ImageNet-HR:
Image files are named with their index (1 of 5,000 images) before the first underscore.
All numbers (separated by underscores) after the image index are acceptable labels for the image. We use the same class indices as ImageNet-1k.

LookHere code:
lookhere.py contains modifications to the base vision transformer implementation (vision_transformer.py). These two files are all that is necessary to train LookHere models with PyTorch.
