ShreddingNet: Coarse-to-Fine Restoration for Multi-Source Shredded Manuscripts

ICLR 2026 Conference Submission22398 Authors

Keywords: manuscript restoration; coarse-to-fine pipeline; computer vision

TL;DR: This work designs a neural network model for multi-source manuscript restoration, employing a coarse-to-fine pipeline to reduce the time complexity, measured by the number of model invocations, to linear.

Abstract: As an important research task of human cultural heritage, the restoration of artworks and calligraphy is of great significance. Seldom existing works have taken the multi-source (*i.e.*, fragments are not ensured to be from the same piece of artworks) fragment-oriented restoration task into account.In this paper, we introduce a restoration algorithm for shredded artworks based on a coarse-to-fine two-stage pipeline.This is an algorithm that can handle the multi-source shredded artworks restoration problem without any restrictive conditions or strong assumptions,and it admits a linear time complexity and robustness to stains, mold, and contour defects. In the proposed coarse matching stage, the algorithm compares the features of each fragment, generating candidate matching pairs. Although a significant number of erroneous matching pairs persist in the candidate set, erroneous matches between fragments from different source images are often rare, enabling high-accuracy clustering of fragments belonging to the same image.In the introduced fine-grained matching stage, the algorithm filters out erroneous matching pairs from the candidate set, producing more precise final matching pairs for global assembly.Experiments conducted on more than 4,000 images from two datasets demonstrate the average reconstruction F1-score achieves 98.37%, which is 5.72% higher than the current state-of-the-art method, confirming the method’s effectiveness and robustness.Source code is available in the supplementary material.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 22398