Fixed small import errors in s2sd.py

criteria/s2sd.py

@@ -64,7 +64,7 @@ def __init__(self, opt):
  # Initialize all target criteria. As each criterion may require its
  # separate set of trainable parameters, several instances have to be created.
  old_embed_dim = copy.deepcopy(opt.embed_dim)
- self.target_criteria = nn.ModuleList()
+ self.target_criteria = torch.nn.ModuleList()
  for t_dim in opt.loss_s2sd_target_dims:
  opt.embed_dim = t_dim
 
@@ -127,8 +127,8 @@ class [0,...,C-1], shape: (BS x 1)
  # If required, use combined global max- and average pooling to produce
  # the feature space.
  if self.pool_aggr:
- avg_batch_features = nn.AdaptiveAvgPool2d(1)(batch_features).view(
- bs, -1) + nn.AdaptiveMaxPool2d(1)(batch_features).view(bs, -1)
+ avg_batch_features = torch.nn.AdaptiveAvgPool2d(1)(batch_features).view(
+ bs, -1) + torch.nn.AdaptiveMaxPool2d(1)(batch_features).view(bs, -1)
  else:
  avg_batch_features = avg_batch_features.view(bs, -1)