Hello,
Is there any possibility of release of the pre-trained models of the other datasets which were shown in the paper, ie, shapes, sprites and PennAction datasets ?
Or otherwise, it'd also be really helpful if you cloud share the training code instead.
Many Thanks !
Hello, Is there any possibility of release of the pre-trained models of the other datasets which were shown in the paper, ie, shapes, sprites and PennAction datasets ? Or otherwise, it'd also be really helpful if you cloud share the training code instead. Many Thanks !