Search papers, labs, and topics across Lattice.
This paper investigates the use of sinusoidal recurrence in implicit neural representations (INRs) to enhance harmonic spectral enrichment through a shared sinusoidal block that iteratively refines latent representations. The authors demonstrate that sinusoidal activations create a harmonic line spectrum, which enriches the spectral support more effectively than traditional feed-forward INRs and non-sinusoidal recurrent models. Empirical results show that their approach achieves superior fidelity on RGB image benchmarks while requiring fewer parameters and optimization steps, and it also performs well on super-resolution, NeRF, and SDF tasks.
Sinusoidal recurrence can achieve higher fidelity in image representation with fewer resources, outperforming traditional feed-forward models.
We study sinusoidal recurrence as an iterative mechanism for harmonic spectral enrichment in implicit neural representations (INRs). Our analysis reveals that sinusoidal activations induce a harmonic line spectrum, providing a spectral account of how recurrent unrolling enriches the effective spectral support. We realize this principle with a shared sinusoidal block that iteratively refines the latent representation. We empirically validate the resulting spectral behavior against feed-forward INRs, non-sinusoidal recurrent variants, and equilibrium-style sinusoidal models. Complementing this analysis, we evaluate the proposed architecture across image and 3D representation tasks. On RGB image benchmarks, our method achieves higher fidelity than feed-forward baselines with fewer parameters and fewer optimization steps, and it further transfers favorably to super-resolution, NeRF, and SDF tasks.
