Regarding creating the ImageNet validation set

vision
1

I am doing a hyper-parameter tuning for self-supervised learning using ImageNet. BYOL [link] states (in Appendix D.1) that they used ‘the last 10009 last images of the official tensorflow ImageNet split’.

When I use torchvision.datasets.ImageNet, the last 10009 images only consist of images from couple classes. I don’t find any image_id for training set in devkit for ImageNet. How can I extract the ‘last’ 10009 images from ImageNet training set? Thank you ahead.

2

Please see the here: https://pytorch.org/docs/stable/data.html#torch.utils.data.random_split
and this post: https://discuss.pytorch.org/t/how-to-split-dataset-into-test-and-validation-sets/33987