train: Fix saving best val loss weights

Author: synml

train.py

@@ -77,7 +77,7 @@
             prev_miou = miou
 
         # Best val_loss를 가진 모델을 저장
-        if val_loss > prev_val_loss:
+        if val_loss < prev_val_loss:
             torch.save(model.state_dict(), os.path.join('weights', '{}_val_best.pth'.format(config['model'])))
             prev_val_loss = val_loss
     writer.close()