In the forward function of CombinatorialMutationsLossModule, is it correct to randomly initialize nn.Conv2D every time op == 'concat'?

[Luious-LYH]

LoMix/trainer.py

Line 184 in fc1a215

conv = nn.Conv2d(cat.size(1), self.num_classes, kernel_size=1).to(device)

def forward(.....):
......
elif op == 'concat':
cat = torch.cat([output_maps[c] for c in comb], dim=1)
conv = nn.Conv2d(cat.size(1), self.num_classes, kernel_size=1).to(device) # Is it correct to randomly initialize?
mutated = conv(cat)