Merge pull request #5019 from BeyondYourself/release/2.4

add a constraint

tools/program.py

@@ -266,7 +266,7 @@ def train(config,
             stats['lr'] = lr
             train_stats.update(stats)
 
-            if cal_metric_during_train:  # only rec and cls need
+            if cal_metric_during_train  and model_type is not "det":  # only rec and cls need
                 batch = [item.numpy() for item in batch]
                 if model_type in ['table', 'kie']:
                     eval_class(preds, batch)