Hi. Thanks for the great work.
In you paper in Stage 2: 4096 × 4096 Region-Level Pretraining you state:
We apply standard dropout (p = 0.10) to all views following work in Gao et al. [29].
But I can't find it anywhere in your code specially in main_dino4k.py.
Can you please show where is it or elaborate on why it's not in the actual implementation?