lmdb.Error: /xxx/xxx/recognition/CVPR2016/: Stale file handle

[HumanZhong]

I have downloaded the CVPR2016 datasets and use rar e CVPR2016.rar to extract the .mdb files.
However, when running scripts/stn_att_rec.sh, this error is reported:
lmdb.Error: /xxx/xxx/recognition/CVPR2016/: Stale file handle
(/xxx/xxx is my personal path on a remote machine)
Because I am not familiar with lmdb, I wonder if the author or someone may tell me what's wrong with this error or tell me some possible solutions(I've also searched this on google but it seems that few people have encountered this error). Is it a read/write authority problem?

[HumanZhong]

The conda envs follow the environment.yml except that I use pytorch-1.2.0, torchvision-0.4.0 and cuda10.0. Could this be the problem?

[HumanZhong]

This problem has been solved.
lmdb.Error: /xxx/xxx/recognition/CVPR2016/: Stale file handle was raised because I store the .mdb files on a remote shared file system which has a different storage format compared with local machine.
By moving the files from the shared file system to local path, the problem is solved.