动态调整学习率，让3D重建更稳定可靠
ReCal3R: Reliability-Calibrated Learning Rates for Streaming 3D Reconstruction
- 根据场景状态可靠性校准每个令牌的学习率
- 在长序列上实现3.7倍的位姿误差降低
- 无需训练，适合实时3D重建系统
流式3D重建依赖紧凑的循环场景状态，在线处理长图像流，时间复杂度线性、内存受限。但重复更新会逐渐污染该状态，使可靠的历史信息被噪声或模糊观测覆盖。本文提出ReCal3R，一种用于循环3D重建的可靠性校准学习率方法。不直接使用候选学习率，而是从维护的场景状态中估计令牌可靠性，并基于令牌对齐、状态重建残差和近期更新压力，校准候选学习率。最终得到逐令牌的学习率，在保守基线率与候选率之间插值，抑制不可靠令牌的激进更新，同时保留对高信息量帧的适应能力。作为训练无关的校准规则应用于CUT3R，ReCal3R在长序列的位姿、深度与重建质量上均表现优异，实现ATE降低3.7×，且运行时与内存开销相当。代码已公开：https://github.com/Powertony102/ReCal3R。
原文摘要 · Abstract (English)
Streaming 3D reconstruction relies on a compact recurrent scene state to process long image streams in linear time and bounded memory. However, repeated updates can gradually corrupt this state, causing reliable historical information to be overwritten by noisy or ambiguous observations. We introduce ReCal3R, a reliability-calibrated learning rate method for recurrent 3D reconstruction. Instead of directly applying a candidate learning rate, our method estimates state token reliability from the maintained scene state and uses it to calibrate a candidate learning rate derived from token alignment, state reconstruction residual, and recent update pressure. The resulting token-wise learning rate interpolates between a conservative base rate and the candidate rate, suppressing aggressive updates on unreliable tokens while preserving adaptation to informative frames. Applied to CUT3R as a training-free calibration rule, ReCal3R reaches strong performance on long sequences in pose, depth, and reconstruction quality, including a 3.7$\times$ reduction in ATE, with comparable runtime and memory. Code is available at: https://github.com/Powertony102/ReCal3R.
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。