DINO-RotateMatch: A Rotation-Aware Deep Framework for Robust Image Matching in Large-Scale 3D Reconstruction
This work addresses robust and scalable image matching for large-scale 3D reconstruction, but it is incremental as it combines existing methods like DINO, ALIKED, and Light Glue with rotation enhancements.
The paper tackled the problem of image matching in large-scale 3D reconstruction from unstructured Internet images by integrating dataset-adaptive pairing with rotation-aware keypoint extraction and matching, resulting in a Silver Award (47th of 943 teams) on the Kaggle Image Matching Challenge 2025 with improvements in mean Average Accuracy (mAA).
This paper presents DINO-RotateMatch, a deep-learning framework designed to address the chal lenges of image matching in large-scale 3D reconstruction from unstructured Internet images. The method integrates a dataset-adaptive image pairing strategy with rotation-aware keypoint extraction and matching. DINO is employed to retrieve semantically relevant image pairs in large collections, while rotation-based augmentation captures orientation-dependent local features using ALIKED and Light Glue. Experiments on the Kaggle Image Matching Challenge 2025 demonstrate consistent improve ments in mean Average Accuracy (mAA), achieving a Silver Award (47th of 943 teams). The results confirm that combining self-supervised global descriptors with rotation-enhanced local matching offers a robust and scalable solution for large-scale 3D reconstruction.