DCARL: A Divide-and-Conquer Framework for Autoregressive Long-Trajectory Video Generation
This addresses the challenge of scalable and stable long-video generation for world modeling applications, representing a strong incremental improvement over existing methods.
The paper tackled the problem of long-trajectory video generation, which suffers from visual drift and poor controllability in autoregressive models, by proposing DCARL, a divide-and-conquer framework that achieved superior performance in visual quality and camera adherence for videos up to 32 seconds.
Long-trajectory video generation is a crucial yet challenging task for world modeling primarily due to the limited scalability of existing video diffusion models (VDMs). Autoregressive models, while offering infinite rollout, suffer from visual drift and poor controllability. To address these issues, we propose DCARL, a novel divide-and-conquer, autoregressive framework that effectively combines the structural stability of the divide-and-conquer scheme with the high-fidelity generation of VDMs. Our approach first employs a dedicated Keyframe Generator trained without temporal compression to establish long-range, globally consistent structural anchors. Subsequently, an Interpolation Generator synthesizes the dense frames in an autoregressive manner with overlapping segments, utilizing the keyframes for global context and a single clean preceding frame for local coherence. Trained on a large-scale internet long trajectory video dataset, our method achieves superior performance in both visual quality (lower FID and FVD) and camera adherence (lower ATE and ARE) compared to state-of-the-art autoregressive and divide-and-conquer baselines, demonstrating stable and high-fidelity generation for long trajectory videos up to 32 seconds in length.