Superpixel-informed Implicit Neural Representation for Multi-Dimensional Data
This work addresses a limitation in INR methods for multi-dimensional data like images and weather data, offering an incremental improvement by incorporating semantic priors.
The paper tackles the problem of implicit neural representations (INRs) ignoring semantic information in multi-dimensional data recovery by proposing a Superpixel-informed INR (S-INR) that uses generalized superpixels as basic units with attention-based MLPs and a shared dictionary. The result is validated through extensive experiments showing effectiveness compared to state-of-the-art INR methods.
Recently, implicit neural representations (INRs) have attracted increasing attention for multi-dimensional data recovery. However, INRs simply map coordinates via a multi-layer perception (MLP) to corresponding values, ignoring the inherent semantic information of the data. To leverage semantic priors from the data, we propose a novel Superpixel-informed INR (S-INR). Specifically, we suggest utilizing generalized superpixel instead of pixel as an alternative basic unit of INR for multi-dimensional data (e.g., images and weather data). The coordinates of generalized superpixels are first fed into exclusive attention-based MLPs, and then the intermediate results interact with a shared dictionary matrix. The elaborately designed modules in S-INR allow us to ingenuously exploit the semantic information within and across generalized superpixels. Extensive experiments on various applications validate the effectiveness and efficacy of our S-INR compared to state-of-the-art INR methods.