 EMREDI is a novel deep learning framework developed to improve the quality and interpretability of Creo-Electron Microscopy, Creo-EM, Maps. The framework consists of two main components, a swing-convunet-based local modeling module and a non-local modeling module. Both modules are integrated into a multi-scale unit architecture, which enables simultaneous optimization of local and non-local image processing. This allows the framework to effectively reduce noise and improve the contrast of Creo-EM maps while preserving their structural information. Additionally, the framework incorporates a loss function that minimizes the local smooth L1 distance and maximizes the non-local structural similarity between processed experimental and simulated target maps. This ensures that the resulting maps are more interpretable and suitable for use in de novo model building. This article was authored by Jawahi, Tao Li, and Cheng Yu-Huang.