DINO-RotateMatch: A Rotation-Aware Deep Framework for Robust Image Matching in Large-Scale 3D Reconstruction

2025-12-03

AI Summary

To address the poor robustness of feature matching under arbitrary rotations in large-scale Internet-image 3D reconstruction, this paper proposes a rotation-aware deep learning matching framework. Methodologically, it integrates self-supervised DINO-based semantic retrieval with rotation-augmented local feature matching: a data-adaptive image-pairing strategy is introduced, coupled with rotation-invariant keypoint detection (ALIKED) and orientation-sensitive feature description, and efficient matching is achieved via LightGlue. Key innovations include rotation-aware keypoint extraction, orientation-enhanced local descriptor modeling, and synergistic optimization combining semantic guidance with geometric constraints. Evaluated on the Kaggle Image Matching Challenge 2025, the method achieves second place (47th out of 943 teams), with significant improvement in mean Average Accuracy (mAA), demonstrating high accuracy, strong robustness under complex viewpoint variations, and excellent scalability.

Technology Category

• Computer Vision: Image and Video Retrieval
• Search and Optimization: Learning to Search
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs

Abstract

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.

Problem

• Improves image matching for large-scale 3D reconstruction
• Addresses challenges from unstructured Internet image collections
• Enhances robustness with rotation-aware feature extraction and matching

Innovation

• Dataset-adaptive image pairing strategy for relevant pairs
• Rotation-aware keypoint extraction with augmentation techniques
• Combining self-supervised global descriptors with enhanced local matching