[Fix]: Fix get train_pipeline method of val workflow (#8575)

Author: Lehsuby

tools/train.py

@@ -217,8 +217,10 @@ def main():
 
     datasets = [build_dataset(cfg.data.train)]
     if len(cfg.workflow) == 2:
+        assert 'val' in [mode for (mode, _) in cfg.workflow]
         val_dataset = copy.deepcopy(cfg.data.val)
-        val_dataset.pipeline = cfg.data.train.pipeline
+        val_dataset.pipeline = cfg.data.train.get(
+            'pipeline', cfg.data.train.dataset.get('pipeline'))
         datasets.append(build_dataset(val_dataset))
     if cfg.checkpoint_config is not None:
         # save mmdet version, config file content and class names in