Masking only the student's global view during DINO training on mini-ImageNet gives +0.76 linear-probe top-1 and essentially no k-NN change.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Rethinking Random Masking in Self-Distillation on ViT
Masking only the student's global view during DINO training on mini-ImageNet gives +0.76 linear-probe top-1 and essentially no k-NN change.