ScaLe-INR: Scale and Learn Implicit Neural Representations

A new multi-branch architecture called ScaLe-INR aims to resolve fundamental limitations in Implicit Neural Representations (INRs) when modeling continuous signals, according to a preprint posted to the arXiv repository on June 26, 2026 [1][2]. The paper, authored by Mario De Silva Chethana Lakshan, introduces the Scale and Learn INR framework to address what the researchers describe as persistent spectral bias and information cross-talk in standard INRs [1][2]. When a single network attempts to capture multi-scale phenomena, high-frequency weight updates can destructively interfere with low-frequency structural approximations [2]. ScaLe-INR tackles this by explicitly matching a signal's frequency spectrum with the optimal operating region of the INR through a multi-branch design [2]. The architecture draws on the Fourier inverse scaling theorem, demonstrating that applying directional coordinate scaling expands a network's representational bandwidth along specific spatial axes [2]. To enforce functional disentanglement between branches, the authors propose a Directional Edge Guidance Loss, a spatially-conditioned sparsity prior derived from ground-truth gradients that constrains high-frequency branches to act as localized edge-filters [2]. In evaluations across reconstruction and inverse tasks, ScaLe-INR improved upon the nearest baselines by 5.16 dB in image reconstruction and 0.65 dB in image denoising [1][2]. The preprint reports a figure of 50.02 dB on audio reconstruction and 0.999 Intersection Over Union on 3D reconstruction, results the authors state beat all state-of-the-art models tested [1][2]. The paper appeared on arXiv, an open-access repository that hosts electronic preprints across mathematics, physics, computer science, and related fields [6]. As of November 2024, the repository receives approximately 24,000 submissions per month and has surpassed two million articles [6]. Submissions are moderated but not peer-reviewed before posting [6]. The repository also supports community-developed tools through its arXivLabs framework, which provides features such as bibliographic exploration and code linking on article pages [4][5].