add : lars optimizer and weight scheduling according to barlow twin

barlow_twin_like/main.py

@@ -58,7 +58,7 @@ class LARS(optim.Optimizer):
 
 
 def adjust_learning_rate(args, optimizer, loader, step):
-    max_steps = args.epochs * len(loader)
+    max_steps = args.epoches * len(loader)
     warmup_steps = 10 * len(loader)
     base_lr = args.batch_size / 256
     if step < warmup_steps:
@@ -69,8 +69,8 @@ def adjust_learning_rate(args, optimizer, loader, step):
         q = 0.5 * (1 + math.cos(math.pi * step / max_steps))
         end_lr = base_lr * 0.001
         lr = base_lr * q + end_lr * (1 - q)
-    optimizer.param_groups[0]['lr'] = lr * args.learning_rate_weights
-    optimizer.param_groups[1]['lr'] = lr * args.learning_rate_biases
+    optimizer.param_groups[0]['lr'] = lr * 0.2
+    optimizer.param_groups[1]['lr'] = lr * 0.0048
 
 def save_model(model, path):
     print('Model saved')