[论文] GigaBench: Benchmarking Learnable Representations for Gigapixel Image ...

论文概要

研究领域: CV
作者: Matteo Fusconi, Alex Costanzino, Pierluigi Zama Ramirez, Luigi Di Stefano
发布时间: 2026-10-22
arXiv: 2610.31644

中文摘要

可学习表示（LR）可以将参数化函数族拟合到多种信号上。这些新兴的神经原生数据格式因其紧凑性、分辨率无关性和可直接进行神经处理的特点正受到科学界关注。与此同时，十亿像素（gigapixel）图像在科学、文化和工业领域越来越普及。本文提出 GigaBench——首个专注于在十亿像素图像上部署可学习表示的基准和实验评估，涵盖图像压缩和超分辨率两个任务。我们改编了一组涵盖隐式、混合式和显式方法的代表性 LR，使其能在消费级硬件上可靠地在十亿像素规模运行。通过跨重建质量、模型大小、压缩和推理时间的系统评估，我们识别出跨方法的一致趋势，并为未来研究提出可操作的指导建议。

原文摘要

Learnable Representations (LRs) can fit parametric families of functions to diverse kinds of signals. These emergent neural-native data formats are gathering interest due to their compactness, resolution-agnosticism and amenability to direct neural processing. Gigapixel images are becoming increasingly widespread across scientific, cultural, and industrial domains. We present GigaBench, the first benchmark focused on deploying LRs with gigapixel imagery, addressing image compression and super-resolution. We adapt a representative set of LRs spanning Implicit, Hybrid, and Explicit methods to operate reliably at the gigapixel scale on consumer hardware. Through systematic evaluation across reconstruction quality, model size, compression and inference times, we identify consistent trends and ...

自动采集于 2026-10-24

#论文 #arXiv #CV #小凯