2D Gum-Net: Unsupervised Dense Elastic Image Registration via Symmetric Correlation and Spectral Pooling
arXiv cs.AI · 10 Nov 2026, 04:11 AEST
2D Gum-Net is an end-to-end architecture that predicts dense elastic deformation fields fully unsupervised. It improves structural correlation by up to 15.7% on FVC2004 while maintaining the integrity of imposter scores.
Key facts
- Dense non-rigid registration, the recovery of a pixel-wise deformation field that warps one image into geometric agreement with another, is fundamental across medical imaging, structural biology, and correspondence estimation, yet learning-based approaches face a severe bottleneck: ground-truth deformation fields are notoriously difficult to obtain for real data.
- We present 2D Gum-Net, an end-to-end architecture that predicts dense elastic deformation fields fully unsupervised.
- Trained purely on synthetic impressions and evaluated zero-shot on real benchmarks, our method improves structural correlation by up to 15.7% on FVC2004 and raises genuine match scores under an independent matcher while leaving imposter scores essentially unchanged.
Read aloud
2D Gum-Net is an end-to-end architecture that predicts dense elastic deformation fields fully unsupervised. It improves structural correlation by up to fifteen point seven percent on FVC2004 while maintaining the integrity of imposter scores. Dense non-rigid registration, the recovery of a pixel-wise deformation field that warps one image into geometric agreement with another, is fundamental across medical imaging, structural biology, and correspondence estimation, yet learning-based approaches face a severe bottleneck: ground-truth deformation fields are notoriously difficult to obtain for real data. We present 2D Gum-Net, an end-to-end architecture that predicts dense elastic deformation fields fully unsupervised. Trained purely on synthetic impressions and evaluated zero-shot on real benchmarks, our method improves structural correlation by up to fifteen point seven percent on FVC2004 and raises genuine match scores under an independent matcher while leaving imposter scores essentially unchanged.
Sources in this cluster
- 2D Gum-Net: Unsupervised Dense Elastic Image Registration via Symmetric Correlation and Spectral Pooling
arXiv cs.AI · 10 Nov 2026, 04:11 AEST