Pretrained Torch vision models

[OBouldjedri]

I was wondering on what size of ImageNet, the pretrained models of torch vision were pre-trained ?

ImageNet 1K
or
ImageNet21K
or
ImageNet22K

[ptrblck]

Based on this older post:

Is there pretrained CNN (e.g. ResNet) for CIFAR-10 or CIFAR-100?

they’re all pre-trained on Imagenet-12

they are pretrained on 2012 version of it and I would assume on the 1K classes version, since all pretrained classification models have 1000 output logits.