超越像素：从视频先验到4D世界

摘要

4D生成能够从文本或图像等条件合成动态3D场景。现有方法要么使用独立的4D模型重建生成的RGB视频，要么调整特定的视频生成器直接预测几何信息。前者存在分布不匹配和误差传播的问题，而后者则将4D预测与特定生成器绑定，在生成器或条件机制发生变化时可能需要重新训练。我们探究共享变分自编码器（VAE）的视频模型的最终去噪潜变量能否作为可复用的接口，用于显式4D预测。基于这一洞见，我们提出了直接从潜变量到4D的生成方法，并将其实现为Latent-to-4D，该方法绕过RGB，将视频潜变量与预训练4D解码器的词元网格对齐，并通过逐帧和全局时空注意力进行精化。仅使用约1K个现有重建片段进行训练，单个检查点即可在同一VAE家族内的多个视频扩散Transformer之间无需修改地迁移。在Text4D-200和I4D-200数据集上，Latent-to-4D在基于投影的DINO-F1指标上分别超过匹配同潜变量Wan+4RC级联方法2.88–3.45分和5.81分，同时人工评估者在几何质量、时间稳定性和整体质量方面也更为偏好该方法。

English

4D generation synthesizes dynamic 3D scenes from conditions such as text or images. Existing methods either reconstruct generated RGB videos with a separate 4D model or adapt a particular video generator to predict geometry directly. The former suffers from distribution mismatch and error propagation, whereas the latter ties 4D prediction to a specific generator and may require retraining when the generator or conditioning regime changes. We ask whether the final denoised latents of video models that share a variational autoencoder (VAE) can instead provide a reusable interface to explicit 4D prediction. Building on this insight, we introduce direct latent-to-4D generation and instantiate it as Latent-to-4D, which bypasses RGB by aligning a video latent with the token grid of a pretrained 4D decoder and refining it through frame-wise and global spatiotemporal attention. Trained on roughly 1K existing reconstruction clips, a single checkpoint transfers unchanged across multiple video diffusion transformers within the same VAE family. On Text4D-200 and I4D-200, Latent-to-4D surpasses matched same-latent Wan+4RC cascades in projection-based DINO-F1 by 2.88--3.45 and 5.81 points, respectively, while also being preferred by human raters for geometry, temporal stability, and overall quality.