Question about your tech report

[GewelsJI]

Hey, @zhuyiche

Thanks for open sourcing your nice work.

I have a small question about the details of conducting compared experiments in Figure 4 (left figure). (arXiv)

You said that you 'standardized the use of ViT-Large with a 14-patch size and an input resolution of 336 pixels'. But the original checkpoints would not support such operation, otherwise interoperate them before loading. Such as DINOv2 model only provides 224px model in https://huggingface.co/facebook/dinov2-large.

Looking forward to your quick reply.

Best,
Daniel