Zero-Ablation Overstates Register Content Dependence in DINO Vision Transformers

arXiv:2604.14433v1 Announce Type: new Abstract: Zero-ablation -- replacing token activations with zero vectors -- is widely used to probe token function in vision transformers. Register zeroing in DINOv2+registers and DINOv3 produces large drops (up to -36.6,pp classification, -30.9,pp segmentation), suggesting registers are functionally indispensable. However, three replacement controls -- mean-substitution, noise-substitution, and cross-image register-shuffling -- preserve performance across classification, correspondence, and segmentation, remaining within {sim}1,pp of the unmodified baseline. Per-patch cosine similarity shows these replacements genuinely perturb internal representations, while zeroing causes disproportionately large perturbations, consistent with why it alone degrades tasks. We conclude that zero-ablation overstates dependence on exact register content. In the frozen-feature evaluations we test, performance depends on plausible register-like activations rather than on exact image-specific values. Registers nevertheless buffer dense features from exttt{[CLS]} dependence and are associated with compressed patch geometry. These findings, including the replacement-control results, replicate at ViT-B scale.

Related

• Can Cross-Layer Transcoders Replace Vision Transformer Activations? An Interpretable Perspective on Vision
• Vision Transformers Need More Than Registers
• VisPCO: Visual Token Pruning Configuration Optimization via Budget-Aware Pareto-Frontier Learning for Vision-Language Models
• Does Visual Token Pruning Improve Calibration? An Empirical Study on Confidence in MLLMs

Source: arXiv cs.CV | 2026-04-17