AdaGaR：面向动态场景重建的自适应Gabor表征

摘要

基于单目视频重建动态三维场景需同时捕捉高频外观细节与时间连续运动。现有采用单一高斯基元的方法受限于其低通滤波特性，而标准Gabor函数存在能量不稳定问题。此外，时间连续性约束的缺失常导致插值过程中出现运动伪影。我们提出AdaGaR这一统一框架，在显式动态场景建模中同时解决频率自适应性与时间连续性问题。通过引入自适应Gabor表征，我们扩展高斯基元至可学习频率权重与自适应能量补偿，以平衡细节捕捉与稳定性。针对时间连续性，采用带时间曲率正则化的三次埃尔米特样条确保平滑运动演化。结合深度估计、点追踪与前景掩码的自适应初始化机制，在训练初期建立稳定点云分布。Tap-Vid DAVIS数据集实验表明，该方法在峰值信噪比（35.49）、结构相似性（0.9433）和感知相似度（0.0723）上达到最优性能，并在帧插值、深度一致性、视频编辑与立体视图合成任务中展现强大泛化能力。项目页面：https://jiewenchan.github.io/AdaGaR/

English

Reconstructing dynamic 3D scenes from monocular videos requires simultaneously capturing high-frequency appearance details and temporally continuous motion. Existing methods using single Gaussian primitives are limited by their low-pass filtering nature, while standard Gabor functions introduce energy instability. Moreover, lack of temporal continuity constraints often leads to motion artifacts during interpolation. We propose AdaGaR, a unified framework addressing both frequency adaptivity and temporal continuity in explicit dynamic scene modeling. We introduce Adaptive Gabor Representation, extending Gaussians through learnable frequency weights and adaptive energy compensation to balance detail capture and stability. For temporal continuity, we employ Cubic Hermite Splines with Temporal Curvature Regularization to ensure smooth motion evolution. An Adaptive Initialization mechanism combining depth estimation, point tracking, and foreground masks establishes stable point cloud distributions in early training. Experiments on Tap-Vid DAVIS demonstrate state-of-the-art performance (PSNR 35.49, SSIM 0.9433, LPIPS 0.0723) and strong generalization across frame interpolation, depth consistency, video editing, and stereo view synthesis. Project page: https://jiewenchan.github.io/AdaGaR/

AdaGaR：面向动态场景重建的自适应Gabor表征

AdaGaR: Adaptive Gabor Representation for Dynamic Scene Reconstruction

摘要

Support