Solv3R looped transformer solves multi-view geometry
Original title: Solv3R: 3D Looped Transformer as a Multi-view Solver
Solv3R uses a looped transformer for iterative refinement in multi-view reconstruction without task-specific modules.
Solv3R learns to continue solving multi-view geometry from its current state with an update operator independent of iteration count. It supports offline, incremental, relocalization and streaming reconstruction. Representations of solved views can be cached for efficiency.
- Update operator independent of iteration count
- Preserves accurate reconstructions under continued iteration
- Supports offline incremental relocalization and streaming
- Code and models released at ubc-vision.github.io/solv3r
- Single learned solver for all tasks
- AI models
- Solv3R
From the original article
1 min read at huggingface.co
Classical structure-from-motion reconstruction proceeds iteratively, refining camera poses and scene geometry from a current estimate. This allows maps to easily expand without recomputing already computed views, or geometry estimates to improve as more observations are provided.
By Jeong-gi Kwak, Andrea Tagliasacchi, Sho Kagami · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.