freeze pretrained kor w2v

evaluate.py

@@ -17,7 +17,7 @@ def load_metadata(file_path):
  lines = f.readlines()
  for line in lines:
  line = line.strip()
- label = line.split(" ")[5]
+ label = line.split(" ")[2]
  labels.append(label)
  return labels
 
@@ -40,8 +40,8 @@ def load_metadata_from_proto(meta_file_path, proto_file_path):
  lines = f.readlines()
  for line in lines:
  line = line.strip()
- file_name = line.split(" ")[1]
- label = line.split(" ")[5]
+ file_name = line.split(" ")[0]
+ label = line.split(" ")[2]
  if file_name in protos:
  index = protos.index(file_name)
  labels[index] = label
@@ -167,8 +167,8 @@ def calculate_EER(scores, labels):
  # and bonafide otherwise
 
  # create two lists: one for the labels and one for the predictions
- # labels = metadata
- labels = load_metadata_from_proto(args.metadata_file, args.protocol_file)
+ labels = metadata
+ # labels = load_metadata_from_proto(args.metadata_file, args.protocol_file)
  predictions = []
  for i, file_name in enumerate(proto):
  score = scores[i]
@@ -194,5 +194,5 @@ def calculate_EER(scores, labels):
  print(f"TN = {cm[1][1]}")
  print(f"FP = {cm[0][1]}")
  print(f"FN = {cm[1][0]}")
- 
+
  calculate_EER(scores, labels)

oc_classifier.py

@@ -73,9 +73,11 @@ def __init__(self, protocol_file, dataset_dir, eval=False):
  for line in lines:
  line = line.strip()
  line = line.split(" ")
- if line[4] == "bonafide":
- self.file_list.append(line[1])
- self.label_list.append(line[4]) # bonafide only
+ self.file_list.append(line[0])
+ self.label_list.append("bonafide")
+ # if line[4] == "bonafide":
+ # self.file_list.append(line[1])
+ # self.label_list.append(line[4]) # bonafide only
 
  self._length = len(self.file_list) 
 

oc_training.py

@@ -318,6 +318,11 @@ def collate_fn(self, batch):
  device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
 
  aasist = AModel(None, device).to(device)
+ aasist.train()
+
+ aasist.ssl_model.eval()
+ for param in aasist.ssl_model.parameters():
+ param.requires_grad = False
  # ssl = SSLModel(device)
  # senet34 = se_resnet34().to(device)
  # lcnn = lcnn_net(asoftmax=False).to(device)
@@ -349,7 +354,8 @@ def collate_fn(self, batch):
  print(f"Epoch {epoch + 1}\n-------------------------------")
 
  # Training phase
- aasist.train()
+
+ #aasist.train()
  # ssl.eval()
  # senet34.train()
  # lcnn.train()