IMC 2024 Methods & Solutions Review

TL;DR

This paper surveys six SfM-focused categories from IMC 2024 and reviews leading methods for image matching and 3D reconstruction, with emphasis on efficiency, generalization, and handling challenging scenes such as transparent objects. It presents an ensemble-based solution that leverages ALIKED, LightGlue, OmniGlue, and COLMAP, combined with task-specific adaptations, achieving a private-leaderboard score of 0.153449 and placing 160th among over 1,000 participants. The work contrasts dense and sparse matching paradigms (LoFTR, SuperGlue, LightGlue, OmniGlue) and provides practical insights into edge-case handling, rotation/orientation issues, and scene-specific preprocessing. Collectively, the paper offers actionable guidance for future participants and researchers aiming to improve robust 3D reconstruction from varied real-world imagery. The combination of a thorough method review with a concrete ensemble solution advances understanding of image matching pipelines and their applicability to challenging SfM scenarios.

Abstract

For the past three years, Kaggle has been hosting the Image Matching Challenge, which focuses on solving a 3D image reconstruction problem using a collection of 2D images. Each year, this competition fosters the development of innovative and effective methodologies by its participants. In this paper, we introduce an advanced ensemble technique that we developed, achieving a score of 0.153449 on the private leaderboard and securing the 160th position out of over 1,000 participants. Additionally, we conduct a comprehensive review of existing methods and techniques employed by top-performing teams in the competition. Our solution, alongside the insights gathered from other leading approaches, contributes to the ongoing advancement in the field of 3D image reconstruction. This research provides valuable knowledge for future participants and researchers aiming to excel in similar image matching and reconstruction challenges.

Figures (9)

• Figure 1: General Flow of how data was processed by most of the competitors
• Figure 2: using segmentation models significantly improved keypoint detection on transparent
• Figure 3: as clearly visible, background keypoints detected in transparent scene class. To solve this DINOv2 provides seperation between foreground & background which helps detect foreground keypoints for transparent scene
• Figure 4: detection and segmentation of foreground can be seen.
• Figure 5: LigthGlue is faster at matching easy image pairs (top) than difficult ones (bottom) because it can stop at earlier layers when its predictions are confident.