Implicit Neural Representation-Based Continuous Single Image Super Resolution: An Empirical Study

2026-01-25

AI Summary

This work addresses the lack of systematic empirical evaluation in existing implicit neural representation (INR)-based methods for arbitrary-scale image super-resolution (ASSR). We establish a unified framework to comprehensively compare multiple INR-based ASSR approaches under diverse training strategies, loss functions, and optimization settings, revealing their high sensitivity to training configurations. To improve texture fidelity, we propose a novel intensity variation penalty loss that yields significant gains. Our analysis further demonstrates that recent complex methods offer only marginal improvements over simpler baselines. Notably, we provide the first evidence of scaling laws in INR-based super-resolution, driven by both model size and data diversity. To support reproducible research, we publicly release our implementation code.

Technology Category

• Computer Vision: Representation Learning for Vision
• Machine Learning: Representation Learning
• Natural Language Processing: Sentence-level Semantics, Textual Inference, etc.

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Security and Privacy: Large-scale security measurements
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Implicit neural representation (INR) has become the standard approach for arbitrary-scale image super-resolution (ASSR). To date, no empirical study has systematically examined the effectiveness of existing methods, nor investigated the effects of different training recipes, such as scaling laws, objective design, and optimization strategies. A rigorous empirical analysis is essential not only for benchmarking performance and revealing true gains but also for establishing the current state of ASSR, identifying saturation limits, and highlighting promising directions. We fill this gap by comparing existing techniques across diverse settings and presenting aggregated performance results on multiple image quality metrics. We contribute a unified framework and code repository to facilitate reproducible comparisons. Furthermore, we investigate the impact of carefully controlled training configurations on perceptual image quality and examine a new loss function that penalizes intensity variations while preserving edges, textures, and finer details during training. We conclude the following key insights that have been previously overlooked: (1) Recent, more complex INR methods provide only marginal improvements over earlier methods. (2) Model performance is strongly correlated to training configurations, a factor overlooked in prior works. (3) The proposed loss enhances texture fidelity across architectures, emphasizing the role of objective design for targeted perceptual gains. (4) Scaling laws apply to INR-based ASSR, confirming predictable gains with increased model complexity and data diversity.

Problem

• Implicit Neural Representation
• Arbitrary-Scale Super-Resolution
• Empirical Study
• Training Recipes
• Image Quality Metrics

Innovation

• Implicit Neural Representation
• Arbitrary-Scale Super-Resolution
• Empirical Study
• Perceptual Loss Function
• Scaling Laws