Source code for MA-LaMa: Restoration of ancient Chinese paintings using GANs with multi-scale defect fusion architecture and multi-pooling enhanced spatial–channel attention

作者：Chaoyang Zhang, Xiang Li, Mingder Jean · 发表于：Zenodo (CERN European Organization for Nuclear Research) · 年份：2026 · DOI：10.5281/zenodo.21456876 · 研究领域：Computer science、Artificial intelligence、Programming language、Computer graphics (images)、Computer vision、Natural language processing、Engineering drawing

This repository contains the source code supporting the MA-LaMa framework proposed in the manuscript “Restoration of ancient Chinese paintings using GANs with multi-scale defect fusion architecture and multi-pooling enhanced spatial–channel attention.” The repository includes the implementations of the multi-scale defect fusion architecture (MDFA), the multi-pooling enhanced channel–spatial module (MECS), the MA-LaMa model, training and inference scripts, evaluation scripts, configuration files, dependency information, and trained model weights.