InfiniteNature-Zero: Learning Perpetual View Generation of Natural Scenes from Single Images
This enables perpetual view generation for applications like virtual reality and content creation, representing a novel paradigm rather than an incremental improvement.
The paper tackles the problem of generating unbounded flythrough videos from a single image of a natural scene, achieving this by learning from a collection of single photographs without requiring camera poses or multiple views, and demonstrates superior performance compared to supervised methods.
We present a method for learning to generate unbounded flythrough videos of natural scenes starting from a single view, where this capability is learned from a collection of single photographs, without requiring camera poses or even multiple views of each scene. To achieve this, we propose a novel self-supervised view generation training paradigm, where we sample and rendering virtual camera trajectories, including cyclic ones, allowing our model to learn stable view generation from a collection of single views. At test time, despite never seeing a video during training, our approach can take a single image and generate long camera trajectories comprised of hundreds of new views with realistic and diverse content. We compare our approach with recent state-of-the-art supervised view generation methods that require posed multi-view videos and demonstrate superior performance and synthesis quality.