OmniStream: Mastering Perception, Reconstruction and Action in Continuous Streams

TL;DR

OmniStream is a unified, real-time streaming visual backbone that integrates perception, reconstruction, and action capabilities, enabling versatile and general-purpose visual understanding in continuous video streams.

Contribution

The paper introduces OmniStream, a novel causal spatiotemporal attention-based model trained on multiple tasks and datasets, achieving broad generalization without specialized fine-tuning.

Findings

Competitive performance across diverse tasks and datasets

Effective online processing with a frozen backbone

Demonstrates potential for general-purpose visual agents

Abstract

Modern visual agents require representations that are general, causal, and physically structured to operate in real-time streaming environments. However, current vision foundation models remain fragmented, specializing narrowly in image semantic perception, offline temporal modeling, or spatial geometry. This paper introduces OmniStream, a unified streaming visual backbone that effectively perceives, reconstructs, and acts from diverse visual inputs. By incorporating causal spatiotemporal attention and 3D rotary positional embeddings (3D-RoPE), our model supports efficient, frame-by-frame online processing of video streams via a persistent KV-cache. We pre-train OmniStream using a synergistic multi-task framework coupling static and temporal representation learning, streaming geometric reconstruction, and vision-language alignment on 29 datasets. Extensive evaluations show that, even with a strictly frozen backbone, OmniStream achieves consistently competitive performance with specialized experts across image and video probing, streaming geometric reconstruction, complex video and spatial reasoning, as well as robotic manipulation (unseen at training). Rather than pursuing benchmark-specific dominance, our work demonstrates the viability of training a single, versatile vision backbone that generalizes across semantic, spatial, and temporal reasoning, i.e., a more meaningful step toward general-purpose visual understanding for interactive and embodied agents.