:
proto = proto.reshape(self.args.shot, self.args.validation_way,
-1)
proto = poincare_mean(proto, dim=0, c=self.e2p.c)
data_query = self.e2p(self.encoder(data_query))
logits = -dist_matrix(data_query, proto,
c=self.e2p.c) / self.args.temperature
else:
if self.training:
proto = proto.reshape(self.args.shot, self.args.way,
-1).mean(dim=0)
else:
proto = proto.reshape(self.args.shot, self.args.validation_way,
-1).mean(dim=0)
logits = euclidean_metric(self.encoder(data_query),
proto) / self.args.temperature
return logits
def forward(self, data_shot, data_query):
if self.args.hyperbolic:
proto = self.encoder(data_shot)
if self.training:
proto = proto.reshape(self.args.shot, self.args.way, -1)
else:
proto = proto.reshape(self.args.shot, self.args.validation_way,
-1)
proto = poincare_mean(proto, dim=0, c=self.encoder.get_c())
data_query = self.encoder(data_query)
logits = -dist_matrix(data_query, proto, c=self.encoder.get_c()
) / self.args.temperature
else:
print('Does not support non-hyperbolic!')
return logits