Thera, developed by researchers at ETH Zurich and University of Zurich, is an innovative arbitrary-scale super-resolution method with a unique physical observation model. The method utilizes a hypernetwork to estimate parameters of neural heat fields, ensuring correct anti-aliasing. Results show superior performance compared to a leading competitor. Due to its principled observation model, Thera achieves state-of-the-art performance on super-resolution benchmarks. The details of how Thera works are outlined, emphasizing the advanced techniques used. For more information, the paper “Thera: Aliasing-Free Arbitrary-Scale Super-Resolution with Neural Heat Fields” is available for citation.
https://therasr.github.io/