What is the difference between trainer.model and predictor.model in detectron2?

[Nurungyi]

Hi,

I used same dataset in train and test for experiment through cocoevaluation of both difference with same configuration.
result1) train : dataset1, test: dataset1, and trainer.model was used with same configuration.
result2) train : dataset1, test: dataset1, and trainer.model was used with same configuration.

However, both results are different. In document and tutorial, both looks similar in function.
Why they are different?

Thank you