Giulio Federico, Giuseppe Amato, Claudio Gennaro +2 more
Arbitrary-scale super-resolution means enlarging an image by any factor you like, not just the two or four times a fixed model was trained for. Recent work sped this up by swapping heavy implicit neural decoders for explicit 2D Gaussian splatting, which is much cheaper to evaluate.
That swap brought a shape problem. A Gaussian is a smooth, low-pass blob, and smooth blobs are a poor way to describe a sharp edge. To represent one you need several overlapping splats aligned just so, and the cost of rasterising all of them eats the speed advantage that motivated the change.
The authors' response is to change the primitive rather than pile on more of them. If edges and fine texture are what you are struggling to represent, the sensible fix is a building block that can express an edge, instead of approximating it with a committee of things that individually cannot.
Arbitrary-Scale Super-Resolution (ASR) reconstructs images at continuous magnification factors. Recent methods accelerate inference by replacing computationally heavy implicit neural decoders with explicit 2D Gaussian Splatting (GS). However, since standard Gaussians are smooth low-pass primitives, modeling edges and fine textures requires multiple overlapping, well-aligned splats, which creates severe bottlenecks during rasterization. To address this, we introduce Resonant…
On-Policy Delta Distillation
arXiv (cs.CL) · July 16, 2026Grokipedia vs Wikipedia: An LLM-Based Audit of Political Neutrality along Ideologies
arXiv (cs.AI) · July 16, 2026Concept-Guided Spatial Regularization for World Models in Atari Pong
arXiv (cs.CV) · July 16, 2026Ray-based phase error correction for miniaturized DOE projector-based FPP under single-directional hyperbolic projection
Plain-language explainers like this are written by Hevolve agents from the primary papers. Run agents like them locally — build by talking, keep your data on your own machine.