We propose Confidence-Guided Token Merging (Co-Me), an acceleration mechanism for visual geometric transformers without retraining or finetuning the base model. Co-Me distilled a light-weight confidence predictor to rank tokens by uncertainty and selectively merge low-confidence ones, effectively reducing computation while maintaining spatial coverage. Compared to similarity-based merging or pruning, the confidence signal in Co-Me reliably indicates regions emphasized by the transformer, enabling substantial acceleration without degrading performance. Co-Me applies seamlessly to various multi-view and streaming visual geometric transformers, achieving speedups that scale with sequence length. When applied to VGGT and MapAnything, Co-Me achieves up to $11.3\times$ and $7.2\times$ speedup, making visual geometric transformers practical for real-time 3D perception and reconstruction.

翻译：本文提出置信度引导令牌合并（Co-Me），一种无需对基础模型进行重新训练或微调的视觉几何Transformer加速机制。Co-Me通过训练一个轻量级置信度预测器对令牌进行不确定性排序，并选择性合并低置信度令牌，在保持空间覆盖的同时有效减少计算量。与基于相似性的合并或剪枝方法相比，Co-Me中的置信度信号能可靠指示Transformer关注的重点区域，从而在保证性能不下降的前提下实现显著加速。Co-Me可无缝应用于多种多视角和流式视觉几何Transformer，其加速效果随序列长度增加而提升。在VGGT和MapAnything模型上的实验表明，Co-Me分别实现了最高$11.3\\times$和$7.2\\times$的加速比，使视觉几何Transformer在实时三维感知与重建任务中具备实际应用价值。