same test, but different result

[Zjsxstar]

The test was repeated several times using the test file and found that the subpixel output was slightly different most times. sometimes the same, but it should always be the same.
And I try some solutions: 1. In test.py, I add random seed and add model.eval(). 2. In lightning_loftr.py . I add self.matcher = self. matcher.eval() behind "self.dump_dir = dump_dir".