finish compactness+descriptiveness branches

models/sslassist.py

@@ -502,6 +502,7 @@ def __init__(self, args, device):
         self.pool_hT2 = GraphPool(pool_ratios[2], gat_dims[1], 0.3)
 
         self.out_layer = nn.Linear(5 * gat_dims[1], 2)
+        self.emb_layer = nn.Linear(160, 100)
 
     def forward(self, x):
         #-------pre-trained Wav2vec model fine tunning ------------------------##
@@ -588,10 +589,11 @@ def forward(self, x):
         S_max, _ = torch.max(torch.abs(out_S), dim=1)
         S_avg = torch.mean(out_S, dim=1)
 
-        emb = last_hidden = torch.cat(
+        last_hidden = torch.cat(
             [T_max, T_avg, S_max, S_avg, master.squeeze(1)], dim=1)
 
         last_hidden = self.drop(last_hidden)
+        emb = self.emb_layer(last_hidden)
         output = self.out_layer(last_hidden)
 
         return emb, output

oc_training.py

@@ -377,8 +377,8 @@ def collate_fn(self, batch):
             des = outputs_senet34[1]
             # Calculate the losses
             # c_loss = euclidean_distance_loss(com)
-            c_loss = 0.0*compactness_loss(com)
-            d_loss = 1.0*descriptiveness_loss(des, labels.squeeze(0)) # because labels.shape = torch.Size([1, 8])        
+            c_loss = 0.1*compactness_loss(com)
+            d_loss = 0.9*descriptiveness_loss(des, labels.squeeze(0)) # because labels.shape = torch.Size([1, 8])        
             loss = c_loss + d_loss
 
             loss.backward()