RAFT small model evaluate?

[plateau1]

Thank you for a great job!

I used the RAFT small-model(raft-small.pth) for evaluation and the results I got are as follows:
Sintel clean: 2.19
Sintel Final: 3.38
KITTI2015 epe: 8.46
KITTI2015 F1: 26.68%

But the accuracy provided in your paper is:
Sintel clean: 2.21
Sintel Final: 3.35
KITTI2015 epe: 7.51
KITTI2015 F1: 26.9%

May I ask what is the reason for this?