🤖 AI Summary
Existing arbitrary-scale single-image super-resolution (ASR) methods employ neural field modeling, but their point-wise sampling fails to align with the pixel’s point spread function (PSF), causing aliasing artifacts; approximate integral rendering mitigates this at the cost of reduced fidelity and generalization. This paper introduces Neural Thermal Field (NTF), the first approach to model the PSF as a physically consistent heat diffusion process, enabling analytical, lossless anti-aliasing for arbitrary scaling factors. NTF yields a differentiable, zero-overhead representation that supports end-to-end optimization and inference while ensuring theoretical rigor and parameter efficiency. Experiments demonstrate that NTF significantly outperforms state-of-the-art methods across multi-scale reconstruction tasks, achieving consistent improvements in PSNR and SSIM without additional computational cost.
📝 Abstract
Recent approaches to arbitrary-scale single image super-resolution (ASR) use neural fields to represent continuous signals that can be sampled at arbitrary resolutions. However, point-wise queries of neural fields do not naturally match the point spread function (PSF) of pixels, which may cause aliasing in the super-resolved image. Existing methods attempt to mitigate this by approximating an integral version of the field at each scaling factor, compromising both fidelity and generalization. In this work, we introduce neural heat fields, a novel neural field formulation that inherently models a physically exact PSF. Our formulation enables analytically correct anti-aliasing at any desired output resolution, and -- unlike supersampling -- at no additional cost. Building on this foundation, we propose Thera, an end-to-end ASR method that substantially outperforms existing approaches, while being more parameter-efficient and offering strong theoretical guarantees. The project page is at https://therasr.github.io.
