checkpoint_dir is a required argument

pytorch3dunet/unet3d/trainer.py

@@ -142,7 +142,7 @@ def __init__(self, model, optimizer, lr_scheduler, loss_criterion, eval_criterio
         elif pre_trained is not None:
             logger.info(f"Logging pre-trained model from '{pre_trained}'...")
             utils.load_checkpoint(pre_trained, self.model, None)
-            if 'checkpoint_dir' not in kwargs:
+            if not self.checkpoint_dir:
                 self.checkpoint_dir = os.path.split(pre_trained)[0]
 
     def fit(self):