不训练即可减少视觉模型内存占用，保持精度同时提升长序列推理效率
Evict3R: Training-Free Token Eviction for Memory-Bounded Streaming Visual Geometry Transformers
- 推理时自动剔除冗余视觉标记，仅保留关键信息
- 在7-Scenes数据集上峰值内存减半至9.39GB，精度损失不足0.003
- 适合资源受限场景下的长时序视频理解任务
流式视觉变换器如StreamVGGT虽具强3D感知能力，但键值（KV）记忆持续增长，限制可扩展性。本文提出一种无需训练的推理时标记淘汰策略，通过丢弃冗余信息、保留最具判别力的标记来控制内存。在7-Scenes长序列测试中，峰值内存从18.63 GB降至9.39 GB，精度与完整度仅下降0.003；在严格内存约束下，该策略支持更高帧率采样，显著提升重建精度。跨视频深度估计（Sintel、KITTI）、3D重建（7-Scenes、NRGBD）及相机位姿估计（Sintel、TUM-dynamics）实验表明，本方法以极低内存开销接近StreamVGGT性能，使长时序流式推理更具可行性。
原文摘要 · Abstract (English)
Streaming visual transformers like StreamVGGT achieve strong 3D perception but suffer from unbounded growth of key value (KV) memory, which limits scalability. We propose a training-free, inference-time token eviction policy that bounds memory by discarding redundant tokens while keeping the most informative ones. Our method uses significantly less memory with little to no drop in accuracy: on 7-Scenes with long sequences it reduces peak memory from 18.63 GB to 9.39 GB while accuracy and completeness drop by only 0.003. Under strict memory budgets, eviction enables denser frame sampling, which improves reconstruction accuracy compared to the baseline. Experiments across video depth estimation (Sintel, KITTI), 3D reconstruction (7-Scenes, NRGBD), and camera pose estimation (Sintel, TUM-dynamics) show that our approach closely matches StreamVGGT at a fraction of the memory and makes long-horizon streaming inference more practical.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。