Why Eval with flip and rotation

[robotshanxi]

It looks to me the network training has no flip or rotation, why they are added as evaluation? Thanks

[sunset1995]

Hi,
The network was actually trained with flip and rotation augmentation (see dataset.py).

[robotshanxi]

Thanks for the reply, I was thinking args.no_flip is true. My mistake.