Video World Models with Long-term Spatial Memory

TL;DR

This paper introduces a geometry-grounded long-term spatial memory framework to improve scene consistency in video world models, enabling better long-term environment generation and addressing forgetting issues.

Contribution

The paper proposes a novel long-term spatial memory mechanism inspired by human memory, with custom datasets for training and evaluation of 3D memory in video world models.

Findings

Enhanced scene consistency and quality in generated videos

Improved long-term context retention compared to baselines

Demonstrated effectiveness on custom datasets with 3D memory mechanisms

Abstract

Emerging world models autoregressively generate video frames in response to actions, such as camera movements and text prompts, among other control signals. Due to limited temporal context window sizes, these models often struggle to maintain scene consistency during revisits, leading to severe forgetting of previously generated environments. Inspired by the mechanisms of human memory, we introduce a novel framework to enhancing long-term consistency of video world models through a geometry-grounded long-term spatial memory. Our framework includes mechanisms to store and retrieve information from the long-term spatial memory and we curate custom datasets to train and evaluate world models with explicitly stored 3D memory mechanisms. Our evaluations show improved quality, consistency, and context length compared to relevant baselines, paving the way towards long-term consistent world generation.