The MOSE valid test results differ significantly from the paper

Hi, thank you for releasing SAM2 and the excellent documentation.

I have a question about SAM2 testing on the MOSE valid. I tested the MOSE validation set using your code, checkpoints, and all the same content as you open sourced, but J&F was only 0.66, which is far from the 0.746 you released. I would like to know if there are any possible issues that I did not notice and hope to receive your reply.

I would greatly appreciate any guidance or reference materials.