Ranked 5th (Top 0.5% of 943 teams) in Image Matching Challenge | Kaggle Solo Gold Medal 🏅

Ranked 5th out of 943 teams (Top 0.5%) in the Image Matching Challenge 2025 on Kaggle, run in conjunction with the Image Matching Workshop at the IEEE/CVF Conference on Computer Vision and Pattern Recognition. The competition targeted end-to-end, real-world Structure-from-Motion (SfM) from messy photo collections: participants had to cluster images into coherent scene groups (or label outliers) and then build sparse SfM reconstructions, estimating camera poses/geometry for each cluster. It was challenging because performance depended on both correct grouping and geometrically accurate reconstruction (scored via a combined metric coupling a clustering score with reconstruction accuracy, e.g., mAA), under substantial viewpoint/appearance variation and many distractors.