Retrieve What's Missing: Coverage-Maximizing Retrieval for Consistent Long Video Generation

2026-06-01

AI Summary

Long-term autoregressive video generation often suffers from structural distortions due to insufficient geometric consistency over extended sequences. To address this, this work proposes COVRAG, a novel framework that introduces, for the first time, a depth-based coverage maximization strategy. Leveraging a pretrained 3D prior, COVRAG constructs a target-view coverage map as a lightweight memory and iteratively retrieves historical frames that maximally complement currently missing regions through residual coverage gain. The framework further incorporates a sliding-window depth caching mechanism, which significantly enhances geometric consistency and scalability in long-form video generation while maintaining low latency. Experimental results demonstrate that COVRAG substantially outperforms existing baseline methods on the RealEstate10K and DL3DV10K datasets.

Abstract

Maintaining long-term geometric consistency remains challenging for long-horizon autoregressive video generation. Memory-augmented generative models address this by retrieving historical frames, but their effectiveness depends on two key design choices: what 3D-geometric evidence should represent past observations, and how memory frames should be selected from this evidence. Existing methods often rely on camera poses or field-of-view overlap, which are lightweight but too coarse to reason about pixel-wise visibility, or use explicit 3D reconstruction, which provides fine-grained evidence but is costly to maintain over long rollouts. We propose Coverage-Maximizing Retrieval-Augmented Generation (COVRAG), a depth-based memory retrieval framework that uses pretrained 3D priors to construct a target-view coverage map as lightweight 3D memory evidence. For frame selection, COVRAG maximizes residual coverage gain, iteratively retrieving frames that explain target-view regions not covered by the current context or previously selected memories. To improve scalability in long-video generation, we introduce sliding-window depth caching for efficient geometry estimation. Experiments on RealEstate10K and DL3DV10K show that COVRAG improves long-horizon geometric consistency while maintaining low latency compared to baselines.

Problem

• long video generation
• geometric consistency
• memory retrieval
• 3D reconstruction
• autoregressive generation

Innovation

• coverage-maximizing retrieval
• 3D priors
• memory-augmented generation
• geometric consistency
• depth caching