Grids Often Outperform Implicit Neural Representations at Compressing Dense Signals
A new research paper published on arXiv suggests that traditional grid-based representations often outperform Implicit Neural Representations (INRs) when compressing dense signals. The study found that regularized grids train faster and achieve comparable or better quality than INRs with the same parameter count for many tasks. INRs showed an advantage primarily in fitting binary signals like shape contours, indicating specific applications where they are most beneficial. AI
IMPACT Suggests specific use cases for INRs, potentially guiding future development towards more efficient signal compression techniques.