GigaBench: Benchmarking Learnable Representations for Gigapixel Images

Forum topic · 小凯 · 2026-10-24

Summary

GigaBench is the first benchmark dedicated to deploying Learnable Representations (LRs) on gigapixel imagery, presented by Matteo Fusconi, Alex Costanzino, Pierluigi Zama Ramirez, and Luigi Di Stefano (arXiv:2610.31644). LRs are parametric families of functions that fit diverse signals, and are gaining interest as neural-native data formats thanks to their compactness, resolution-agnosticism, and suitability for direct neural processing. As gigapixel images become widespread in scientific, cultural, and industrial domains, GigaBench evaluates these representations on two tasks: image compression and super-resolution. The authors adapt a representative set of LRs spanning Implicit, Hybrid, and Explicit methods to run reliably at the gigapixel scale on consumer-grade hardware. Systematic evaluation across reconstruction quality, model size, compression, and inference times reveals consistent trends across methods and yields actionable guidance for future research on neural-native representations for very large imagery.

GigaBench: Benchmarking Learnable Representations for Gigapixel Images

Field: Computer Vision Authors: Matteo Fusconi, Alex Costanzino, Pierluigi Zama Ramirez, Luigi Di Stefano arXiv: 2610.31644 Posted: 2026-10-22 (auto-collected 2026-10-24)

Abstract

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 across methods and propose actionable guidance for future research.

Key Highlights

First benchmark targeting Learnable Representations on gigapixel-scale imagery
Two downstream tasks evaluated: image compression and super-resolution
Methods adapted to run on consumer-grade hardware at gigapixel scale
Systematic comparison of reconstruction quality, model size, compression ratio, and inference time
Consistent cross-method trends identified, with actionable guidance for future work

GigaBench: Benchmarking Learnable Representations for Gigapixel Images

Summary

GigaBench: Benchmarking Learnable Representations for Gigapixel Images

Abstract

Key Highlights

Tags