Reproduce “Continuous Space-Time Video Super-Resolution with 3D Fourier Fields” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Alexander Becker, Julius Erbach, Dominik Narnhofer, Konrad Schindler
Researchers at ETH Zurich developed V³, a continuous space-time video super-resolution method that employs 3D Video Fourier Fields (VFF) for a unified video representation. This approach achieves state-of-the-art performance in various SR tasks with enhanced temporal consistency and improved computational efficiency.
Key insights
- Representing video as a unified 3D Fourier Field inherently models spatio-temporal coherence, eliminating the need for explicit warping and its associated errors in motion compensation.
- Hard-wiring an analytical Point Spread Function provides a principled anti-aliasing mechanism, which offers theoretical guarantees and better generalization compared to implicitly learned filters.
- Utilizing a neural encoder with an extensive receptive field allows the model to jointly reason over multiple frames, capturing long-range dependencies and complex motion patterns more robustly than flow-based methods.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.