VMem: Consistent Interactive Video Scene Generation with Surfel-Indexed View Memory
This addresses the challenge of long-term scene coherence in video generation for applications like virtual exploration, though it is an incremental improvement over existing methods.
The paper tackles the problem of generating consistent interactive video scenes by introducing a memory module that indexes past views geometrically, enabling efficient retrieval and reducing computational cost while maintaining scene coherence.
We propose a novel memory module for building video generators capable of interactively exploring environments. Previous approaches have achieved similar results either by out-painting 2D views of a scene while incrementally reconstructing its 3D geometry-which quickly accumulates errors-or by using video generators with a short context window, which struggle to maintain scene coherence over the long term. To address these limitations, we introduce Surfel-Indexed View Memory (VMem), a memory module that remembers past views by indexing them geometrically based on the 3D surface elements (surfels) they have observed. VMem enables efficient retrieval of the most relevant past views when generating new ones. By focusing only on these relevant views, our method produces consistent explorations of imagined environments at a fraction of the computational cost required to use all past views as context. We evaluate our approach on challenging long-term scene synthesis benchmarks and demonstrate superior performance compared to existing methods in maintaining scene coherence and camera control.