ainewsblitz.com

Breaking

RoMa v2 pairs frozen DINOv3 features with a two-stage matcher for denser image correspondences

RoMa v2 pairs frozen DINOv3 features with a two-stage matcher for denser image correspondences
  • Research & Papers
  • Foundation Models
  • Open Source

Researchers from Linköping University, Chalmers University of Technology, Lund University and collaborators have released RoMa v2, a dense feature matcher that estimates pixel-level correspondences between image pairs more accurately and more quickly than the original RoMa, with public code and an ECCV 2026 camera-ready paper. The model is described in RoMa v2: Harder Better Faster Denser Feature Matching and implemented in the RoMaV2 repository.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.

$20
Read this article
$29/month
Unlimited — all 27,649 articles, the full archive, and comprehension quizzes
Save 72%
$98/year
≈ $8.17/month
Unlimited, billed once a year