Newest papers

Stream4D: Streaming 4D Reconstruction with Training-free Motion Decoupling via Causal KV-Cache Attention

Wei Zhang, Ying Feng, Mengtong Wu, Shiqiang Gong, Huaqi Zhang, Qi Zhang · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.31320 · cited by 0 works (OpenAlex) · open-access PDF

Portuguese

A reconstrução de cenas 4D requer um gerenciamento robusto de objetos dinâmicos, mas os métodos de transmissão existentes geralmente assumem cenas estáticas, levando a deslizamentos de posição e artefatos fantasmas em longas seqüências. Introduzimos o Stream4D, um framework de reconstrução de transmissão livre de treinamento e de desligamento de movimento que pode ser integrado diretamente em backbones de Transformer causal (por exemplo, StreamVGGT, Stream3R). Nossa principal observação é que a atenção causal codifica implicitamente sinais de movimento temporal ricos dentro da cache KV. Nós agregamos de forma adaptativa essas sugestões a partir de várias camadas de atenção e através de janelas temporais, e as fundimos com antecedentes de profundidade para lidar com casos desa

Machine translation (Meta M2M100); the English original follows.

Abstract

4D scene reconstruction requires robust handling of dynamic objects, yet existing streaming methods commonly assume static scenes, leading to pose drift and ghosting artifacts over long sequences. We introduce Stream4D, a training-free, motion-decoupled streaming reconstruction framework that can be directly integrated into causal Transformer backbones (e.g., StreamVGGT, Stream3R). Our key observation is that causal attention implicitly encodes rich temporal motion cues within the KV cache. We adaptively aggregate these cues from multiple attention layers and across temporal windows, and fuse them with depth priors to handle challenging cases such as occlusions and textureless regions, producing precise dynamic masks that decouple static and dynamic regions. We then apply a soft key masking scheme that leverages these dynamic masks to suppress interference from dynamic regions in causal attention, thereby improving camera pose estimation accuracy. Stream4D inherits the efficiency of causal attention and remains compatible with existing streaming architectures. On diverse dynamic-scene benchmarks, Stream4D significantly improves camera pose estimation and static-scene reconstruction quality, yields reliable dynamic masks, and effectively mitigates long-sequence ghosting.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Computer science, Ghosting, Computer vision, Artificial intelligence, Key (lock), Fuse (electrical)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange