FlexGlue: open-source pixel-perfect image matching that rejects doppelganger pairs
ducha_aiki · x · 2026-11-04
Philipp Lindenberger, Paul-Edouard Sarlin, Ondrej Miksik and Marc Pollefeys (ETH Zurich / CVG) released FlexGlue, an open-source feature matching model (MIT-licensed code plus paper).
Key points:
- One model, four matching modes: sparse-to-dense matches tailored to each use case
- Robust under extreme viewpoint changes, low-texture scenes, rotation and lighting shifts
- Covisibility classifier detects and rejects unmatchable doppelganger pairs — a common SfM failure mode
- State-of-the-art SfM and matching with fast inference
Repo cvg/FlexGlue ships inference code, ONNX weights (built on DINOv3), Colab demo, and works with any keypoint detector.
More from Research
- ParallACT Benchmark: Frontier Models Hit Just 16% on Multi-Principal Agent Tasks — EliasEskin · 2026-11-04
- TAMU professor releases 400+ page AI-assisted tutorial deck on generative models — ShuiwangJi · 2026-11-04
- DeepMind mafia seeding a new wave of AI for Science startups in the UK — ersatzben · 2026-11-04
- MIT CSAIL fixes small models by hand-editing weights — hype or old idea rebranded? — ziv_ravid · 2026-11-04
- Robots need failure data: learned sims handle rendering, MuJoCo keeps contact honest — chris_j_paxton · 2026-11-04
- Robotics researcher: training neural sims needs ~10x more failure than success data — chris_j_paxton · 2026-11-04