Description
Reconstructing an object from a few photos leaves unseen regions missing or distorted. ReconViaGen, an ICLR 2026 work, brings generation into multi-view reconstruction to fill unseen parts for more accurate 3D objects.
It also estimates camera poses and has a demo on Hugging Face.
Multi-view:Few photos.
Generative fill:Missing regions repaired.
Pose estimation:Cameras inferred.
Demo:Hugging Face.
It also estimates camera poses and has a demo on Hugging Face.
Features
Multi-view:Few photos.
Generative fill:Missing regions repaired.
Pose estimation:Cameras inferred.
Demo:Hugging Face.
