Search papers, labs, and topics across Lattice.
This paper introduces NaviCache, a novel test-time self-calibration method for Video Diffusion Models (VDMs) that addresses the computational inefficiencies associated with offline calibration methods. By re-conceptualizing feature evolution as an Inertial Navigation System (INS) problem, NaviCache effectively models the coupling between input and output variations, allowing for adaptive tracking of feature changes. Experimental results across multiple video datasets show that NaviCache significantly improves error judgment for computation skipping while maintaining high performance levels.
NaviCache redefines how we approach computational efficiency in video generation, achieving superior error judgment and performance without the burdens of traditional calibration methods.
Video Diffusion Models (VDMs) is constrained by immense computational costs. While offline calibration-based acceleration suffers from calibration data dependency, prohibitive calibration duration, and susceptibility to distribution shifts, offline calibration-free methods eliminate these hurdles. However, since they rely on instantaneous zero-order approximations where the mapping between input and output differences varies in real-time, they are susceptible to observational noise and ignore the intrinsic momentum within the diffusion trajectory. In this paper, we propose NaviCache, a plug-and-play test-time self-calibration method re-conceptualizing feature evolution as an Inertial Navigation System (INS) problem. NaviCache bridges the fundamental domain gap and the non-stationary nature of diffusion by modeling the relative coupling between input and output variations. We introduce a dual-state estimation architecture that adaptively tracks the feature change ratio and its latent drift, initialized via a specialized Initial Alignment phase. By integrating a time-dependent noise schedule with an uncertainty-aware Measurement Update mechanism, NaviCache provides a theoretically grounded mechanism for error-bounded computation skipping. Extensive experiments on the HunyuanVideo, Wan, and Open-Sora series demonstrate that NaviCache exhibits more accurate error judgment for computation skipping and achieves outstanding comprehensive performance.
